<%BANNER%>

Analysis of neonatal heart rate variability and cardiac orienting responses

University of Florida Institutional Repository
xml version 1.0 encoding UTF-8
REPORT xmlns http:www.fcla.edudlsmddaitss xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.fcla.edudlsmddaitssdaitssReport.xsd
INGEST IEID E20110109_AAAAZI INGEST_TIME 2011-01-10T01:49:43Z PACKAGE UFE0001215_00001
AGREEMENT_INFO ACCOUNT UF PROJECT UFDC
FILES
FILE SIZE 41600 DFID F20110109_AADDTO ORIGIN DEPOSITOR PATH lebrun_d_Page_046thm.jpg GLOBAL false PRESERVATION BIT MESSAGE_DIGEST ALGORITHM MD5
e6c9f866b2664c331eea328279d65d87
SHA-1
97be52d3ef8f5549b8a4b97d8f9bb9114e605ca2
75096 F20110109_AADDTP lebrun_d_Page_047.QC.jpg
37b639226c0f0dcd4c28964d4bc7ef71
0218238031919d6c40f2e97223d3e992d66225ec
76189 F20110109_AADDUD lebrun_d_Page_069.QC.jpg
8c2e6a0dabdbb762a47a7ee21f4606fe
e1f8dbef0d0eb52105c2c8f763d2c10b9628c576
40114 F20110109_AADDTQ lebrun_d_Page_049thm.jpg
39c26904ca3a5fef284917acd3c94a3b
3ef103c19d362f24ee25a3c3499a9afe3dfb8e71
54795 F20110109_AADDUE lebrun_d_Page_071.QC.jpg
650037ea7fffd9fb5ba698e2a2eb5bf0
87eb8f9de5d953eddb3af40eb868b672f699636b
73911 F20110109_AADDTR lebrun_d_Page_052.QC.jpg
4a1a592500e6c97adea13b63cae0462e
448dea92fe9683476aaac502f69e60a3089624a7
72922 F20110109_AADDUF lebrun_d_Page_072.QC.jpg
dd95e9a3ce67cd27a915a77799b2d5de
021948711b947ea54eadbd2b637831a31889f472
35092 F20110109_AADDTS lebrun_d_Page_053thm.jpg
f75b4195125938def606e9db2babaa01
ebf3777e3d7dac9cb48fec512a8030a549f1d4f7
40478 F20110109_AADDUG lebrun_d_Page_072thm.jpg
e4a0e7bf09fbe60c27df6fb5d5015736
82f0dae8f839c9cbaaf5d7b7d2f301010a769fe0
42568 F20110109_AADDTT lebrun_d_Page_054thm.jpg
6dda4b7663f6e0ddebb1fa802b2f7989
445e832147c2faf6b8a1bd7156a1765e9260b6e7
42154 F20110109_AADDUH lebrun_d_Page_075thm.jpg
5fbca597c3882852d25b90b8293aa4c1
bdab127727aca976e828413954ae23e649907aff
11497 F20110109_AADDTU lebrun_d_Page_055.QC.jpg
1834e2490bf2f514ea937a10d81baa68
931ea0677722f2cb4f552d6076ab7705dd64938a
84608 F20110109_AADDUI lebrun_d_Page_076.QC.jpg
eac4953dfbb67d878cdba8d79affd302
68396d85c5fcf0a698e5b5ee35183c063554fae6
4708 F20110109_AADDTV lebrun_d_Page_055thm.jpg
004039e30fc8580e7c5556278fd1c5ac
418eacb47d5bfc9edb30b6b10532dba9196b814d
46484 F20110109_AADDUJ lebrun_d_Page_076thm.jpg
6da8f1019727eac2af256b49d37b2486
6e90ab266b31c6a77cb9cca0b07ced6dff430946
80686 F20110109_AADDTW lebrun_d_Page_058.QC.jpg
5bf4f99157b0b564a11996545eb097a9
9ed5e479c8b81d49298ac5b42276dc613623a2a2
42634 F20110109_AADDUK lebrun_d_Page_077thm.jpg
f83c95b68ea02df660b59d58fbb212ba
0638672b9aae70da416c5515fdd689567a554cbf
83461 F20110109_AADDTX lebrun_d_Page_059.QC.jpg
0d93264630bff1baae70cc6d5f2c0fa0
abd3a788f95fb0e20f9f5732b938d7ab1661f585
22637 F20110109_AADDVA lebrun_d_Page_087thm.jpg
622da6ddfe137ef776bd02d5373b901a
a268bdcbedf18fdcaad3dd302611a733542d2f21
86484 F20110109_AADDUL lebrun_d_Page_078.QC.jpg
253ce4de1ca72108919c8258fbff5c3f
e1940af99321942a70a0081df91c451f63a27682
38237 F20110109_AADDTY lebrun_d_Page_060thm.jpg
97228e7e8947ab192eb4071102633691
a188b8b606793b9ab0acbbc5f877a3289a5c71a6
63322 F20110109_AADDVB lebrun_d_Page_088.QC.jpg
dba9029ad09355328ae1a4934a86dd72
55675f62791ff311c2704d11ad261f0820938b76
46013 F20110109_AADDUM lebrun_d_Page_078thm.jpg
87a6fac88e94f3077ed6092a4c39b4b2
fd92621163aceda4021e9ac315669ac843b40157
68443 F20110109_AADDTZ lebrun_d_Page_061.QC.jpg
4bab3e158abd8bb47e39791bfe702f2a
7ed08d44c796ebe2de7a576946b64269b3b1e795
20820 F20110109_AADDVC lebrun_d_Page_088thm.jpg
d01175d66d7139fda2b98bd999c940d4
4f621bdedb35e08a0aa06c7a459fbe0303128a8a
81732 F20110109_AADDUN lebrun_d_Page_079.QC.jpg
3d77b8d579879e4209c108334a6cc733
8b25170ac9ce65df5ba5e49d1cef27524ba75ab1
38881 F20110109_AADDVD lebrun_d_Page_089thm.jpg
0165d4ff688c13be9921d21be0c24972
c97428d0e989c8febf4a41cc544211e5a70d76e9
24521 F20110109_AADDUO lebrun_d_Page_079thm.jpg
4bc993370593c6cd2d6546df529b4296
3cc5451e872f7a17cb53a445d0cf4c1045dbe540
91340 F20110109_AADDUP lebrun_d_Page_080.QC.jpg
0e819a1ec950147f1e43775e84110b6c
4edf2f7ea4353f4ac7db4e1be5ed87bf366e90be
77494 F20110109_AADDVE lebrun_d_Page_090.QC.jpg
9bf620a1fa54fbd656189fd13f748a66
7b39de0db5da71842da2f5dfbed86766762cd221
46246 F20110109_AADDUQ lebrun_d_Page_080thm.jpg
394ac28d7262e6477ee402373b305c15
9fbfb706b0077504d73618101af969b4fc0afa46
73219 F20110109_AADDVF lebrun_d_Page_091.QC.jpg
6645f29fc26e77050728e1ff7160e369
24aba8eeff0717ca83df0c808c4999150b5a52e7
55496 F20110109_AADDUR lebrun_d_Page_081.QC.jpg
fade217e7776c0a3b77952ae4ef19001
aa9f19b8fb75658ed9b5134b05ef8ed7f27ba9c1
71143 F20110109_AADDVG lebrun_d_Page_092.QC.jpg
c00b2327ed60524d0f561e0df3b7fd9f
3ba17d5b4878f3cfffad356029d57b333539b2f9
48765 F20110109_AADDUS lebrun_d_Page_082.QC.jpg
31b0611145ed2b5317105e8f620f08d1
df388e68280416312494c8fe32eb1b7c934069fc
42867 F20110109_AADDVH lebrun_d_Page_092thm.jpg
546492b36308429323b4cf621a81806b
311f2abd4dca179b98e0847257f283bd679b951a
21801 F20110109_AADDUT lebrun_d_Page_082thm.jpg
4a63526b2cddb31dabe638b87384305a
efc582c63077bcd636b82be6f7f2e089df7bd062
42955 F20110109_AADDVI lebrun_d_Page_093thm.jpg
37ef979fa8ef156cacadbfd443898215
ad8042aef369dd1d2d965823904b9b41c239d301
67039 F20110109_AADDUU lebrun_d_Page_083.QC.jpg
7a95a6ea17cbe278fae383c0e43dd0cf
65533511fef3b6c2433afc12c4b486074d5aa4c8
74587 F20110109_AADDVJ lebrun_d_Page_095.QC.jpg
ef14b16d588fcdfc37a76a6d97c42b47
ab1af10660ec1d786a05d14029c9a45760faf7a4
21134 F20110109_AADDUV lebrun_d_Page_083thm.jpg
ef2a134b472e5492ba2c1e863dea1054
c02ffbcfac79138d1e889dc9c139bd77502f0652
43424 F20110109_AADDVK lebrun_d_Page_095thm.jpg
650f13ce1496f02ab4e799431664ea53
80b9b052161a62e99d9c3ac0d07d3c8d1798b41d
70196 F20110109_AADDUW lebrun_d_Page_084.QC.jpg
1bf69f62de6e155f957e4e5d8c41647a
15887cde37fbe8260a6c6eef0a02fcb750699738
39260 F20110109_AADDVL lebrun_d_Page_097thm.jpg
d6dc8a155fdadcc71b66c483a52b45b3
9c9d1a934c1b5a0cd3e11f401505c480ba9e64b7
22555 F20110109_AADDUX lebrun_d_Page_084thm.jpg
a44543fd8cf6ea7aa48123d0c8a8c156
653ab8e8eecbb431578370495d93ff6099092fd4
F20110109_AADDVM lebrun_d_Page_098thm.jpg
bd87ed0e004f43587201293c8df5cfbc
53cc249a9fbeb8682d26355f20a4f4948055a189
27821 F20110109_AADDUY lebrun_d_Page_085.QC.jpg
72b1e2bafe9b16fd9986950f122609c3
0d27e527e494f0419c859f86d6afeaa1ae5c3d11
67152 F20110109_AADDVN lebrun_d_Page_099.QC.jpg
423ec229390c449299913617d6a0f3d4
e90fd72b5c61ec1aa4987685197834d8dfcd0d19
19986 F20110109_AADDUZ lebrun_d_Page_086thm.jpg
a28f2bd575fceb8959a21c0696d7d6b5
79a315a06c5c2becb7db1bfb45d12a535d0d695b
83470 F20110109_AADDVO lebrun_d_Page_102.QC.jpg
85a963cf07f159f99060f6dbd85015fd
d3d6e5b1b82a8bc76be0d6560bd68c0001dc3247
10664 F20110109_AADDVP lebrun_d_Page_103.QC.jpg
9abf7e734369b53cb928a33156f4c4ed
c616c8fdf10beba2c2d662c534df3000dfe4bed2
4144 F20110109_AADDVQ lebrun_d_Page_103thm.jpg
7787c848e67c0490448e377b6569ab87
a2b4f7050e2f116b4f56dc662b6720df3dc06126
7313 F20110109_AADDVR lebrun_d_Page_104thm.jpg
b77f90ed9643e5ddc99c3c93453773d1
517e9fc5b35cc2e52a75530c9eb8623807a172ef
120892 F20110109_AADDVS UFE0001215_00001.mets FULL
c1a54ae285d0572015b1283689fe56d1
15a8bb90ade318200acfff66fa6cbff7f1caf1cc
39021 F20110109_AADCUA lebrun_d_Page_086.pro
a0edd1eb2a8bdb2813469449db23e700
f70f88818a694e3b26039b3413b5d6aa89af3336
147138 F20110109_AADCUB lebrun_d_Page_041.jpg
bedd3fcac03b5e7fbb6898bf032cc8f4
e1a92bb4440b2eea3fe9ab5831caf65c963eb537
162114 F20110109_AADCUC lebrun_d_Page_043.jpg
785f22f0a3b9bb5cf9a57ca90fba2a99
f8eb0a45f06d534dc070d82f394e59617c1574d9
469935 F20110109_AADCUD lebrun_d_Page_008.jp2
eaefdad43af03b3554ea918f76b29286
7ca838f30fdc0f29c59ab0b895561358718ba356
47813 F20110109_AADCTO lebrun_d_Page_028thm.jpg
75b8b70a1fdb3aed5f0d0afa1aa0516c
cd5d09cbf97deebe2f554dd64151fda044e44fff
170826 F20110109_AADCTP lebrun_d_Page_048.jpg
e44778138a4a15aac93d6b567ba889aa
bfb507914e2ac5382902a3dfb1da3e9bdc60a9ce
85295 F20110109_AADCUE lebrun_d_Page_015.jp2
68cae164d2dc148d8244e49bcfc9a8bb
5b7e307582080c9927c727e858d8e6f737b5984f
41792 F20110109_AADCTQ lebrun_d_Page_047thm.jpg
ec485f387abf1fea6e31850d17f5932c
14051b905a8e7fd14a73aa895a77c655dafb76b4
973 F20110109_AADCUF lebrun_d_Page_051.txt
ec50186cab51b5a1e0fa268404a418fe
c33a1ab2922f2e0afdbaff84650bb7fa61693292
41313 F20110109_AADCTR lebrun_d_Page_048.pro
14a80666472d1095e4fcb94b96c20302
9d1a4ff1ca4e5510a7236f5356e5dbf81e5ce049
85478 F20110109_AADDAA lebrun_d_Page_034.QC.jpg
bc78e49842320f26f0c10fff6954de73
c272f20595e96cda259d9b5d075e982b8e7bc60c
86693 F20110109_AADCUG lebrun_d_Page_053.jpg
21182abcf022d80429728e2e56733b07
e9c5755d5bd505258a766538ba55691678974e1b
196051 F20110109_AADCTS lebrun_d_Page_033.jpg
1832f755608e258aa4b579aa4afc6134
3d8824677007155fc1e5b8197854e72165fcc876
168524 F20110109_AADDAB lebrun_d_Page_045.jpg
1722bd678ad126cb7281ff2fdd50e667
fd5e29c40a8e1a990ed453ac314610995521f13e
16608 F20110109_AADCUH lebrun_d_Page_056thm.jpg
9a31b55c1abe83986ba235cc826e6750
e0e6df5edfb181cc3c6b56ab7b6d0f5c2358ab27
16148 F20110109_AADCTT lebrun_d_Page_032.pro
2b31e3bf0a29ed8e50cbbd731176d930
6a2d8470e1aa844a506a36f0910630f0136b52c6
69267 F20110109_AADDAC lebrun_d_Page_085.jpg
4fd46d972d3f342797adfa774e2ef850
7d84814d925d50d9d6914e12e4314efd02dd98dd
27478 F20110109_AADCUI lebrun_d_Page_074.pro
e43f422e6c87a135420c171a49f742cc
98cf9e1931efb30f70953de54011e3d4b1c7d163
1053954 F20110109_AADCTU lebrun_d_Page_068.tif
b73da26bb63caab29dd7be03d538bef3
e8a0ab3fe4ce036710d459bdd3e6040c6bb017a3
654 F20110109_AADDAD lebrun_d_Page_085.txt
f2a52034bb018d5df576edacee3eca2e
88a4aa65706f53411c358abda814a87c492dc2c9
15283 F20110109_AADCUJ lebrun_d_Page_085.pro
4af5047cdec997bea9dc5c0cd2891c68
35b3068e36d4e76d109f19fc36f403c004c25045
23692 F20110109_AADCTV lebrun_d_Page_102thm.jpg
932aa9f016e64445225bcaeb576d5711
36c26ac4bb68337e541ff6d68d897bff3c9ead21
896882 F20110109_AADDAE lebrun_d_Page_099.jp2
ca2df3bc4b70ca8e3aa5b7feeef592b4
67f7d6cbcfc87f75dfa2e93e8a6612a95569fc3d
6312 F20110109_AADCUK lebrun_d_Page_027thm.jpg
0f281cf8111f731f18a95c947e6ac6e9
9bb443eb9fb528ebd88832594bc88896ca1b1034
7100 F20110109_AADCTW lebrun_d_Page_003.jp2
a24d0bae1fbb2e1a6e81c2fd0bb2d722
99e9053ff96a13235ca5562d596e6652ca319d07
189081 F20110109_AADDAF lebrun_d_Page_084.jpg
f7e8d424464302f9fe24ecd05662b937
51676f64bae0d1ac8d6dd0fb2e2a49e3d67461c9
25271604 F20110109_AADCVA lebrun_d_Page_090.tif
523ebe3f02451ef21fa85024f3224eab
83842ac63c452f85abdc7c556aced71b2da6a86c
16337 F20110109_AADCUL lebrun_d_Page_091.pro
2a0dbb1fb7d7abdb46e3bd03eb0683e2
c3b57a6efe97e40bb74c6b58a7d00744f3b5c4c5
48650 F20110109_AADCTX lebrun_d_Page_008.QC.jpg
99bdea5d85d408d6854d06177a839444
549a0238c5ae27fdb64496a706afa59b67096799
146221 F20110109_AADDAG lebrun_d_Page_051.jpg
d5207ccb2cc8830f688ce1e753ea2ea9
1fc14eaec0e4b64a33f10f4a8fba166bb818fefa
21424 F20110109_AADCVB lebrun_d_Page_050thm.jpg
d7c23e7d53bc366acf9940a9f4008c1e
5ba42ac09c95eec2893272d62f318c57786ccaf1
42946 F20110109_AADCUM lebrun_d_Page_068.QC.jpg
4363ec9a652c1958ce778276e29c1262
60974742f53c31d65deb10e0a0628e15cfdfa4b2
F20110109_AADCTY lebrun_d_Page_094.tif
dbcceefcd151b7121a445a2e384bae81
ef4cc900a972872aa97f10ce7328566b5e58706a
44248 F20110109_AADDAH lebrun_d_Page_059thm.jpg
8b3cad7b0d776b278cdc64cbdadd4576
23a3363030708205afc54329fb285cd48e205590
1791 F20110109_AADCVC lebrun_d_Page_024.txt
70873b74b97c3898e2668aea477531d4
e35aff5fccae617d3a7091cb3418704672b852d3
67328 F20110109_AADCUN lebrun_d_Page_094.QC.jpg
84d986b0b58c261b3cb1373d243c6e18
1b5e33dc455bd5d5ffc93d4baff2b22a114d10fc
1560 F20110109_AADCTZ lebrun_d_Page_040.txt
6ad3b2e309393abd723788c315420db3
f8438520748121eac109d919ea7f42554fccf960
49760 F20110109_AADDAI lebrun_d_Page_056.QC.jpg
5d4d70126a6efbfc0305d8d60e39815c
be4efd35b11e651142f9e361571dae25af00f518
1524 F20110109_AADCVD lebrun_d_Page_057.txt
61577be4b54e28271920e5b61ba1e4d3
be05e5212ea03d8c7cdb6a2278bd4bb8f791344d
1573 F20110109_AADCUO lebrun_d_Page_026.txt
58d37f7a493f49d1e22acd9dbe247831
26a4bdff640812a780f9294f469bbb3dbabf11ba
94831 F20110109_AADDAJ lebrun_d_Page_063.jp2
5e826f1f9aa8d2300066fce18d407f03
3d113a1c4d93f45c891b94e7dfb26ccd198fd2dc
1571 F20110109_AADCVE lebrun_d_Page_042.txt
037baa990572014586a75e39dbddef6b
39b63bc5abaac33df62079275c6b9d9788b0c8e9
172958 F20110109_AADCUP lebrun_d_Page_088.jpg
c715daa3608111902d53844a2481d5f8
75b78920e2d86663a66fa333463eff0efc75284c
186619 F20110109_AADDAK lebrun_d_Page_063.jpg
23192fa9db4fdc5cef14f80437045808
d2cf77f9e4c654d919e4f0311c84ae1cff364705
72571 F20110109_AADCUQ lebrun_d_Page_064.QC.jpg
d790b9f75a521981851fe6d65550fdb5
95e3597c63515224a62f4990b37672300442aa13
49587 F20110109_AADDAL lebrun_d_Page_069.pro
8bfb8949ee65ba45033a657af497b959
395b3896bd56274d7ae613ce7dcbafe4dbb049a9
F20110109_AADCVF lebrun_d_Page_040.tif
e21bc86f6f90c8a318327e333108aa2f
caacfd3e9131cf60529a0c9d519eb6471d012c75
96333 F20110109_AADCUR lebrun_d_Page_050.jp2
b64454431f5f12c8f014a5296fcd6721
a19880e8f3290b5fb6902aecd32ff6d13e7338e0
1175 F20110109_AADDBA lebrun_d_Page_002.pro
347e23d7a4a280dc070d59b7b97e831a
3c3c8b763356df20885dd7b1301c3e04cfaeb996
82122 F20110109_AADDAM lebrun_d_Page_074.QC.jpg
f3022d3aed9fc4a65875ff84c4c86bc0
5034b011132aa6a7b58b15c70060758ad0204c79
F20110109_AADCVG lebrun_d_Page_038.tif
aea6fb53e35b31df65606d607b946c59
cabc8c152be1ead83730b9310499b7eb6feb925d
197638 F20110109_AADCUS lebrun_d_Page_044.jpg
522662994f57da5be1d5ba5beb67698a
2317590e8c864cfd3414e7f5d01374b3cf216861
31417 F20110109_AADDBB lebrun_d_Page_030.pro
8bb4f3a29edfb1dc7cbcf922959f7dcd
e4b671c4bfd30a4a3a8e3c0358aa5d21ad5cef02
72778 F20110109_AADDAN lebrun_d_Page_100.QC.jpg
3bed3a2bf0f03f9296be5bf73f75c662
90e5367de0e74d0823515bce3580913c1bb31902
16286 F20110109_AADCVH lebrun_d_Page_027.QC.jpg
1fac0b1cbb5ec97cc1900abeda16caa3
0603e5ef0512c1b267f97224e51089d74c4183e3
1925 F20110109_AADCUT lebrun_d_Page_003.pro
1d071cba4e9ed0341ccf36cce5637b4e
6e6f5f446711cc4da31e4283667d92e882b9b733
120065 F20110109_AADDBC lebrun_d_Page_016.jpg
a88c11fc44fc31572bc4a0bb9945be3d
21be655fbc38cf6975d8ef292246bbe3ecdd5d16
664529 F20110109_AADDAO lebrun_d_Page_041.jp2
e83809f83e797473cfba1ca71af41469
b03f7ed09c515ebb25554cc4ea48a2d9b16b299a
531 F20110109_AADCVI lebrun_d_Page_008.txt
78af6616893fc59a8d1ff95d30aed0ad
1e43454a33fd06068693d515d31601a39feb0aed
43499 F20110109_AADCUU lebrun_d_Page_063.pro
c713ccf3151dc9787ca6007186646a35
5ab1f1ee24c439212c4d795d3d8ff74d29eccec8
12626 F20110109_AADDBD lebrun_d_Page_055.jp2
37e02dce9d846ac0fed0ad03c00a3815
1e4b6862a15b59bc261f5f07b977505d1caf0ebb
205916 F20110109_AADCVJ lebrun_d_Page_014.jpg
8e836f4b84caa52d9315275b20427ec5
5c453105cf99e6c6a80625b3881a0c739df61d56
72516 F20110109_AADCUV lebrun_d_Page_067.QC.jpg
5ac1773231975c5cbf318d015a2c0b95
2a48df4148f412108d000d3ae345b504121b7e6d
24909 F20110109_AADDBE lebrun_d_Page_101thm.jpg
f9ea225b054a58e43040e910808dacf8
156a85eb5fce6db78996c8bc029caabd47d23ed1
72470 F20110109_AADDAP lebrun_d_Page_087.QC.jpg
4edf4edc38c8fc05c9761c12fb617bed
d749820a70843028a4bcaa22c141cbe45c875583
566916 F20110109_AADCVK lebrun_d_Page_082.jp2
64037abb3262c85591bff5317ed6e940
deedabd0a8319a6b0b6441412ade54c73f4c4448
607505 F20110109_AADCUW lebrun_d_Page_061.jp2
8102453d3468d0f36994c1e18b836603
e3324e5278232a20855eda6b0514b6d6237388f3
961869 F20110109_AADDBF lebrun_d_Page_077.jp2
5c2475a239811af33b34d3c34343d5de
fcc5e2a0b4d56fa849a6c2af2057bb876d5d0128
973554 F20110109_AADDAQ lebrun_d_Page_038.jp2
a548b723cb73dc58f290db2858d68ae9
be5bd1a7883fdc14c3f0d41a3b2dd8a324038aaa
1671 F20110109_AADCVL lebrun_d_Page_086.txt
85443052002e735b6decb62476729298
eab12d3d6c3857f7e2615e5b21a3b6dd1739a5d0
992857 F20110109_AADCUX lebrun_d_Page_091.jp2
84443b6e836f21e9831f63fb104b100e
226d13d738495996d98e0898f1568e4dc6f2807e
F20110109_AADDBG lebrun_d_Page_039.tif
6b4f356e97f213dc25723420bca77e5f
4ae3ee5326f095337f0319392f5a177ccbecaa06
F20110109_AADCWA lebrun_d_Page_087.tif
50b1a80f84b3a34767d78835c3bc09ff
9ce4dbd34f173b6d4458bb9d332c6fede5a84125
1845 F20110109_AADDAR lebrun_d_Page_087.txt
8aaa47509adb1136c8cb4a275e53dcb0
f84cb117e270bf132d9655eb158edbd0f8ca0300
5757 F20110109_AADCVM lebrun_d_Page_002.QC.jpg
5f3def145bb7e419e38e2bace9cf5185
fe4a00284d0281acddf664ddb472cfc5dea2044e
F20110109_AADCUY lebrun_d_Page_042.tif
948719b8e2e6596b62e7bce8e080d7a0
0dc460de5599ac340c0c4a95b8c3a108837a8fa3
4025890 F20110109_AADDBH lebrun_d.pdf
8594a671f67361ad8731b5c4fcbeb9a4
2fa961afad4effd2a05c9f627e399d0394a36e1d
F20110109_AADCWB lebrun_d_Page_058.tif
17fe0bc69005aa8e26e0143d4e898cde
86f4f18831b04b40131081af26f34a8d53b220b4
7084 F20110109_AADDAS lebrun_d_Page_001thm.jpg
5b73ec7098540f83bfb3b01016ba36fe
45ee1f203572b38390227cce01b61f3785f1c300
29560 F20110109_AADCVN lebrun_d_Page_035.pro
9c95cf5f3a2de1401296c9a2d9a70896
ddb8c861ca1db2d11ae20c47be21387940782327
108510 F20110109_AADCUZ lebrun_d_Page_014.jp2
e102d8a37e51f712b9a2f70fce5540db
022e55dea5ff5659abb89f796619c95c36f6ace0
72023 F20110109_AADDBI lebrun_d_Page_041.QC.jpg
7a03cfde10a15b82fa013c9bf12ded03
655427f1d997168570d2b445c9bf69cda76239ac
182879 F20110109_AADCWC lebrun_d_Page_037.jpg
967c4a7bd52eef43140d4fafe298af4e
dac9cae8f08ad37f568f73204d50d776823d9bae
173337 F20110109_AADDAT lebrun_d_Page_083.jpg
e7a6291c0012d62c6fe8e22aaba69ce8
745f93d00cce41580930d800f0a79ad22132e166
818419 F20110109_AADCVO lebrun_d_Page_094.jp2
a6015f4f90b11788e365cff5d9aedd90
97c6d776dc88c0f93d09409c487fa576e2c478ee
13653 F20110109_AADDBJ lebrun_d_Page_097.pro
48f5ee2219cb71feee54df711e7b0e2d
cda0b6e579edd178ded2d2e75d310cd974049a49
F20110109_AADCWD lebrun_d_Page_102.tif
f8574b8c5d09e8a6f5112001bbb5689c
e5a38d8b04a8199a720d3a837e7b5fa75cf7a3dc
19027 F20110109_AADDAU lebrun_d_Page_093.pro
d76309483d635935988ec1d565ab2a90
26d866e61b76c805b3fef694406167edb8db2b9c
80755 F20110109_AADCVP lebrun_d_Page_029.QC.jpg
3adc474c0c41ed60ad8b59fc5020fda2
f93913e4c1b7d059dc89424e846ffa99f5bf21a9
60285 F20110109_AADDBK lebrun_d_Page_060.QC.jpg
4821ba74d6267868836d6303b8425fe5
4897f9a174bb402e4f888f3da2e203a3afc24ac5
10058 F20110109_AADCWE lebrun_d_Page_032thm.jpg
a84ed156d6accf9cd5ded626836e2a5d
c07ae0ccd64900463ab55d65717754db33ec737b
8423998 F20110109_AADDAV lebrun_d_Page_016.tif
a5f80aac0de56e19ce61973a80f2481c
9002ad89d96a0144f5384321c254774b26193f93
F20110109_AADCVQ lebrun_d_Page_019.tif
e929a504968220fc6313ecd696c4ad90
f202a964a7a769f41dffdc248e2eeb978ad68692
1341 F20110109_AADDBL lebrun_d_Page_065.txt
42356b508abd72afde9677245f59be16
50316ef8b8087d91473cf74ae63956a6e01481fc
133287 F20110109_AADCWF lebrun_d_Page_056.jpg
c271f331fa7d22a786792b93b0306a5a
087e4352725da5c039c6ba37f5b59fa14ae2f186
132420 F20110109_AADDAW lebrun_d_Page_034.jp2
a88a865e41839dc74c105dad35db58fe
6390796b83095bf7b20af3da53f75faab4ca3a6f
3786 F20110109_AADCVR lebrun_d_Page_006.txt
d3a73aca49fc3a7df5fc328950b017df
5c40001305f267e91d6ad0659eedca9b30a12504
1413 F20110109_AADDCA lebrun_d_Page_054.txt
c2f801639e4558fc27eff97a1a0a60ca
122a731882d0ed5c41112bd77dccc8e695f53563
26610 F20110109_AADDBM lebrun_d_Page_055.jpg
9c9e31ecb1bb87e7104dba9de95aba06
7bb7da20e773b1df73a6d7276bb7035e4840b689
51362 F20110109_AADDAX lebrun_d_Page_104.jpg
41f83f221f7a0d978b4d04f49fabcaab
d1711ac5c3e2b7db9e97d93d7f058843a5ab880d
62749 F20110109_AADCVS lebrun_d_Page_086.QC.jpg
9539212f9f2d24b3fbcfecd1dfa08be4
4931ba5330c08aeff129f653291687069520a55e
74135 F20110109_AADDCB lebrun_d_Page_093.QC.jpg
3a0445223cc89559613f82c1f928d3a6
128e632e298c98db2fc4d23832f4acd423391e9b
19864 F20110109_AADDBN lebrun_d_Page_104.QC.jpg
9cffdb7029ee751e07125fd959f8fb6c
fa7a0ab30ba415c6c5d219b38d8f62f981395ea5
223835 F20110109_AADCWG lebrun_d_Page_100.jpg
12ced18d15a1a6f9e300fe08c7e6427f
fad5f64bb70f0184ba5c29754284f8327b7cd696
166916 F20110109_AADDAY lebrun_d_Page_095.jpg
ba7c063bb462734a57838986ae0fa886
05c384bb8019f5eea154e32ada6a9fc530767ed7
23668 F20110109_AADCVT lebrun_d_Page_018thm.jpg
70bc4f16a3ab69fd9a2adf9d66b268a9
20ce38e25c6e66c91f7fbe5e10d1ed568aed1ced
F20110109_AADDCC lebrun_d_Page_046.tif
7129bc8f43368f66d5d64d3679228808
c1fefbe42570c71de65dfc908cdfef61a60f30eb
1833 F20110109_AADDBO lebrun_d_Page_050.txt
9428ff9594311da123eab466c877fa47
2277640c8678fcd811f67f644b9bfdfbebbeacfa
148621 F20110109_AADCWH lebrun_d_Page_071.jpg
2bef9882fc94f4cc041f593cc77df355
b3c8cd0c0ac612309982a1b49b255859ef32e8c7
24922 F20110109_AADDAZ lebrun_d_Page_069thm.jpg
f94965a3279cf411f4f46b130ddaec8c
cd06bc19e113becf454f8caf7f0ba76eb92f2c68
F20110109_AADCVU lebrun_d_Page_085.tif
5116c346398a62057d238e8495756b9e
119182cf535c8c94202c1d7db6d68322b47a523f
20629 F20110109_AADDCD lebrun_d_Page_092.pro
c1b4d9b8e5fdffa4a89878bde63c9ebf
8542526b59c53c9915e32fe2775b1c5301efbaea
77562 F20110109_AADDBP lebrun_d_Page_043.QC.jpg
177c9323f520ca78c3531014846aa772
7f4400c69fb682c1a7f2bd0bfafe75fc021f4eab
F20110109_AADCWI lebrun_d_Page_013.tif
bdc4c903d3b6a03cb13631c7015ec9dd
88200c1448425ca36d20f5720741ae66c0e38f30
F20110109_AADCVV lebrun_d_Page_070.tif
f2874483748f848cd49a1e97e7f2a937
f147df0a85dfd182033a7aaee5e918062c8f4068
24409 F20110109_AADDCE lebrun_d_Page_014thm.jpg
6b5b322835ee3927af0b03835a3296fb
0cd058cb121e8cf3b8e01f792e2d2170b138e4f7
44201 F20110109_AADCWJ lebrun_d_Page_084.pro
932c19c2354598a5797a87e5d609a5b7
6c1d75cdaaecfafd4074487e99d975b252bfe675
24145 F20110109_AADCVW lebrun_d_Page_020thm.jpg
0fd7a04db8bb705c9df0ec3a3d1db8bb
f9e49514e37258c9561e4c759243b0e7a4e6a064
52817 F20110109_AADDCF lebrun_d_Page_001.jpg
da4aafa855a5c35e1ffdf5251243e376
989f9bc449957abac7574b9eedd8c9d47492bca7
108337 F20110109_AADDBQ lebrun_d_Page_058.jp2
b63fad7fc275f8136dd53453ce7cfbf7
69c6c197e9f2b1a855166248ba7dbe75398f2679
F20110109_AADCWK lebrun_d_Page_083.tif
7e7640401797813de0950cdc24cfa955
6c627f8e05c191bd1234dddf507892c482038f18
850946 F20110109_AADDCG lebrun_d_Page_026.jp2
873b836a65fad08f09e87a46328a44ed
445c9fe3b8d88ef01ec883d7d9ab04ab4cbcd258
714516 F20110109_AADCXA lebrun_d_Page_030.jp2
975451e88b2d377ad0ef86f350d4240d
ac5b0bcd6de84231956a1e6556dd97dccf0ee7b4
26993 F20110109_AADDBR lebrun_d_Page_070.jpg
45eceb943f7749b8856fd5cc0128b21d
9f36e4b253070ff6af2d1ab8591a53f836ecece1
104216 F20110109_AADCWL lebrun_d_Page_006.QC.jpg
94d7c4271013a61dee4585b7f98d875d
07b29f0ac225a31c55c0b1ea40e87a1cf42c7f28
41814 F20110109_AADCVX lebrun_d_Page_091thm.jpg
4d1c1d418e3cc1b9bef79737fccd5092
8c82ad0d9563b61ecca6080ae7b7870fc7bf677c
89641 F20110109_AADDCH lebrun_d_Page_005.QC.jpg
9c3dd6942983b4f9df0e22b9c2a2ce6d
bd9f57e5ed312502231148bdc84ed7d1c24e27be
1051972 F20110109_AADCXB lebrun_d_Page_005.jp2
291891a056e63ad364ea11f9130e74c6
72cea90dc6f6ac0b8417e9751ea9f6bffb99fc0e
821526 F20110109_AADDBS lebrun_d_Page_098.jp2
6fe29b265465a56ec680eec99054d035
b10a787668eb1fb24317ec62e060560f5c6d3d41
98875 F20110109_AADCWM lebrun_d_Page_033.jp2
f990eb311a92b3dbd45f99620cec79f8
e3f1e31905af9bf15741448f262a4b1efa8fb2a6
72271 F20110109_AADCVY lebrun_d_Page_051.QC.jpg
cbd16b981e866e49f82b84934ee8d753
08d994c55fa469257c0cf0be9d7a7de935fd8258
F20110109_AADDCI lebrun_d_Page_050.tif
7cd404d852e70af12c15ae7b7c8ac7b6
4ce7fc89ace88422542ef33bebe96976d007ac69
415947 F20110109_AADCXC lebrun_d_Page_007.jp2
12eccfcdd084a1f2a72529dcb143934c
8fff879534244307c7bf0696f659f653b48cd6ff
51534 F20110109_AADDBT lebrun_d_Page_025.pro
43b9396e344077e719f6ae8fc0ffd690
a84c202ccce37a509f601d0b17b1c614b8738602
842 F20110109_AADCWN lebrun_d_Page_041.txt
a31c082335e3fa42984be10a4a1a5da9
d840f23f46e21f2e663643bcdf41f504b9e55e6a
160143 F20110109_AADCVZ lebrun_d_Page_047.jpg
1a0fdac54e0a14d2283ce4248674227d
822bb971bc62b4dab11dbc176c4715e4c899c001
22352 F20110109_AADDCJ lebrun_d_Page_062thm.jpg
82445f98200389579f13b52b976577b4
a985ce87416c70772142760ce14098c90fd7676c
107886 F20110109_AADCXD lebrun_d_Page_017.jp2
8ec056cd21c47536820ee5f3992f9b5f
5888ab6d89edaddbc301b5a41ed70f8ed8082c25
76881 F20110109_AADDBU lebrun_d_Page_020.QC.jpg
f423dbc9d6ca961e8282042491127804
95a51f685e138416225cb3c7fa8d55922db406c0
82139 F20110109_AADCWO lebrun_d_Page_035.QC.jpg
9090798e9f0c376c5deb02e62f314a98
fad249af1334e9a20e7cb24b5b8be8afa7c1db4e
21422 F20110109_AADDCK lebrun_d_Page_048thm.jpg
2556100808b627f928dc662243c9d931
ada617c4991feeffe522420c4e9251eaa06802d4
4632 F20110109_AADCXE lebrun_d_Page_070thm.jpg
094c3754dd3a5ba4d37a96e869077abc
0c72338ffb93bb6dc94c7968bdd319c9b7e93c6d
44246 F20110109_AADDBV lebrun_d_Page_045thm.jpg
b971e7940b5d728ac07baf844f75973d
e1b6ffe04e2a08ad4913de6eaf7c829ee12ef069
1051937 F20110109_AADCWP lebrun_d_Page_090.jp2
3bebdee9b9d799b9ca7746094dd62151
6016375a6e680f50601cfdf4d2e65e21011d792b
33975 F20110109_AADDCL lebrun_d_Page_078.pro
5f4c414be567fabef25c7340593ffba7
9db24300af97744262519c1c176b68ecb57209a7
32663 F20110109_AADCXF lebrun_d_Page_008thm.jpg
0f7020fa10ebf82e3901630d06d40916
c9663ab492d2222a7238548d7d174ac6a033fdde
19158 F20110109_AADDBW lebrun_d_Page_071thm.jpg
b0f66ab24189bf1caaa6db3398b65780
5a14e1ede768e44b8d832d6e95b8aacfa14176be
862 F20110109_AADCWQ lebrun_d_Page_095.txt
326ab5ef33f9f5661131a85a99bab577
783e3dc8ded0809e5b5a3a5d62fbda830301a052
47820 F20110109_AADDCM lebrun_d_Page_016.QC.jpg
b832b8791b0a784cc5a0bc5382810506
c76d6ff65de547db8cf827a642449f519dbf0254
F20110109_AADCXG lebrun_d_Page_015.tif
706fa641eda26b22b8c7b3bf7791fe30
171b29182b571fcd5248ad12b4d960d122a0712b
78129 F20110109_AADDBX lebrun_d_Page_031.QC.jpg
0eeca3804106f31c445cfc11f2d0e678
5d0f85ff431ec4410e6be1e59eabda114f56d116
46352 F20110109_AADCWR lebrun_d_Page_087.pro
dbb3a38cae716cc962069e186c087698
df8372c737f0d265ddf0b2ac9634dc664267723a
29859 F20110109_AADDDA lebrun_d_Page_081.pro
ec575e7ff0ab8c36c94c707b120b8434
43fe409ad2c2c5328dbbf9e21398926aab7a0cf4
193884 F20110109_AADDCN lebrun_d_Page_064.jpg
7e12bf2fa969c9e5b25fcdd6d7e1c1cd
c5077cedbc5335d59e362eff456436aa0b467608
93685 F20110109_AADDBY lebrun_d_Page_022.jp2
cc422bbecda9a8d25f83fa8dd730b7dc
0bd7ab288bc1c156a1e32f35629d431bb5489424
23563 F20110109_AADCWS lebrun_d_Page_064thm.jpg
a26bbd3c9ed5c60b3d46b8e09a68a584
9babd57174754fc568f2e6b25b73d7b1851383e0
10840 F20110109_AADDDB lebrun_d_Page_096.pro
71fb29108168ee2cc4b77e682fce5117
13cc2010f627a4e30cc629893b76c60c01e9d233
49008 F20110109_AADDCO lebrun_d_Page_006thm.jpg
6bf7b8530e488dfcdaaf4984eff2fc23
c37609f8c1b82dbf5e37f188f8d434ed63fa4da1
81102 F20110109_AADCXH lebrun_d_Page_101.QC.jpg
7a8f670131ee1c5a7b8b16cfc8330a10
39916fd01ebca340ab3ade233de7eecb0b6c059f
1648 F20110109_AADDBZ lebrun_d_Page_083.txt
7b880460d2fddd6cf92cd4e45f942341
ef2dfb584983a6eeaaf0e09070d859e9ed51d292
40671 F20110109_AADCWT lebrun_d_Page_094thm.jpg
9f3d9daf18d8c021f6b3d4632913d6d2
ab2d2877eae689d18c20782533d440bade04e151
212023 F20110109_AADDDC lebrun_d_Page_058.jpg
69fa83f1e1c5f498cbaeb3bbbde0e47a
0203fd2b36ac00599980049d102575a76084b0a6
1832 F20110109_AADDCP lebrun_d_Page_063.txt
03dbf48dbbbd99a191ccf9e9bc5362af
5dde7ba7c88e5b6a1266edbb483e41606bc639c1
828 F20110109_AADCXI lebrun_d_Page_097.txt
dcdf47c6aef713bae79f494270e833df
f81107995bcc8afa66a48fd4aa504fc20456148b
22417 F20110109_AADCWU lebrun_d_Page_067thm.jpg
f125cc1ccd1ae65128d07ea41c45af6b
eb028b42195342a08b887ee61038a500070dd7b0
F20110109_AADDDD lebrun_d_Page_066.tif
6a39e215f56c8d3583e7cd71dac9b461
a76e460a00a2eab0a4085ca15f4f1ae3df71a515
F20110109_AADDCQ lebrun_d_Page_103.tif
32b8b0bb8b6c1f9a1db0fb8fac8697cd
352f52331469c08499b3fa0aa56b599ff773c7f3
1726 F20110109_AADCXJ lebrun_d_Page_066.txt
87f847860bb7485a49d1df947fa14667
2f9e06856f46b2d759b495abd918b54152589c90
1494 F20110109_AADCWV lebrun_d_Page_056.txt
3525384091e58875c5a5c8b8144701cc
7c2a63c1fa37cb18a6a50b505053345b70dcdd25
1238 F20110109_AADDDE lebrun_d_Page_047.txt
b840f6c02c68a7ccb5ac134373e829c4
9d1d137fbab054a64e2af0107363c607304d7e02
27393 F20110109_AADCXK lebrun_d_Page_004.pro
580a371ec55ed90259827e1fd4608e4a
6c4f56888c02d2f29d690a3c9093eece5830bb1f
23151 F20110109_AADCWW lebrun_d_Page_065.pro
736b9a4b64a68eb9a9b0fdaf3450b71e
a9f7c44ff245b97d6f59ac83758dfd00e48e9c07
F20110109_AADDDF lebrun_d_Page_051.tif
9cea6f029879aa49403b0af6718801e1
30dde245b33b712d76afd653b7b1e4b794abbbfe
19196 F20110109_AADDCR lebrun_d_Page_016.pro
a5e62067298102332b822d13458df915
51b20f310c7c443116672e1d571ac1eb201bf5d2
87086 F20110109_AADCXL lebrun_d_Page_038.QC.jpg
6aeb6378d8b66bcdae4164a8cd80576c
1266fd69818cfc00ff313d4ab3e1e0c331d9d323
214 F20110109_AADCWX lebrun_d_Page_070.txt
259544594706beea7ad5abc393d699b2
dd971bf3e93e7ae3c04d9abc190b9f2306ba0bb7
121148 F20110109_AADDDG lebrun_d_Page_004.jpg
332a7cca1d41e7d97cb25a89d8e31ceb
f43c088f324447fa86601ab8ae2048fc94489649
114643 F20110109_AADCYA lebrun_d_Page_060.jpg
29eece87627720414b36786c698af0e1
dd0d0a8a4d1580c1e4942de8afc13b0108647923
24802 F20110109_AADDCS lebrun_d_Page_080.pro
b31eef34851953f28969480189b25cad
cd3e7d35c4960c6cef55cec50f032ccbc8117eeb
22337 F20110109_AADCXM lebrun_d_Page_024thm.jpg
5a61a647eb2983d7936103ed077eedba
01835079f907349a4540e3c85aaa85c0b86d30de
47886 F20110109_AADCWY lebrun_d_Page_019.pro
411f3333617722349e8ccb6a7df759da
2979673e7264957a6029f3bea6f52714cba651d6
622486 F20110109_AADDDH lebrun_d_Page_046.jp2
1d6d231f1d324e6e7c7ecdf2f054bb07
3764913312e856ad50cd673be7ef1fc37e6af04f
19195 F20110109_AADCYB lebrun_d_Page_057thm.jpg
8f5802b1fe62ce4c61b8301c581a8b35
3ef02a83e7959234a0c875b7ca8089d4405b388a
54752 F20110109_AADDCT lebrun_d_Page_057.QC.jpg
8da7d3e06b9e6b14bd9ba1644f9be74e
5ca5f9267eb96db31e63d5ab54070efa2fd558b9
4366 F20110109_AADCXN lebrun_d_Page_070.pro
52e0d57840b75b6f3c35c3588cb62b69
e62d3e75079b3a4b8f932c604912ae41369c20a1
23453 F20110109_AADCWZ lebrun_d_Page_046.pro
8fd8c6572d348c29246995917f10434e
85162e6f33511c4c3f46a440217432badb563a3b
F20110109_AADDDI lebrun_d_Page_063.tif
209e8a6597e6f27f161cdb8388764730
e388cf210a54e37411c6ba412595da9b681863e5
188857 F20110109_AADCYC lebrun_d_Page_067.jpg
0710acd9789ba0295061d32282330117
c278d85bad143c401083745269a491d3409660fb
41989 F20110109_AADCXO lebrun_d_Page_052thm.jpg
07c013e1fd512c792a2faa6730d6daef
17e040740acc67a5883e3a0f6f5558a7920d582c
F20110109_AADDDJ lebrun_d_Page_008.tif
a89eef1c649109e4d6e9638e3feaabfd
6e2d1b9d6c51a6086d4da04a7898a1ecdc7e1cfd
3094 F20110109_AADCYD lebrun_d_Page_002thm.jpg
1403f38e6b8851c0d75b9b3acfd0badd
872ff906a6329dd612edd66b42fd8ee6e4647a2e
45506 F20110109_AADDCU lebrun_d_Page_026thm.jpg
05a3e80111056238b1a2d62833a38b4a
6900a93a3856d36b0226c72d6922f101f414c51c
1051982 F20110109_AADCXP lebrun_d_Page_006.jp2
537b43b03384c13bab2eb353fdc00646
d72883dd9f0ef82c4aa8f0974338c29e00210238
17574 F20110109_AADDDK lebrun_d_Page_049.pro
89981a294b12df66783a897ab7af7329
a4019fa984e298d298ab6920fdf0a5e1f51cde04
3058 F20110109_AADCYE lebrun_d_Page_005.txt
b338a3467b799aa1f9b8eaf8422a5e53
42c9e699e18c89b65866c8e5a2b45f9fc0440ab1
143123 F20110109_AADDCV lebrun_d_Page_046.jpg
2edba80c8043ad5196a0d605b6a255a2
aabba7842b9585a3b801f48f3a12708322b7394b
43480 F20110109_AADCXQ lebrun_d_Page_013.pro
4d2630f674c23027b0c27a9c50b7eeda
1fe50b2eacc4baeee0098c37a86a14b0f965d19f
1745 F20110109_AADDDL lebrun_d_Page_059.txt
7b87cbef72a54d68420f798557830a0f
76f8d91e2aed966942bd6d1154ecabff41feb567
1253 F20110109_AADCYF lebrun_d_Page_092.txt
a11617a4f1fbb754a19e9a868e04d177
63e3a8048c38c87a8399198ebcbc4bdbd76b3976
52869 F20110109_AADDCW lebrun_d_Page_100.pro
43af4125d6141d76aba491041a4e0faa
528de09a29de99cb8c95f7b09eeea37f5ed324d6
15918 F20110109_AADCXR lebrun_d_Page_065thm.jpg
9e808628cb1dc1ea3d588d3dcc660660
446d9515112f18dcf4afdd92705cd7f6c07db2d5
41131 F20110109_AADDEA lebrun_d_Page_083.pro
001bf81c4c45d71a25123431dfb67beb
60561a132f4ec9e99845470c6f7a49e798a512df
F20110109_AADDDM lebrun_d_Page_077.tif
fd074aed355efe7df1ba48e1bfd3ac1f
b0bf67514a0eeb6f00b49b6e25689ee661e96769
2023 F20110109_AADCYG lebrun_d_Page_025.txt
84f1e98da96969c823919c76e290c631
a34e7ff9c6e61b57f2c21707129df568850f747d
1821 F20110109_AADDCX lebrun_d_Page_084.txt
156dc68a1cf9957ccf356c6f80b3277b
e5c4c574b8a80edd2852a600b0a7222cd1df8e62
1945 F20110109_AADCXS lebrun_d_Page_044.txt
04070dac1b31b13a528b38f57e4c49a8
3b7dc1887499d3b99f681949edd3cc9cc057fe86
F20110109_AADDEB lebrun_d_Page_104.tif
93b11a7c0a6668f9dfcd6b1f0a6566cb
38b87f444dc67d2b24868a3f56959cba78a317d0
F20110109_AADDDN lebrun_d_Page_004.tif
c466e21dbe58ce80e88dfe56338bcd3c
81a1f1c267d358d22518bd4efaf9600bb95dcc77
39157 F20110109_AADCYH lebrun_d_Page_099thm.jpg
f3dc73bef43083e4dd6e1a3cd90f2b3e
43cff5fa6141e49f75bd7b13ebea329b77d32938
1440 F20110109_AADDCY lebrun_d_Page_068.txt
077d28f73f68f68097b96f3a860101bf
a611cda458f84c87ed56b80abc1d13ea8ebbd017
478 F20110109_AADCXT lebrun_d_Page_053.txt
96ade616c5c974d9e29e7ce9d6b72261
4e5f577a73b41a1145f8e7fa4b571cb66909d439
1242 F20110109_AADDEC lebrun_d_Page_061.txt
e19afcbdc0215277ab0f4b51f7488810
94f929dbee18ed09c30d0cb6e1a64ee36b1c2471
13038 F20110109_AADDDO lebrun_d_Page_103.jp2
57cb91c73d671ba7e09ca4c9b3afb8e7
e52ed02a166fe3c8105ee88b6ac46e37925da16a
F20110109_AADDCZ lebrun_d_Page_018.tif
1b4c7c16320e7c1d21273059f385f89f
2ab30de88f5311775ba3519ef21ba5047acbabb0
806619 F20110109_AADCXU lebrun_d_Page_043.jp2
56829f7c2349e37620a7571157bbaa5d
0fa1ab3f83d4148cef313b13e7783d5aec956d24
29372 F20110109_AADDED lebrun_d_Page_054.pro
c0b651f22a98e4b65eb12cb0c114e116
e87d6eb6ae269bf98f2d766e1b4ba6a8f4ba772c
23775 F20110109_AADDDP lebrun_d_Page_081thm.jpg
5eb42dce3f2878a6b73555efb657b58d
4c79004383c16bdb5ed8ee72b4c313ef0cf2f339
24877 F20110109_AADCYI lebrun_d_Page_058thm.jpg
3dd640ed4c7651e63bd4d1f40211bef2
af67a38419e8f6fba7047f6cf2bd946fedc8543f
2044 F20110109_AADCXV lebrun_d_Page_018.txt
91e3e57a0dfe30dd78fffbc8eca16020
f48925d19dddaa2a8cd78354b5e976de96217095
1003 F20110109_AADDEE lebrun_d_Page_046.txt
60a2f81674bf355b43e1d7a3b977cf70
14f27b3beefee839747c5cb5b1e679ef71d762ce
699008 F20110109_AADDDQ lebrun_d_Page_051.jp2
4f20fea95cf11691f964faf3e23e8c26
8b4822e0072caadd5542223ec0f22738b9546d27
158004 F20110109_AADCYJ lebrun_d_Page_066.jpg
88f929406449feba919c8ee0f051d75e
2de71d1190e6a3112b235e477ddfe2c4ed23fe77
68567 F20110109_AADCXW lebrun_d_Page_046.QC.jpg
9bbd2e4af24ede6e8a57a5852b0df46b
4e2793d5acfa2e2d941e121bd1d4e635f3bfbde6
21149 F20110109_AADDEF lebrun_d_Page_100thm.jpg
17369014b32c89b8cb30555649ace6b6
f6bf044618b76827dbec3b4f054e322a2f6380f7
67106 F20110109_AADDDR lebrun_d_Page_098.QC.jpg
bd07ff66cb0337dfe8bfbdc90848c582
aeddde386afb75233644df87f47271ce23f07d7b
882815 F20110109_AADCYK lebrun_d_Page_080.jp2
8764b107fa4aad8cf316520b53484457
64d7595ac1c8717746bf39e74d0e403a2ab6c122
47687 F20110109_AADCXX lebrun_d_Page_033.pro
dfc8c66b6d7ffb54d1fa0577c1703cde
a68a7cb34ced95872a067cdad86a8b34613374eb
F20110109_AADDEG lebrun_d_Page_022.tif
7dab09c5d3188be2ea54035641b1372f
fb734b57b70bc973619015c358fb5cf8fd5cfcea
F20110109_AADCZA lebrun_d_Page_062.txt
384091003b71ac4b0f97c532571de835
5c28f635f0db813f269451de9f46efa580a2ad68
53088 F20110109_AADCYL lebrun_d_Page_029.pro
50a2893a0d6281eb50e57cfe5b1ea232
b07bb70c3d9fb848bf1f7ca52ae3e6cf86d1c548
9112 F20110109_AADCXY lebrun_d_Page_085thm.jpg
32b6618d0f7c254a33376a475e238517
c61050825ddda5a773d4e2d1dc0ce71519611b0a
1565 F20110109_AADDEH lebrun_d_Page_075.txt
20c7627868f43e36ae391c7bd3b5ff0e
cf5c5f0b235aed0c054332847587d00ded01d8c2
19088 F20110109_AADCZB lebrun_d_Page_094.pro
3da85b0fa2f3271ee429ceac43033224
4ab09624e7ccd5bafe1e14d42824fed6ff3035b9
F20110109_AADDDS lebrun_d_Page_078.tif
14f91a0151dda8a48f41ddc98a1f4034
3a812a15b7a6f1595920b8780e42741defb108fd
4565 F20110109_AADCYM lebrun_d_Page_103.pro
cdd4d6ace9fe2f84ecdd43f8f2f62447
5e07ea494b3eef86fc2eebd3151aa72b1799571b
214591 F20110109_AADCXZ lebrun_d_Page_079.jpg
ab062f568b070fcd916b9d2050cdc7d0
b661809b992421295bfb83e029948d449f92e5c2
181918 F20110109_AADDEI lebrun_d_Page_036.jpg
00275b47c402101d3582f698bad40c6f
f165dee4f8d65bfc79d495fca1f375691fbd6ab7
866259 F20110109_AADCZC lebrun_d_Page_097.jp2
5e2dc1e7d60ad69df773dd5663a84fc3
792e537a2439106d4277fe4e07caf3770e34f2dc
22726 F20110109_AADDDT lebrun_d_Page_012.QC.jpg
4b74f7e952370de0452859547d5e0241
aefb6e520f309666d27d2271c4e63bf70884395d
365 F20110109_AADCYN lebrun_d_Page_027.txt
e94766321478c7fb681e69d2e3c0b7c2
dd33073585ff9c55246ed0bc012f825e49bfe77e
F20110109_AADDEJ lebrun_d_Page_062.tif
2ca3fa79de005d9379aef5f7d24b94e8
a4b9cdd8f92e91755366af333a88d47bd15f36b7
64358 F20110109_AADCZD lebrun_d_Page_048.QC.jpg
cce7b034e9311913bbaaec8297f82cc5
a04ed291b43fd9652391eef130db8a0b9d5f6d8a
67022 F20110109_AADDDU lebrun_d_Page_050.QC.jpg
b0d50b3e3e8182068f7da9a245af19d1
807266b17d9e733b6df2b6230c484315810f7568
28728 F20110109_AADCYO lebrun_d_Page_103.jpg
9c1db0515d879f85f3a15191c620815b
81179cbe5496262929588d38239ba91387cc68fc
57099 F20110109_AADDEK lebrun_d_Page_096.QC.jpg
3d10e9c7f07f8b03708184cc9cc84ef2
c3f7b23105edbb95c4b9e20ab4dfd2650d11c191
68133 F20110109_AADCZE lebrun_d_Page_024.QC.jpg
03ba1fa16551c53cc6623e3857d7b93b
2469ca6f566982ef75e5a3cc8cd3a7150c700cad
862727 F20110109_AADDDV lebrun_d_Page_039.jp2
ff9cf68b3c860e6f0bb8b41a9118fa90
3dc191da1b2304e19b443a00a06a1edd0e09c047
222462 F20110109_AADCYP lebrun_d_Page_028.jpg
db254f4d4dc0c30192f04bd193d46d34
13ae4b9be5abf421877f46eef234284ed9551886
75558 F20110109_AADDEL lebrun_d_Page_030.QC.jpg
6700c1c4119ee3a68cb76b41503d0605
8d31dce4c63c717171a937167af6fcf82565711c
2089 F20110109_AADCZF lebrun_d_Page_029.txt
75aead1f3af097d6b5f4e0ec821edf10
f46ee20d59602a691cd2f98ec547c3619373d446
68142 F20110109_AADDDW lebrun_d_Page_063.QC.jpg
8dde15419f0b940dcae699b1174e18d3
fcf607718ba4929308b996a9e78f36c1d0b44c77
69602 F20110109_AADCYQ lebrun_d_Page_033.QC.jpg
06abec4cbbe415daed86dd1fac2f20c0
61176b1d6af38224c436b80df54b5bb86e9da59e
35731 F20110109_AADDEM lebrun_d_Page_096thm.jpg
c6e6ebe00b7eeedf5317de2b0307f500
3ac0d173ac46fc1e5093115f502b7664a15da515
58300 F20110109_AADCZG lebrun_d_Page_042.QC.jpg
9e5d75d9e73495d545ea023be2f7948d
90bd8a8a4c13b6b96978f8b0b1d1e7b0df7afa33
467 F20110109_AADDDX lebrun_d_Page_012.txt
83cfc1a319e48fd1fb0421cdd7ccdc1a
285c44fa4f04e6d130ba11472dc243712c5b2485
22350 F20110109_AADCYR lebrun_d_Page_023thm.jpg
3f41a0e1d052556ebe6e80cd797c450a
9ac56538877cb1222e561c2a0c113fe10448e029
51025 F20110109_AADDFA lebrun_d_Page_079.pro
51a32a0adfb34a4542fb491a33023445
e72cd2e9b2879b543d1f9acc43bf097e3f22a8de
24610 F20110109_AADDEN lebrun_d_Page_025thm.jpg
e26da7a94fc3dac04446b855432aa6a3
7e55b9ac39e2e46c3f620a54dbad2197a4d058a4
438476 F20110109_AADCZH lebrun_d_Page_060.jp2
d330f50ccd048c2edc442ff01c0f2033
9928d8b0f474ab8b6c9f7fe0b03ed92970f309c2
64701 F20110109_AADDDY lebrun_d_Page_034.pro
31c2163ef8351091fc1ef422ffe7690e
a8e4523763bb13bc8b74034722f4fe622f435256
1221 F20110109_AADCYS lebrun_d_Page_094.txt
f547a189ed88a3739e872514a16ed7d7
b8f8617df4aabd887eae1317599f5fa2f285409a
35388 F20110109_AADDFB lebrun_d_Page_076.pro
b29d075f9577ee58afb877d5b528be6b
670ebace14f6d5f29c3004b9b1397c754d1e7585
75143 F20110109_AADDEO lebrun_d_Page_032.jpg
8487b3d61b28292325864bcea94f72f1
2b0f5854fe6b18b8212bab9cecfdc8026342798a
26430 F20110109_AADCZI lebrun_d_Page_043.pro
7bf70e5e5cdc9dbf8633f91fde8ce291
d074e77a34d9748d1ba0cd7f1a4e8d318133294c
69883 F20110109_AADDDZ lebrun_d_Page_044.QC.jpg
50afaa5abc8906584d8763aa770cd066
2b04f5aa6297cc76459036e2a98ee29719f037bc
27141 F20110109_AADCYT lebrun_d_Page_007.pro
9ce8fbc268ee2e4a9bab70b795950fd6
cb5275a7be55525fa02629e0fc70f1f5bb7963ad
46588 F20110109_AADDFC lebrun_d_Page_004.QC.jpg
1102f4f4068696348a36d5b373b6428e
f9506e6df198181bc3d526a4172ebb170a5b947b
74935 F20110109_AADDEP lebrun_d_Page_057.jp2
fb2da8ebe874eeee05c7e7078d04b4cf
3aba87a78d9fed1801c674a186cf3c077ad6c669
45045 F20110109_AADCYU lebrun_d_Page_064.pro
8758f3324886ec523de12ff2b46667c4
98c390b2b0d1b8d577cef9b060d63bd1fbab5b2c
F20110109_AADDFD lebrun_d_Page_037.tif
61734c12668bd985bdde4539f8a9f376
5afaa2edf9161d022d3a7de5d4eae6fd54f194be
36137 F20110109_AADDEQ lebrun_d_Page_042.pro
ef6c3622fab1ebf24eb7eca30a549205
a8fe26058f635e44978990ed1741e335583906a5
46392 F20110109_AADCZJ lebrun_d_Page_074thm.jpg
02043a78166aab643cce7a74a0a2e055
ccb58a1684db50bf8fb0693e7d11e64cc342ac4f
20484 F20110109_AADCYV lebrun_d_Page_040thm.jpg
8276c17645ddec3c6b466471754de5dc
3e54697b00fd032232516ed811febd1ee736036b
F20110109_AADDFE lebrun_d_Page_030.tif
86cc200243ee9e3288a09afa9a1eb172
e78595c7f148f81157d771fd16abac28b2b2a794
20734 F20110109_AADDER lebrun_d_Page_099.pro
773147fb36d56c585217f26fa1dde62d
78f45d703cc00ae32e89b5c28889ff2db1201349
F20110109_AADCZK lebrun_d_Page_074.tif
6dddbb8f094155bada0fbacd0c91660f
52a5884e36a94555e2364223e9be09cfbe1df96b
1758 F20110109_AADCYW lebrun_d_Page_022.txt
b74706c0079dc5adeab7144169eabb7a
166f34bb8d7e9a0bf493f822e25f9a94a5504fb4
147569 F20110109_AADDFF lebrun_d_Page_052.jpg
799ad92d52a5ea69526b7aa99dbb7d68
1d29f8c73df8d7f70bd1b16f644c9be965416fb8
F20110109_AADDES lebrun_d_Page_101.tif
fe0d84965d776e02c9814eaa2fc6b5f0
13aa2546e98b4c7c0facf5a9abd49034c196c62d
F20110109_AADCZL lebrun_d_Page_028.tif
c6c335e8830abdc375ae0bf0027ac4e6
e0ecd1b2606450adb29297e83a34f295bd8a27f9
57356 F20110109_AADCYX lebrun_d_Page_066.QC.jpg
67f9155a9b0a65b78e286a4c2c94024e
f0663afc6daf6716f0d3814ed18ba7017d199755
98030 F20110109_AADDFG lebrun_d_Page_028.QC.jpg
1da8ea47bd27e8c072655a68f0ab3635
e46fbc2ed93cb03f44f3ccd3a5c17f53af316155
157947 F20110109_AADCZM lebrun_d_Page_030.jpg
fa653bdb67930cdd9deb49d82a800ced
3494081a21537149444f81294a218ec0417fdb36
2203 F20110109_AADCYY lebrun_d_Page_100.txt
0b3c833a865edd598315577292c6cf0f
fabe42567e955641cfef547031ecee5624d718ef
F20110109_AADDFH lebrun_d_Page_091.tif
bf813c4698331db1118d79b04903f64d
6acf81b5aad3a566da17173184e90687abe9b162
54450 F20110109_AADDET lebrun_d_Page_065.jp2
6fed3d6402e1bb6a53a6559fc8bd1106
aa07dc36a7c0224306c672cff3c63f7230ecd327
993761 F20110109_AADCZN lebrun_d_Page_078.jp2
d671b3159748482351ae291e9c2721c9
a86809fab8ac437bacb056691a89098ecef82821
48058 F20110109_AADCYZ lebrun_d_Page_009thm.jpg
1c456e4789390e832e5a8bfb7bd45d1e
d7b42881eae6eea5a5c7fdef7c4083e5095c0dea
F20110109_AADDFI lebrun_d_Page_097.tif
5142b38400c7200ab260bcbfe8ad065e
d4a127f73dff131a8ea633afcd95aad22556243f
62459 F20110109_AADDEU lebrun_d_Page_049.QC.jpg
fd37958678300dacbf742399c53ebe1f
8ce74bf5d7eca0ef70d74ee21a348c2a1ada8125
101580 F20110109_AADCZO lebrun_d_Page_023.jp2
b6fe2ecc72b190150ab10dcb4db9d146
b6651c85c1f451c166dd8bf12b5de9bc967310d5
F20110109_AADDFJ lebrun_d_Page_044.tif
e1cb49074ed31f1fc04bada5288dfdd8
1435ad9dcd16941ec888a2525a07eedc10adb4bd
752 F20110109_AADDEV lebrun_d_Page_031.txt
183439de86eab7b30ae84e70a8478646
38f10183cb276b49aacf17527ecc7aae14fc35b0
83382 F20110109_AADCZP lebrun_d_Page_011.jp2
9a75464b497d21602106569df0abfe00
45912e4e816fd56222bd1aeb972d3c884ec2ad9a
158975 F20110109_AADDFK lebrun_d_Page_042.jpg
a3439978cf36edd3a0c26023cfd0048a
527db379f3c1685810e57821c3912d89cf57ce9b
206565 F20110109_AADDEW lebrun_d_Page_013.jpg
2161152f6b00dc7f6e36d37daae5fd85
10c6e8a97b032c39971c233b156162cd036ff9a4
236708 F20110109_AADCZQ lebrun_d_Page_005.jpg
281db8c49a9fffd29fdf35e972d01c42
b7c82f464c96377da63fc9fef7df2f9dd04a0931
78249 F20110109_AADDFL lebrun_d_Page_066.jp2
5b30df8c6ea2984e443a29f9ec209904
71ef4bf60f5af8227f84f0fadec377fcb42734cc
1605 F20110109_AADDEX lebrun_d_Page_071.txt
dede94e1174c4fb838a2d1003272daec
b65da9a94a1d12752a3c8049565a42368d401925
12641 F20110109_AADCZR lebrun_d_Page_070.jp2
d6b4e11a58a53cd8c58736caec952405
df838c718578d2fd23b06e7cea3d6599868cc035
269087 F20110109_AADDGA lebrun_d_Page_009.jpg
0020907c1e396a73ef84fc78661a5ecc
d35010a6af9ce75fd0075e98eecae9721107e295
81054 F20110109_AADDFM lebrun_d_Page_077.QC.jpg
58e502fba3de5689ff0d0416b3ca17d8
cdf2ce41365267285c98029ef4a1c3fddebb9769
F20110109_AADDEY lebrun_d_Page_053.tif
177eed0b050a90b682fc2380906688f0
b7e4be1c762fd6776fc1e7e66c4fc0c3c4ceafbd
F20110109_AADCZS lebrun_d_Page_029.tif
f16ff373c227658225426684f4d40314
c073297094c8a7844bfcd6f89f0834825cb9cb3b
757194 F20110109_AADDGB lebrun_d_Page_047.jp2
6adeaf8089e3657000fa91aaf07b6b28
c6da8f1aee4071386ede6c371d6e26cb3fccfb7a
20232 F20110109_AADDFN lebrun_d_Page_041.pro
4c7313d202beb0ac169afd7f15621b10
53a146cc37884ba6b5ca3ae06047889d82bacd61
22798 F20110109_AADDEZ lebrun_d_Page_063thm.jpg
8a4ab7015fea9954bbbf71f5b40a362b
2a74ee7f490b7dd112578d84bf8a56602e5539a3
148405 F20110109_AADCZT lebrun_d_Page_097.jpg
09d2622cecf110403482d7eb7571b2cf
2f6a3832ef4a1f97c8beb6df482061dc901a60ca
51081 F20110109_AADDGC lebrun_d_Page_058.pro
b9bdec1b86a2e1741d53cfeb4f3f193d
6aa21553a93d942f4199c170abe30b2eff434cbe
F20110109_AADDFO lebrun_d_Page_096.tif
8df22b29c6bac773808ae2ac2b3b3087
b02d057e7878c34af23fa701425bf2aca9705243
F20110109_AADCZU lebrun_d_Page_055.tif
864747d6375f3faa158d1b66805b037d
fb7df804f71696c889e1f93c89ae527f34599da2
21788 F20110109_AADDGD lebrun_d_Page_036thm.jpg
21c2c25e0cd5f35a396bd9e1b97e537e
5b287fd8cbabb073934d749f2e2e6d2615c2ce90
165300 F20110109_AADDFP lebrun_d_Page_086.jpg
3b60981e7446c0b25159f97019654212
66492f90aa5800e91a320c63be8f851240b12707
181583 F20110109_AADCZV lebrun_d_Page_059.jpg
c7c99ec6cc4b94df4575ecb4df42f1aa
19751f55ed72416606117240f49bab81f4d3443c
47579 F20110109_AADDGE lebrun_d_Page_073.QC.jpg
8ec0838cfb83d2302a5ecc20f7763e6a
fdd41baf944d99a304795b42801c1d977e66dc3f
1642 F20110109_AADDFQ lebrun_d_Page_011.txt
63ca9109db8577489b987ba1ee4aca53
95cc1df57d29c90a025e201de63f1297228739ab
2458 F20110109_AADCZW lebrun_d_Page_009.txt
3ee6175e28318a1cfad7cf8a516b3866
6808a6f0d0eb1019378eb85ab7b57bb6e23f12c4
69769 F20110109_AADDGF lebrun_d_Page_062.QC.jpg
ac4a53fb0cfe83df162cf5e1b0b7168d
e4ac16bdc2cff6302d8ca2ebe8a1b77284e386f6
179021 F20110109_AADDFR lebrun_d_Page_035.jpg
1566767f4604c8c5e7b784c1916d5f25
b5ed1df207720d7bcdd936cd6a541805c64bbe6a
1558 F20110109_AADCZX lebrun_d_Page_088.txt
553bb4b38a55559897be64e1a02ceb13
b9c8637da27c2dfa0f8fa4150a2cb2cd26e85a20
115010 F20110109_AADDGG lebrun_d_Page_096.jpg
4ef7f781d261d2d0f9a042b32be368d1
2da54ce5cf89a3e29c648522aaf2fb222a12e4ff
454 F20110109_AADDFS lebrun_d_Page_001.txt
01dca691a3bb4f945cf835f3499893e0
2ef957f67d9e4fde16bae12c28e910770e307f33
67549 F20110109_AADCZY lebrun_d_Page_097.QC.jpg
0efcf8e778150c0b41e60602794acf92
56889946005e27697313c28d1400f11c79369701
20056 F20110109_AADDGH lebrun_d_Page_090.pro
b902792bee3e3a7b2916d2141234b1dc
359648aeca6e7ea7823ccc9422f7e18fd3d7b791
1051950 F20110109_AADDFT lebrun_d_Page_095.jp2
9e7e372621c4982968bdd4bbab2fca14
fc6d71bf5d0aaab1231f9a17773cffb9f3a9f5f0
1891 F20110109_AADCZZ lebrun_d_Page_028.txt
ec8d45a29b9d58b81561a56a187ffe63
b0e1fb399bf65b828ccf647d971cd588158cbfb3
60846 F20110109_AADDGI lebrun_d_Page_089.QC.jpg
ae61857354ac52067debe4a9aeba1223
9673a6c52e3c0007a49e0c1084e8f58dc17dc83e
197178 F20110109_AADDGJ lebrun_d_Page_076.jpg
a9045606a03893a4e5579b7de4660de9
a581b8f6497ac42ce42c4183f67edf1702c19da4
F20110109_AADDFU lebrun_d_Page_014.tif
66aff1daf27c8d2419cc8cfc5854204b
de2a772881ae82f9eda2fe200a974790116c0a3b
98869 F20110109_AADDGK lebrun_d_Page_064.jp2
7b436d30fea0f0eae27eedebd112c5f7
a48b5181e8afb3af9f5d703294f174318f4ba12b
959 F20110109_AADDFV lebrun_d_Page_049.txt
8a536e6ec6c7a4013dc4d4f3279dc3da
5246e6f325152ce4f7fbd075734a80f8ad64274a
23134 F20110109_AADDGL lebrun_d_Page_001.jp2
4e7cf20bf9b3eaa237a59917d3af47d5
bd40037d2a2e2f9384ebfa24307863dcf15d4ab3
19922 F20110109_AADDFW lebrun_d_Page_015thm.jpg
6718b7386647d2b8ba943942865a4e73
dd2ec70c2c700e565ff72fecb55a19da80d4f6d7
75784 F20110109_AADDHA lebrun_d_Page_054.QC.jpg
852453352d39cbcb8a6385fbfc0c98dc
dea25dbd4ddc6dc7cd3798837cbd84fd93c0bda4
201606 F20110109_AADDGM lebrun_d_Page_019.jpg
2d58c66947a35dd7d25501e66be68711
4e9b62b9f4a771490c27ebb31372a8e1f52e4184
80234 F20110109_AADDFX lebrun_d_Page_045.QC.jpg
ed8aaebce0529caad8effc352c7c885b
aa00ac3595d35cb1661934741aee6d068f98131a
1731 F20110109_AADDHB lebrun_d_Page_010.txt
fbd616cf8b31425f6bc1f1b2b830cc60
2d3a5f989d0529378afb73dac127f1f3fdb61cb0
21636 F20110109_AADDGN lebrun_d_Page_016thm.jpg
f9f58ba22a2367fff60e8e349ac9c0bf
f74f60f00a3f41d25e25dc6a808345866aa16d59
24421 F20110109_AADDFY lebrun_d_Page_019thm.jpg
d9b09a725bb9be16b3aa0b5bc4bca37d
1d8cfc04dd97e3343b80922743ba3952bc4d1195
F20110109_AADDHC lebrun_d_Page_100.tif
c6d95dad52ea62c82e00d70a8214f755
544f6b9c41182519011b11e48db0175da8d52337
F20110109_AADDGO lebrun_d_Page_072.tif
8a420e21745f3f50e9018363069cf9f4
3c3cc06629bc7d0823101400f7363cf05d715c05
56213 F20110109_AADDFZ lebrun_d_Page_012.jpg
379a9cc0d10f9fcc5f43a5f94b68fcbd
9781c5629cc41a33ae33c3260bb24cafc510a8f7
1624 F20110109_AADDHD lebrun_d_Page_037.txt
e60d7f93e1c6aca74dfcda3fc95612c5
43cafe321a5f7651b9c155d706e18d116a4b2211
232 F20110109_AADDGP lebrun_d_Page_103.txt
39bae02e884299d6a0e619970176c0ce
3fd5514dd248f798764c19c3ccd94d5a10452a79
43272 F20110109_AADDHE lebrun_d_Page_090thm.jpg
b199d2dd27215333ee9668bc149187a5
30ba7f3440da33b4329d81686d8f3d3456e52c6f
F20110109_AADDGQ lebrun_d_Page_071.tif
70e3a79b8bd009a678d6e5fe6d34984f
6e55967444c5648717a7f6c4187691bba3aefaea
49206 F20110109_AADDHF lebrun_d_Page_053.QC.jpg
a8984d434b717f1c4509d25df13c5693
757a3571e6120a782fa2331a46cd13a18f074029
95104 F20110109_AADDGR lebrun_d_Page_010.QC.jpg
1e8f09c27bcd59231ca1152d82c8c736
ac0e947a6d2f2b257d9d137e6f2a21ba023e50b8
41681 F20110109_AADDHG lebrun_d_Page_041thm.jpg
74e9c59b3da3f5d824bdd05bcf807b64
7c05cb6ce599b749c57d6e21478297890d7e0507
F20110109_AADDGS lebrun_d_Page_024.tif
86888bcb75d872908ea936fb102db597
18984891fb47144f916670c4ba4bf0e975083c49
167992 F20110109_AADDHH lebrun_d_Page_090.jpg
e898765949f1a54c48dc2909448da780
eda3e87bda4da8c318263eef0518d63f2a9bfa0e
247979 F20110109_AADDGT lebrun_d_Page_102.jpg
ba639f52a4a09cf254c4c6a2190b227c
8e1b3a177462c6042711394f80f9f3daa04869fd
2467 F20110109_AADDHI lebrun_d_Page_102.txt
d6377b6b81216f3ed889e44ce56e9516
a38080c62c630864fa056a2fd2520de21a77e618
33078 F20110109_AADDGU lebrun_d_Page_057.pro
470010d58545296d695e28e21a8befb1
2bc27340aa3e367d246ed46f99f5d3ec97d1a0e4
27105 F20110109_AADDHJ lebrun_d_Page_021.jpg
93b5f1edcf200a05ad07bdaf1baeeefd
34a253c1ec763735e3e389e8f5edb6d0c7f6478c
11778 F20110109_AADDHK lebrun_d_Page_070.QC.jpg
3dedd6d4274b5983c5d7b91d83195a37
670b63693507e8fe84ca9b5b2e530f0770d3faf5
24791 F20110109_AADDGV lebrun_d_Page_061.pro
39557c5d2eb975e7e43d7abfd2812fb4
7a7738992a7399dfdf5684b3432cfb20c75f077e
F20110109_AADDHL lebrun_d_Page_084.tif
4ccb94efa533e56259e3200c05bbb6d7
fe6c023b43cd875091aed71f8eddf2e93c0e298b
655799 F20110109_AADDGW lebrun_d_Page_081.jp2
94317957616e0a13abcd6a3d0aa74a9e
ab4e1460e308a9bdb3e27028445668ce2901c297
147411 F20110109_AADDHM lebrun_d_Page_099.jpg
37a3803113a402a1d70eb550da6133cd
25f26ac11f80e9adc81597363a642310dbb0f9de
4740 F20110109_AADDGX lebrun_d_Page_021thm.jpg
30797bda45fcfddd157b7d7695302797
390cb6fdfddb36f979f3f341dda58ecfda90f738
30799 F20110109_AADDIA lebrun_d_Page_056.pro
6a04f33aa23203fde1480eb97f0eaaf7
d6ca85266c3e9d9aaa43def0bea93240da744613
75537 F20110109_AADDHN lebrun_d_Page_075.QC.jpg
c7f5ed70240b9ff07fbc10ce1cc1f3d4
309a2a2d46b3cf80597ddcd31907d864011d0547
185939 F20110109_AADDGY lebrun_d_Page_026.jpg
5527a5d29910c2657f7a8f60291155ce
ba1fd0d371e46916a00a4d698f14f9629f598017
799403 F20110109_AADDIB lebrun_d_Page_035.jp2
434442992f6502b00a6ae55a38e8b8e9
8e8e31d00c5ac04f50fc8192ed0d0aa54099abf2
16565 F20110109_AADDHO lebrun_d_Page_073thm.jpg
e661d25e3c8f24ea6aee12154418f998
2090b333a1f6c18b169917eabaaf50b3a52a8b09
F20110109_AADDGZ lebrun_d_Page_081.tif
a257ed85de5285ad482548705ecfa079
1a2824488f3249dccd3f038cec95fa38efb1de61
112160 F20110109_AADDIC lebrun_d_Page_025.jp2
62a90e7e66ba3c359cbd8d24872c1f2d
c221a486a29f2fe2250719d683d4cf1f0a4ba1be
93206 F20110109_AADDHP lebrun_d_Page_006.pro
d5e667b84093bc82dc9962252b575800
4a4c24025d75259af9e1bab7f865b8d9bb4755fa
17735 F20110109_AADDID lebrun_d_Page_003.jpg
b5a344234b42636d2295b787e64e7d8b
382727f7e526a51da3364924ebbf27aed6ae38dd
42210 F20110109_AADDHQ lebrun_d_Page_051thm.jpg
abe2de561ebf7827efdc255e29ab0d59
d4b5c30c312d8fbfefa799eeaeccc711e7651a2e
F20110109_AADDIE lebrun_d_Page_079.tif
f6943ccca1b6ad53c6c8ed536b615726
caa9b54128676c60e09ad3a3d0c8b78bdd68458d
F20110109_AADDHR lebrun_d_Page_049.tif
c964dedcd7c12c7bfa347f95315b5fde
0088cc7bb7d9906b9f9c05f7ba46497eac2b10db
F20110109_AADDIF lebrun_d_Page_005.tif
3f59e03119a93986560724c842a8f1f9
5c85bc926ab5d8ab19894cb40f6c4207b5ca7740
14601 F20110109_AADDIG lebrun_d_Page_004thm.jpg
ddb34b345c348dddf0d8bd3ccbcb9c3e
a2a8aa1cd5fca1ed8a02b432d0cc759a25d84097
F20110109_AADDHS lebrun_d_Page_034.tif
acffaf8f2d828b972ac6e08e456c3d02
c2cbdec1e2030e1f01174793edf697842dd8225f
122703 F20110109_AADDIH lebrun_d_Page_101.jp2
91452da8ec7a74fae84fd21e06b47967
5023e8c82cd72b46a7c4c5e5b8ff150921f9c858
F20110109_AADDHT lebrun_d_Page_036.tif
d13a884af73d240248260055524b67fc
afa0210072b8b4363ce6e5826b8275b41bdf4bf2
F20110109_AADDII lebrun_d_Page_025.tif
73ce056e18899aaf4e58b7e47030aa1b
377f3f15653c99218b622383e662fd3819659f29
F20110109_AADDHU lebrun_d_Page_033.tif
43ab6094382018b5e8c0f99778b91107
b8fe2539238fe2e00d82e1d848a12fa7407bea0a
18488 F20110109_AADDIJ lebrun_d_Page_066thm.jpg
a06d3efcfacef953d43359d7bd3ae04f
bc07621768923058b601a3ff3c7d3c9dbc4b4169
F20110109_AADDHV lebrun_d_Page_007.tif
5d1f057083195105f9c4818d97fc4b47
8507795fad5d2c9c3ca0080870fd63ef05d19896
F20110109_AADDIK lebrun_d_Page_026.tif
9cacc48acfcc5e31a422407755567e92
3f56dd85256d04e9ad26bbc0886b5aa4502d4062
1516 F20110109_AADDIL lebrun_d_Page_015.txt
9943ae6247d03f6e2ce97cd7edd48aa8
3e9b425a45a40a96f4e56f9e805bf1270e6831b8
21097 F20110109_AADDHW lebrun_d_Page_027.jp2
3745980d91c15e65bdb0a89a1843c442
7e05cedc4a40f2f96c090c69a8ee53e383529f90
212564 F20110109_AADDJA lebrun_d_Page_018.jpg
011150998cda99f83cf7e7f71e280a44
fea4ad1dfbd03222163a19b8a9a7e833d20dc523
194616 F20110109_AADDIM lebrun_d_Page_078.jpg
191a0acdcd195701c46a35efbed8165e
151fcff8dceb2556d906686478397053007a7ffe
75553 F20110109_AADDHX lebrun_d_Page_019.QC.jpg
9771add0261e54fc3c5c631ce1b5e0d8
46b7769657466e7c62e5bac0cf4c48f35b7b929a
41501 F20110109_AADDJB lebrun_d_Page_007.QC.jpg
19e15143dc17cc738a6b67ed3fd0acf0
70c9d7ca06f4936c0f3bcc01df19e28e46e60324
242064 F20110109_AADDIN lebrun_d_Page_101.jpg
40df578460846dcf58f085f1cd7771e4
dbfa8df4a2a2f630df12353cc545ecd75ec79d7f
81089 F20110109_AADDHY lebrun_d_Page_039.QC.jpg
d74670514a58cccbbacf714c4c0fe753
0e13d11a06ef6679505c4ac9f588b1b26eed7f87
107994 F20110109_AADDJC lebrun_d_Page_009.QC.jpg
05abac0b3a67fec257299846ea5cab83
3306d5c0b54a85809f3633285107e0c4911547bc
165939 F20110109_AADDIO lebrun_d_Page_011.jpg
9015543d9d0fb34a84354c21be68ef92
420a865610c5ab87f9f7d8ed2c03fa3d788d743a
32200 F20110109_AADDHZ lebrun_d_Page_037.pro
39785d84fd056713a1cbc44f5dd185f5
077978a40d41c26ca56460e7472631f73396ec7c
655910 F20110109_AADDJD lebrun_d_Page_052.jp2
1d8773b3c09e951bc57dd1b540021b0d
e783aa7d304ef089b52e41d79e82a9024a0ed4fb
F20110109_AADDIP lebrun_d_Page_076.tif
192a5013dfb7f26b90232ca77514586d
1211ea77da42442800491ae0cfc127ab90a9c782
146258 F20110109_AADDJE lebrun_d_Page_098.jpg
0b24e72b94d8014f0c2bca603daab644
2e8ce117271735bfc89c6afcd23c8f8e2615301f
F20110109_AADDIQ lebrun_d_Page_073.tif
f0336285f0948d271e5b1bdd0deae7e4
f907f5f65a5be8caf721d8d2579317ff1364206c
122330 F20110109_AADDJF lebrun_d_Page_082.jpg
1cc0e37a596b423dd12952984f289616
89504b590a9655b408f47648874570f01aa12ac7
96901 F20110109_AADDIR lebrun_d_Page_084.jp2
89c997b5f6c2effa77f3b60eeb0ef4d0
92a4ba69287bf5bf1161f9c78659f06018c280e3
38042 F20110109_AADDJG lebrun_d_Page_026.pro
b0287a24359a1bf3667270895a73d17e
6f8d3e3fa613cebe8420cd9ebdca2bfb9779756c
4706 F20110109_AADDIS lebrun_d_Page_021.pro
614f60bb3b59684a0a55a96bed3dfcbc
8289742abd6c985c17dd59db7c2f432606999379
F20110109_AADDJH lebrun_d_Page_086.tif
f64ced42df80031e88df431d622a03a6
4f7d701c244e6df2179bd898e4b30b4a9ef51d52
50845 F20110109_AADDIT lebrun_d_Page_014.pro
5c0029b7935282de1cc12a286b26c935
e887b6cc5feb51d578b82b63a7fb63d9c75eea6c
527 F20110109_AADDJI lebrun_d_Page_089.txt
e9eddf60306a65bc6c8e6226ed806e4d
84fa353b4718647bd1537e86b3897f3717e57e81
30900 F20110109_AADDIU lebrun_d_Page_007thm.jpg
9f706cf0957fb6d44a98279a98efa128
6e34d151698bd33eb27b6d42302cbe1ff9d9d21b
83770 F20110109_AADDJJ lebrun_d_Page_042.jp2
984fe6a486debbf1a4b0a145abc2debc
b01567f9fb57e2047a915f12c583aacfbee7b879
542760 F20110109_AADDIV lebrun_d_Page_016.jp2
b46607adc25e5edb7cae679a5799d2d8
8c4e508629fc92d67ea9818b919228248d4f9bec
18941 F20110109_AADDJK lebrun_d_Page_060.pro
31b821453759d16fe4a07823b3d5c913
2a4a99853d95b04881fc578e3e4370c87bceaf3f
1679 F20110109_AADDIW lebrun_d_Page_039.txt
6565137778b950b28944b03814661ac8
8a48997657085433a567246a8a4835097daf1ce3
1645 F20110109_AADDJL lebrun_d_Page_076.txt
54d510d91ec899af72ad91246a61ec8f
da63c73120e4686bb3b1dcf202e857e688a4afd4
12968 F20110109_AADDJM lebrun_d_Page_021.jp2
ff0c9be161c584c23c1a4d389edd61fd
7e340588f1c557c40a45d2e13e095a0e32ac2998
100279 F20110109_AADDIX lebrun_d_Page_044.jp2
e5c5809f8a26c12addaf83070c965d88
cb3bf014e395011222030bb30e2a88b1b07e9d0f
188889 F20110109_AADDKA lebrun_d_Page_024.jpg
daf42a106056a2d4d6548a1bfb3be8c8
fc8542d06b6706d95e1415ab6dc447671668c23c
156735 F20110109_AADDJN UFE0001215_00001.xml
d74fd9fa684598a5b1c652f8ef2a03f5
2af139d728de3296fd03e35cc21cbf685f9a632c
83923 F20110109_AADDIY lebrun_d_Page_040.jp2
f4a54e316b5b4db623501f01bc1d10df
90c100fd4aa501483cac0c4ce8d4b1a12253d18e
215903 F20110109_AADDKB lebrun_d_Page_025.jpg
357962fe5247c73210bdda09eaac5105
5f1d2e79c6274457da5f6fa0e125cdd48948cd83
46066 F20110109_AADDIZ lebrun_d_Page_023.pro
0ac3d75221cbe7ddae241c079c12832a
689e453dc2c6422e30736c05cbe627205b1f4491
40197 F20110109_AADDKC lebrun_d_Page_027.jpg
d248278b08074f71dd48f73393c28d79
b400fe9d26042924fff36e4a67dc420dcb4c6b7a
219095 F20110109_AADDKD lebrun_d_Page_029.jpg
ebc979f5786ca3716a9267ea0c569aff
dfce155d9292246a7db4708e92b2090c2fcd4fad
14624 F20110109_AADDJQ lebrun_d_Page_002.jpg
ca649d12ea5080b2fd351ef1b970e2a8
ca828a398b9e436f2aa34530b9e431f21549e14d
162487 F20110109_AADDKE lebrun_d_Page_031.jpg
9b7fce6c7eca95eeae139193aac616d7
282a811f2caddccee19bd526af74fd6688ca358c
293138 F20110109_AADDJR lebrun_d_Page_006.jpg
cb7d9da640f5b56add4f2c5281cbb70f
83babfbdbc59e8d926dbc8b6f74106c654f6fca2
260685 F20110109_AADDKF lebrun_d_Page_034.jpg
c74adbe7a8b4cca67a6b3be258c35c5b
472337e5b5a48b91052f9c67080e15da260d7e86
86058 F20110109_AADDJS lebrun_d_Page_007.jpg
41a6e8c6460bf209cd18b40125498ae6
10fdb97d9273192e6a908dc04b82d936c6ca5f9c
202642 F20110109_AADDKG lebrun_d_Page_038.jpg
f4e547a8a0d4ac8fc3b351bf0f6f4a67
1c65540bb6f340cf7b37e73a2b9417617186dbe5
87549 F20110109_AADDJT lebrun_d_Page_008.jpg
ee4d4576b505e12050dc8fa9af320a60
d299eb6668f5cdcce901d727aa0af5092f52faef
177186 F20110109_AADDKH lebrun_d_Page_039.jpg
9b54473fee2475e1ec25563126e5be07
d240643684cb956b32415b631a1238af7715a3cd
232336 F20110109_AADDJU lebrun_d_Page_010.jpg
a99451a3ee7441c170a0197443351f21
ecdf9c264a2273c1f871a7e676c56912ddc53bfa
162325 F20110109_AADDKI lebrun_d_Page_040.jpg
03b263fa44548685c763d0da662409a1
faa1e86bf44b9d05bc673a15d125c08a969011bc
163749 F20110109_AADDJV lebrun_d_Page_015.jpg
4a17b4f4e9394f554596cdd9f3ebcbef
e03152c3c6bb9aed9c77dc1af9bcbe205094bbe8
130766 F20110109_AADDKJ lebrun_d_Page_049.jpg
841bc9e08b427fd73cf1a056539b26a4
f1290587fe2f1fd097915e30cb7deecd660493e9
210824 F20110109_AADDJW lebrun_d_Page_017.jpg
499476f4643bb26a8a1fbc7f6dc6ddde
75170fabfb187a335215811bf02898df961528d8
187936 F20110109_AADDKK lebrun_d_Page_050.jpg
7e0f3d48ce41858763595c57a3c0a678
d02ebf1f3fba9bed4b9016d19022d3b3f1c3a2e7
208741 F20110109_AADDJX lebrun_d_Page_020.jpg
02f52b74edae50db8babb3dbd6379838
75fb3233e23a41b5dd11d65d65556a7534816a3a
159583 F20110109_AADDKL lebrun_d_Page_054.jpg
9b6ad9d75a5fa5dd2c63f4a29f04acec
b9df75a7d345508bdb73057253693b7d71a5f2b3
119840 F20110109_AADDLA lebrun_d_Page_089.jpg
7f6ce6aa739d9b2c1ef5f360b76a885f
1258272e418515e736ff5af2dab6cef1a7e3752e
147342 F20110109_AADDKM lebrun_d_Page_057.jpg
aea633c64d2af56fa5a40087c0bd2bc6
123c81e7d3bce3d1488928682ebce5b780a62393
183641 F20110109_AADDJY lebrun_d_Page_022.jpg
3e0690c5f83a16099750fd3f7b76d48c
a909dd6bfa10cf5137c4012ef5d5a2b9ab407895
156233 F20110109_AADDLB lebrun_d_Page_091.jpg
8118312f0fc4eca2e6da87d4888bd0e8
9fdf9dbc3d4c060edf18f4bdc1082627f45ce72e
140413 F20110109_AADDKN lebrun_d_Page_061.jpg
43a6a2912aac5247333dc64e1f9af43d
f6d02df9454d0ea129c1703c7268525a84bb56d0
197665 F20110109_AADDJZ lebrun_d_Page_023.jpg
659ce23b95d0e94d387c22cf4207dfee
61e09a08df0b283d56827d96a7ae02dfc6f04500
154519 F20110109_AADDLC lebrun_d_Page_092.jpg
23fdaf0f8b2229ca931624ce45e29e34
00ccb4250576931bef3d6b85e29d0f329a4960ad
187359 F20110109_AADDKO lebrun_d_Page_062.jpg
3f19bbf7b3b154445968b8fa561fa937
bc7b183eebeb59671912b9de4eb33dde288f29ec
161572 F20110109_AADDLD lebrun_d_Page_093.jpg
8f83c0c14877ade199efcfaac40dbc8c
3e29448bef04f967bbf84b65b14ff4ef3539c99c
108237 F20110109_AADDKP lebrun_d_Page_065.jpg
b0a77272d0e48e12cd32d3017cfbbbc0
b3efe18d7cc031a09340807ce6e5b15034920a66
138168 F20110109_AADDLE lebrun_d_Page_094.jpg
acc0b5d82539b0baba2295097f34f87e
9550bd4f41bf150c161ac57f13409b87f163e9be
110185 F20110109_AADDKQ lebrun_d_Page_068.jpg
0e26a4ae91d386ecc256ceea552d7098
43f598048733209955f15df3a2ab61893ef0e785
5477 F20110109_AADDLF lebrun_d_Page_002.jp2
4f35d02421e2737ba4e8409d6e88f581
09755a98f36579e57a4750e381107ae74c1cd9e6
210696 F20110109_AADDKR lebrun_d_Page_069.jpg
232172ca56ae67e2f8896261ae881a68
7b9d0dc9dd9ec06a57b794f29639a80c147ff024
62435 F20110109_AADDLG lebrun_d_Page_004.jp2
d2de5b838ae1f5eb6c42b782c640d0e4
53f76c22e927a2c145b53691123221d2805f78f0
166574 F20110109_AADDKS lebrun_d_Page_072.jpg
6c52dd4b9fe35eabf2d965cd0c187196
b3a2078693891285dc9a4c0e3a77891a71394f85
1051981 F20110109_AADDLH lebrun_d_Page_009.jp2
fb2c556dabf996189b105730cbd145f4
2443a0890fe957de347c45e6e50b6cd39240fc4d
126202 F20110109_AADDKT lebrun_d_Page_073.jpg
ae572f320c747f09dbaa1e08ed46b344
44bbfdeae1a37055679d5b95f1d8e13a389d1db7
F20110109_AADDLI lebrun_d_Page_010.jp2
453938172663372819fd201d33a0a289
abf98717a2a3d76c185c69a4c0ce685f403e8c48
185431 F20110109_AADDKU lebrun_d_Page_074.jpg
0c2f9df87be72301c0a09f15e0978449
ab48160298e20069e5cc87d192d87f82496085ad
29508 F20110109_AADDLJ lebrun_d_Page_012.jp2
853ebb1a0a3641387115d6e41344f673
acde4283827a2fb841aa0f04c440bbf3f04d8a90
160104 F20110109_AADDKV lebrun_d_Page_075.jpg
2982cc0f196a1d3b52679ea5b9dc33ff
21090a44305ec576155e59e743095ecae64474f6
957551 F20110109_AADDLK lebrun_d_Page_013.jp2
a46f7f311b285bc95e3c01a6ecad64dd
ee5c4ba3cc82b3fc217a0a389b4024d4be5acec8
180511 F20110109_AADDKW lebrun_d_Page_077.jpg
7be9a9b1d2d5e4e6f169921b05d6ed08
f482ced46ce9c34b041944154f0591cd031dd391
110415 F20110109_AADDLL lebrun_d_Page_018.jp2
945382a38b30061b2255c549f20c5582
c857f4741b5eb86eebfdff4667eda86981991bef
192920 F20110109_AADDKX lebrun_d_Page_080.jpg
503649c41d5bbe046986f8a39389a833
941d9a87a62fe713c1f36f6deb34d5987950d941
69390 F20110109_AADDMA lebrun_d_Page_056.jp2
e0acc9daef9992f7af2a6bbce27005e9
fe65410130d98f1f3472926d867d762fafdcb171
105032 F20110109_AADDLM lebrun_d_Page_019.jp2
f7ce53dafe5fac34d098087065422071
35ce20cde095cc75628403aa07d58bd249abff81
139683 F20110109_AADDKY lebrun_d_Page_081.jpg
aa1c9f5fce1033685930091c1c1d5c08
51eb03c5ee390d751739938fc6c33e250f100b7f
845248 F20110109_AADDMB lebrun_d_Page_059.jp2
e8ab7d32c250b640af42eab752326212
f11bb3083e8155c9457ed340cee673593b8cd47b
107761 F20110109_AADDLN lebrun_d_Page_020.jp2
f5350b1d3eb3f1c20e280b3069403f12
a54fb18ced942fe3df5a343da933d1937eb99e79
96289 F20110109_AADDMC lebrun_d_Page_062.jp2
b50bedf783ad0d02b369bfe9d4eb4842
8fb90bfdf32004476080781428d1831700615c63
97084 F20110109_AADDLO lebrun_d_Page_024.jp2
49d38579849e025164e049546e9fed31
510fd0368a6999d027038db22e2f702f79d2987d
195037 F20110109_AADDKZ lebrun_d_Page_087.jpg
ac37b830552a33b6664064616f02d51c
0d2a5abbbdf67d7750febc286c7b5c3b29ec5ff3
97456 F20110109_AADDMD lebrun_d_Page_067.jp2
f7410fbcb77d59db76705a98f66b04d7
04b413097ee8e9f686e4c89cba1f1956bb5a96f6
1024207 F20110109_AADDLP lebrun_d_Page_028.jp2
36d06f823a5e52f622004e35a91f9ec6
54e040db4fe16ff61bfa69d6aa0cfcd46fecbecd
56247 F20110109_AADDME lebrun_d_Page_068.jp2
bcd9f3d2000deb568c32c751bb52803e
1257811f43b1e4084cfd77e5e904d3eced9c07be
112639 F20110109_AADDLQ lebrun_d_Page_029.jp2
987cf4f6189f48069946e631442547e9
b4d7871c85c293149dcb5c35e4b086fb3f9c2bb7
106921 F20110109_AADDMF lebrun_d_Page_069.jp2
1d953bf129ecf9bcd1644b8c0a2a305b
f82fd97200de55a6629368e46ea8181c66c7eb44
748361 F20110109_AADDLR lebrun_d_Page_031.jp2
528629403511c559e184cbaa0374f0c4
49e725585ce1f1baddf96555b07fb3e61a509eaf
77521 F20110109_AADDMG lebrun_d_Page_071.jp2
5858005e03b33bfa851ae27f29c395f5
c293ac264fa5b2d52c71d0fadd5ce3af87fc2c8b
38798 F20110109_AADDLS lebrun_d_Page_032.jp2
85456a36f2bded5273113bc0cc506082
606ae9c25b223b2603e3c6c117ea278958183651
724077 F20110109_AADDMH lebrun_d_Page_072.jp2
70d60e11df6bab076bea887598ac0585
1dbb87212b58bb55cbc3ab644e6a330f87f54b6c
92225 F20110109_AADDLT lebrun_d_Page_036.jp2
8e91f56072923909b46c99f7bcd65b58
b1a63294fd0c93327655a1981a581f9c1b1a4c31
63511 F20110109_AADDMI lebrun_d_Page_073.jp2
48b89fe6931197e01472a125567a9d1e
03e331d962967016256768c28c4081f3247b419f
875927 F20110109_AADDLU lebrun_d_Page_037.jp2
534676dbe04f2100ce814276ee0d756f
966b63d7508277d9e121c8a2b9fca770e188b01d
1028687 F20110109_AADDMJ lebrun_d_Page_074.jp2
577ce9054263dc342dda51bdb5bebe85
e072b5ea5b8cfd6b6c108256ef91a578d82bcd6e
764031 F20110109_AADDLV lebrun_d_Page_045.jp2
0cd969618620761355157c8b844b5983
4c61f46010c02090cb072a9482a726dc65b3dce6
747689 F20110109_AADDMK lebrun_d_Page_075.jp2
5a062d9cc0dae5553e92d2c0a0de5bb0
ca1887717f0abb2771241ca3a3ffde06f99c0eb8
93162 F20110109_AADDLW lebrun_d_Page_048.jp2
286aac3013b8ccfe1c3f97fe34095c26
272b200231b7c108b89d6d12e96e8e2806139b30
984912 F20110109_AADDML lebrun_d_Page_076.jp2
d9f1a491565ddfae84eb1963a12da6de
8e036725592e3e77cb8a1dd2969e1f061e112b77
592623 F20110109_AADDLX lebrun_d_Page_049.jp2
da2162dd128b4299bfb35978c2c1a8f3
cf9940574b9b19aa92ba46f08ceb66122e58a364
107540 F20110109_AADDMM lebrun_d_Page_079.jp2
d5a110a12b6761d5f564a70eea7cd176
96de06958371efef1d8bcdeeda2c9429c577a414
351130 F20110109_AADDLY lebrun_d_Page_053.jp2
a0ab03f545387ec8cbef7e2256b6b9cf
2ccfe49bd5a686c65ad18c5216ee4f2ddfe56d1f
F20110109_AADDNA lebrun_d_Page_002.tif
80939a9318b2cf1af551e35b5bf5fb19
558f0b12c0dccf99f521d9f56f586ac3b8c6c534
89362 F20110109_AADDMN lebrun_d_Page_083.jp2
8844ddeb3b32ea5b778a8b9ef0d34423
fdaf41a322afaae57c0c0934fc27c73b84a67f50
721238 F20110109_AADDLZ lebrun_d_Page_054.jp2
673f9160f57cc88e5f3e7e32f37ad61f
72da0c8e4732e375ad3088c810e89f755c01309e
F20110109_AADDNB lebrun_d_Page_003.tif
d9c2070119be65398317833f0eb77f24
5b7dda4f1f3d0d7a2628c2bc0b5b0526fb01750d
36827 F20110109_AADDMO lebrun_d_Page_085.jp2
ba0d19a99176968212b5470f17c44430
ebb6f3d9c22c9f643107b2385ca2cea1886b430b
F20110109_AADDNC lebrun_d_Page_006.tif
b920e7f035b36dd138a157db469abc40
4431cd16f499379c4ca336d01cab19772db668c5
F20110109_AADDND lebrun_d_Page_009.tif
46ec29ac738bd8d335dabe285e798320
059cd6945d3e8428fcd85e7007054a5a5bcf87b4
85503 F20110109_AADDMP lebrun_d_Page_086.jp2
becc9baf7a29250045d0fe6f865da6c2
a88e13424a19aa566c18a018ca0a2e4bd3d294fb
F20110109_AADDNE lebrun_d_Page_010.tif
87d4b9488f6ba277a32176f5e6760a20
b097b76ba2fdfdaabfaaefe291db62f05c42de07
100920 F20110109_AADDMQ lebrun_d_Page_087.jp2
48fe70a08ef18aa5dabc6d06fd7d85e8
8e9f92bce2cac487ad876b7dfbd457a62e4252de
F20110109_AADDNF lebrun_d_Page_011.tif
e0c3a39f8eb56ff381875ec8c572b950
99f5cc54fbe83cfc463b792924420dc3ca66ecf3
86598 F20110109_AADDMR lebrun_d_Page_088.jp2
9b61013a4b45b1da7efcc17b7382e245
c9afbea31076bde81d9d9050f101c18a5d4901e3
F20110109_AADDNG lebrun_d_Page_012.tif
389efad3da270149ca7ea5710e2cf7df
9616ff44be7ff6c564395e4bb1003f71c4841b5d
634218 F20110109_AADDMS lebrun_d_Page_089.jp2
e9f8ac3dfd839a110454d40f36b37873
004a39d761bddccf5ae02016c60441277afa88e8
F20110109_AADDNH lebrun_d_Page_017.tif
564ac706e039d89fcee1533c67193931
ebbba5dc14df3490df4862ce3b12d4aa48214e56
980452 F20110109_AADDMT lebrun_d_Page_092.jp2
f2e40364fe4c4ea392670c7967618d43
a6fb5e168de12723de59ca960e98c700114ea045
F20110109_AADDNI lebrun_d_Page_020.tif
a20a8c8c05e93ad2128290c79cefdebf
80808f66a6c63f2fb2de0924f99be05d36f8f1d2
1044251 F20110109_AADDMU lebrun_d_Page_093.jp2
f1a707c595b422e837f89379543cf4d5
48f2828da2aa24c7528d818e82b16acfcc9a7999
F20110109_AADDNJ lebrun_d_Page_021.tif
777d0579931071f6c65bf227a3b021e3
013a1d9b378686bb4d97cb7d8f02edd057a17632
588912 F20110109_AADDMV lebrun_d_Page_096.jp2
a6c1120ae7cb41234fb1c8bbf073199d
56be3eb5df1a34f64f7b0470fbd419ca7093b984
F20110109_AADDNK lebrun_d_Page_023.tif
c51f4d9f60f95cc77c6940f413fcebd9
acf6ec64fda746ef642f51ec62c612f7791d2f7a
111188 F20110109_AADDMW lebrun_d_Page_100.jp2
a3f592ca533cc3289c153bd3aa781917
427f00333ba1c2fbc261e2c06b97ec7c4cc0d3c0
F20110109_AADDNL lebrun_d_Page_027.tif
943ed1f07bb39ef9e1181f0b1152bca7
3730059e322066ae6c14236d72e8b09a64514197
124112 F20110109_AADDMX lebrun_d_Page_102.jp2
92ab59b9aa6788f7563577071acd63cc
0a231b024f927246e6bd92b2b1f138f783d3b5f0
F20110109_AADDOA lebrun_d_Page_061.tif
540451c00e7e5c9cae2d2e9bc32259b5
f72809f279d0cb1d66c114cfafd6db0bee7cf5b2
F20110109_AADDNM lebrun_d_Page_031.tif
cad787b205775d45924ee78fd7b534de
14b64b3f0e3386595b0d0fd3a9fa29f57d89abf6
25844 F20110109_AADDMY lebrun_d_Page_104.jp2
dc804ff246458767f54cb3c6d0352280
c1d7cb37f9c6b3a0067fbb9b7e405dc8b756f4b5
F20110109_AADDOB lebrun_d_Page_064.tif
39a3f03fdcfba341fcb8b52a8a7156d7
e673907deeb0cf25611499fec82b30dcf5ca187d
F20110109_AADDNN lebrun_d_Page_032.tif
98093d78b3fb656a269870e2f24df31b
c5d2a279970340bc2abd7160944114932083726c
F20110109_AADDMZ lebrun_d_Page_001.tif
639c1331bfdda17a4743e71adc4f36ef
5692996d42f950dbd868f300332013c1df365f43
F20110109_AADDOC lebrun_d_Page_065.tif
57b96b400dbbbae73e6387e4d1c3701b
9a74348f2f3a46c0d3436cd3d3370eb6815291f4
F20110109_AADDNO lebrun_d_Page_035.tif
dfa61ef77d402b7911f99eba8b12bad7
2fc0796e0add776e0e8c34a11c0656eed9b6b791
F20110109_AADDOD lebrun_d_Page_067.tif
04f2e9fd8298edf15349f5da317f530f
47efe0f13074d39bdbe0934134e3142ad211ee17
F20110109_AADDNP lebrun_d_Page_041.tif
80fb43bf1f70f4415599120eec8adef0
0f26fde5b86f6a626d142f678c2e0954f3d1e111
F20110109_AADDOE lebrun_d_Page_069.tif
2564230775f0a0e221d36a1eac15a96b
792ff1598e5db247fdff8e01bf954c7efad17a93
F20110109_AADDNQ lebrun_d_Page_043.tif
ed926d3e42cec149292f0dafa68ac908
94405bdf36ff81594298c8702415d7862e4bbce4
F20110109_AADDOF lebrun_d_Page_075.tif
c4342f3fd7cbdc791898570151fc89fa
f0ccd3b51cddb937e3ab32e22f626e41bb755d8d
F20110109_AADDNR lebrun_d_Page_045.tif
8669f09fc26f89b95787ee474f56acb5
630ed8a179f87e3a1c75fb70edb75bb4c96c54ef
F20110109_AADDOG lebrun_d_Page_080.tif
1e668ea15272fbdfe6b2f0740cd701ae
a1afbcd70baa00764bb23fd2bc21e4fb44fa03a3
F20110109_AADDNS lebrun_d_Page_047.tif
bd5a5c2da62d3f8fb9da6e6ab0abc6de
81476c11e8aeeb65f18bfb8d68b1741fd42a324f
F20110109_AADDOH lebrun_d_Page_082.tif
86d0a9dc1a64b3398313929b18483ab1
13399c96d966a50f5650a2cc043f2abc80d4b334
F20110109_AADDNT lebrun_d_Page_048.tif
1863a393b7eec5f9fc312ffa5db1a496
eeaafd6b6e6458b9dbb7613f869f1cfc9c3ce947
F20110109_AADDOI lebrun_d_Page_088.tif
6d2b808fdcbed5c2f8de7037954bacb4
0bf3b121114e30001c23ff80facc5d7da5f50006
F20110109_AADDNU lebrun_d_Page_052.tif
13b222a588107ee35471c6aa3c82692c
e41d1f2cdaa618757587cc08cb380af902a1ee00
F20110109_AADDOJ lebrun_d_Page_089.tif
a7ed51940cc070a600807649e72e2243
89dfa047d7420a98c5cae11f9c1865e0c5344254
F20110109_AADDNV lebrun_d_Page_054.tif
5def1125899bd18a3b95418dd33a2304
194ff46562526fe7e36108ea5e6b596d4f65f2c3
F20110109_AADDOK lebrun_d_Page_092.tif
7751cc4f13ea759f09e8683532f11eca
bffb3532242bf922bf1931bceb808c9e435dbc2c
F20110109_AADDNW lebrun_d_Page_056.tif
96569819ca6cfe96f413ba8c1f72e8d0
c4e671ced0cb8a0eeb426e6a780459ea89c91ed8
F20110109_AADDOL lebrun_d_Page_093.tif
0257c7b7eecae38b1877ed1de4416711
5f033d76a04342c9119c2363898f3cc2848da40e
F20110109_AADDNX lebrun_d_Page_057.tif
3a2bcc1ee511562136e70f611ef4eecd
9a325094be7ad81b1ce135491225d43e1aa441a7
F20110109_AADDOM lebrun_d_Page_095.tif
16da21373bbaac4ddd5fd822e7413e3e
1d7f0834edbbd40b4129032b1940927e2846ee8d
F20110109_AADDNY lebrun_d_Page_059.tif
0e030839c84a8ed54738acf9a29006c1
2a4a49520d837a9da2e56ff94ad834e3aa71046b
42293 F20110109_AADDPA lebrun_d_Page_022.pro
d8260c6168633125da0b3960555ed6a3
bac3430440d6a5727181c16b77de16843d2ed9b3
F20110109_AADDON lebrun_d_Page_098.tif
b61cdb9f091a558353eff429afaf39c1
940b47032c62a6c7d04e6da7d0e684f8f8a9fde8
F20110109_AADDNZ lebrun_d_Page_060.tif
b8374c8b911c00ccd87696cf42ebebdc
9d0331f423366e8daab3f85a3811224c8d0fd23b
44978 F20110109_AADDPB lebrun_d_Page_024.pro
5baa9390dfd36df9201d8bc3a2b52bc1
154272ccc7ba6c1964b6be384670b32f25d2adcc
F20110109_AADDOO lebrun_d_Page_099.tif
ef72e8172a98162f3bda8e57db4271a7
6b826a0380b3c5ebd6f934bd67502fac56ab66a5
8114 F20110109_AADDPC lebrun_d_Page_027.pro
446f0c0d4885158b1eabdb76a0900d08
ee1dda0daed7c7848e5a5b915df0706e77202f04
7838 F20110109_AADDOP lebrun_d_Page_001.pro
856405c672267416b41b1f8e29872f7c
831682c6c35bac9461fe6243415212005b7983b1
46103 F20110109_AADDPD lebrun_d_Page_028.pro
ef1b88fdcc34ff8996880c4bbff1afb3
518be4d31c47c4c9ebef78d1bfa7c8a729c1d097
74347 F20110109_AADDOQ lebrun_d_Page_005.pro
51270fd9b709e0e0d1426e0d23403478
3f43ab6ef8bdc18abf64a6378a11fb03a411965b
18996 F20110109_AADDPE lebrun_d_Page_031.pro
6cb1d0c90ce5ce4e589a78925ec40dec
e308c44ee88f77fa1123d3ec80ecbdaf77d2c707
12075 F20110109_AADDOR lebrun_d_Page_008.pro
d362412c8c98afe0e93cb91662b79801
5dc14a69f47b20d2a123a496bf7b636a09431d64
42486 F20110109_AADDPF lebrun_d_Page_036.pro
670cdf64d8f42a06cfa9a3b96510ed85
f34762e04621390b3393eb590413e4350ad7081d
61172 F20110109_AADDOS lebrun_d_Page_009.pro
c983d983b50f1f4b8d690f0a24b956d9
288ac3390e99b6f026779e930234943ec13b3bfb
39362 F20110109_AADDPG lebrun_d_Page_038.pro
ca7c971aca0261fcb9a9d742a569d48e
83f281b42a49bdfa90ce542bd620ce53f116b4ba
43107 F20110109_AADDOT lebrun_d_Page_010.pro
456c1687f910e64badbede58c67db9ed
1cfb8994e43cb0e391b00486e6cc185517750a97
31547 F20110109_AADDPH lebrun_d_Page_039.pro
71a7d6d2518d3393e97fe9186bc30668
37d4d22cc20cdf64c6f085f44eea0d93773e176b
36518 F20110109_AADDOU lebrun_d_Page_011.pro
51312d2f543093b61ef63d734ab6cc75
d868e518e757f00b3d402fea12c50ff8c5a01838
36437 F20110109_AADDPI lebrun_d_Page_040.pro
1d894638b06a3b2bbf91d7393233480a
cef0fbdf12402cdd5181934b7ea984ccb58ff3fd
11672 F20110109_AADDOV lebrun_d_Page_012.pro
0d14a7be701677cc582f418751234a2e
05331d7b937381476d277570d0cf1d5bcd83e75e
47485 F20110109_AADDPJ lebrun_d_Page_044.pro
e898be3f420346405c444a9fb8d112d6
93b51d6d899351364f3a5f4b82202169e48fdbb1
38254 F20110109_AADDOW lebrun_d_Page_015.pro
309f81f2287bcf667e3600e1dc4b0d56
e513ca6aa34b45d25f55c638de3af5221a363c6f
32354 F20110109_AADDPK lebrun_d_Page_045.pro
9be2def12e19e939c8d3ebc9435c2d44
9578005b91f2033f416235b3559c84820a95ee71
50540 F20110109_AADDOX lebrun_d_Page_017.pro
74fde370c239f7b4993432f23d38d451
e67bd9fb8d4847d6168c3763e13fbdce63990a2a
27805 F20110109_AADDPL lebrun_d_Page_047.pro
9b4a90f98446127188d9a3e5ea5ccecd
0a01f2e24dca8c5edbbe54d922c95af569960ab0
50894 F20110109_AADDOY lebrun_d_Page_018.pro
180f9ae275ffc49a699b695782a39125
c3da2aa80272dfc1a452bc397a6c788b778d4135
22947 F20110109_AADDQA lebrun_d_Page_077.pro
030b9d2ca79d30ddd5fa176d750ccc28
82b1d0db303e96331bdd42aa2d102fe76d2b9179
44081 F20110109_AADDPM lebrun_d_Page_050.pro
d6a700082027ed6fdb1202fdb1054afa
1fc36f20bc0e3cb19366f9fb564ba7e536077d19
49465 F20110109_AADDOZ lebrun_d_Page_020.pro
d8b2cd577622a4d67f31dbad239601bf
60f2d71189ebee9ab31869cee5383a214843e986
24862 F20110109_AADDQB lebrun_d_Page_082.pro
f66cd4e97bdbd8183455fb9ea9530279
4e67ef5d8c7682d412426d2c28865d0db80f07ec
23546 F20110109_AADDPN lebrun_d_Page_051.pro
e5b0cd58d9ffddc039f0dc22d3789ab5
7a0fe503b084be6ef1688a0e520a1f47357bfcce
38908 F20110109_AADDQC lebrun_d_Page_088.pro
bfd96f956b196ca8c4e2c465f68899ff
cd7aa45dfedfd6440e449c12451539480db84938
26443 F20110109_AADDPO lebrun_d_Page_052.pro
b41a71bf1748eda0d3ff1b3d51f87c41
c62d7d0b5be6264be78bc2a00f994e98962a86b9
10289 F20110109_AADDQD lebrun_d_Page_089.pro
5663e76d50201802424b9e011d606bea
33cd0819472e3a96b9bd7db977e5cadd634b26b6
8331 F20110109_AADDPP lebrun_d_Page_053.pro
8997ccc92dee9ded8f538182a6c56411
a5c5b95071273e771e4371f524922d5baecd7a69
16561 F20110109_AADDQE lebrun_d_Page_095.pro
1dace865cabe5320367ed0d410229d11
c8af7eef29780e5acce1ea818119984aa13360ee
4392 F20110109_AADDPQ lebrun_d_Page_055.pro
7b23657d1ab38509e2d25d30c3190a2a
abf00f3fa094ce20a581791301a71e2dea9d7f66
18511 F20110109_AADDQF lebrun_d_Page_098.pro
0c8c9b0315788d7a1efabf36e103ae2f
07d1d28252dfe1d1c09d273efab76ee984ba0673
34805 F20110109_AADDPR lebrun_d_Page_059.pro
07ab3e52b777278ee7b236638988a15c
6715284cd07bc3099e37bf76ebd65df38b7ce234
59718 F20110109_AADDQG lebrun_d_Page_101.pro
7966de633ea54580037e7eb9f4e7eed7
4e92774149e4671384cf2870499109d0364730f8
44210 F20110109_AADDPS lebrun_d_Page_062.pro
a42ba83af5674d4d2c3a9266f936105f
44916287809ee972db0762e5fa71249b878846e6
59345 F20110109_AADDQH lebrun_d_Page_102.pro
2f18ac223f978bae8ee1af9df4d26815
acd72134efeb01006b9bac3424022371c8605585
35634 F20110109_AADDPT lebrun_d_Page_066.pro
dd87475bd6db62262ba51cf508d411e3
61df39c9ebe67b10044b249a9a4691e913650879
9982 F20110109_AADDQI lebrun_d_Page_104.pro
cc1405b76b89470b83907068ae581093
2efd8aa2509b7429a3824b44b58bbe135869fdf6
44845 F20110109_AADDPU lebrun_d_Page_067.pro
c5879c3a85054e2d4e08062fac2db40b
c65b37a2fca2a7746a31a64f1eae8f71414ae63e
110 F20110109_AADDQJ lebrun_d_Page_002.txt
1f1fafaf2abeb748104a5932a1b7f726
c18ccb94a7393504a3080dd657622dba1661a896
24482 F20110109_AADDPV lebrun_d_Page_068.pro
3ecd6c0d04aec7bda0e60db0920e4232
6b26ba1f45bcfb98638f2130cd3e86a484a0c2dc
128 F20110109_AADDQK lebrun_d_Page_003.txt
fb9b043482a0c02ff7c046c444d5da58
66b792508dba09f3bc9f565fc7b39c48f88a222c
34876 F20110109_AADDPW lebrun_d_Page_071.pro
6a236841fc9d9a3c30073b70fd150228
2ec38cb84f0d134cb53bca2dc6d55bf992179d8b
1146 F20110109_AADDQL lebrun_d_Page_004.txt
d450268d983db35dfb380503382c94a5
7fdcb48c3121357009974295a1fdc310471139ce
36933 F20110109_AADDPX lebrun_d_Page_072.pro
c0d2b5c5d3d8d12c9cc071fc760891aa
4fa6d08f79c8e295529d6ce45a5f9d8d20c950bc
1818 F20110109_AADDRA lebrun_d_Page_036.txt
56ab83a22900ef23985f22e7c3fd9043
28839c910e6722dcdadd11f17b561ceefac5401b
835 F20110109_AADDQM lebrun_d_Page_007.txt
c1151a22e6127400b3d8654f6d765d33
a92fc874e5ddc7b004a0ce9f7ce862b9c5d2aa0a
28488 F20110109_AADDPY lebrun_d_Page_073.pro
afece8dfbe54471b2d6b5bc147703e74
e94e4a3251737eb9d1199e907bb3d80e0b765450
1953 F20110109_AADDRB lebrun_d_Page_038.txt
a1040cf09116bd917543397e901b3196
bf119691d664fa86195e73881292813ee23753a2
1810 F20110109_AADDQN lebrun_d_Page_013.txt
227eee51ca8e44dfaac4e3c0c6774f42
7387c69ec5198c6a8a5ecd011057ff73d7f7f06b
32417 F20110109_AADDPZ lebrun_d_Page_075.pro
211c34687957496e4d0001963838e444
6bcef5d2149e79e5ae51f61b1d95dc39ad18f693
1444 F20110109_AADDRC lebrun_d_Page_043.txt
40442dc754ae93c96693d574688cdd5a
2a6a318191f2cb0de649d936253b1da48d2f0960
2027 F20110109_AADDQO lebrun_d_Page_014.txt
d31384ef857aad5a8658348f6a9ace78
9acbf3b49d0e691bf8ccd56bb8f4a6dc800bc917
1471 F20110109_AADDRD lebrun_d_Page_045.txt
52b161008b1d5178cb1cb99fa9de0d3c
7ecb33548b5e796278a84d545b7972547ce22ee9
848 F20110109_AADDQP lebrun_d_Page_016.txt
bf4c8aa901a3f59a464096dda772ccce
20526deafa34236a7f93cb4b2e2b5a6a05b96a31
1691 F20110109_AADDRE lebrun_d_Page_048.txt
835db564b64cae325ba1207f3f738ab4
5fbc4e01ee14066215dd27da9848e2e9e45c432d
1995 F20110109_AADDQQ lebrun_d_Page_017.txt
7e87b2cf226efa56e3abdaea6420a21d
01a1a37c6b515d28cc77f6ad63164b0ed87eebc4
1264 F20110109_AADDRF lebrun_d_Page_052.txt
3f014d4aa8d798fff6bc7bf89acb9a6a
39d9f40302baba28d63aecd3e77a1f630217b05e
1919 F20110109_AADDQR lebrun_d_Page_019.txt
1e2948eb73a2da5d589e9fb2b0cddc53
45aa3cb2aa6c00936fb7922c60e175146e3fa36f
215 F20110109_AADDRG lebrun_d_Page_055.txt
c5722c979253f06dce89fd6dc0e79e7d
796190ff45dee5df1acdbd4af66e2b5142806ddc
2004 F20110109_AADDQS lebrun_d_Page_020.txt
d901eb9a2d888d0c617a8e4b3f8b70aa
b17d0bc519ec064312b015119ec3736324ceb015
2022 F20110109_AADDRH lebrun_d_Page_058.txt
c14de0abf097d3d26d27194717506242
f23d0fa937940bce0bcb3353c77a166ba94a24a8
230 F20110109_AADDQT lebrun_d_Page_021.txt
0ac3cdf8341a593a0a7533c0fd20b0c1
aa8bb31d91b7f008733714a564fee20472cd3a7d
1042 F20110109_AADDRI lebrun_d_Page_060.txt
b250d21a5ae10848e8518367cd0c743a
86853c7ee54afa1c910ef55cee64d271ba48ab15
1960 F20110109_AADDQU lebrun_d_Page_023.txt
c68f7e00940fdcb024ede40c81eebb9e
c678d82dde7cf4e37a19c6caeb5fa23513bea36e
1868 F20110109_AADDRJ lebrun_d_Page_064.txt
35dffa87748ba1bef8aa929814e17b10
d36b1f67ee375d6a6d69378c4cb5b3af84aec9f7
1430 F20110109_AADDQV lebrun_d_Page_030.txt
3d979c782ac7e56852d46882b01896fc
42f4966a3cab4067a4d636384bb60608b451e40f
1838 F20110109_AADDRK lebrun_d_Page_067.txt
694987e6ca34d361b372cf56e2db8b07
0f757c82b215224fd03c21754b43043c7bd68453
649 F20110109_AADDQW lebrun_d_Page_032.txt
f143af4b79a2f38df5eb22b3fca86db6
30d3447ffeeb1195757d464d62c8feab59daaee6
2053 F20110109_AADDRL lebrun_d_Page_069.txt
4e61f865cda2171d7a8e47bebaaa722e
9031b89701b885d64720f01b6367ca2b1a6bbc95
1961 F20110109_AADDQX lebrun_d_Page_033.txt
ec525305ae26b28f652be4c042f07a4f
55cefc30a845ccba8d18dc1188c4ca70158ab8ea
1260 F20110109_AADDSA lebrun_d_Page_099.txt
c6e681376aa5fa1f14cec5f0fbd8190e
27069052a1386cb9ab57c91dda267bd7965817ae
2632 F20110109_AADDQY lebrun_d_Page_034.txt
f66f800105df0f1489b5af0bf7d5c23b
75e65b1f5a14e9cf015f8e7945f46e01081fa0d9
1636 F20110109_AADDRM lebrun_d_Page_072.txt
12558af9ebe1b54c9e81930f9650e242
af57f96954abd5474ad006342a583574fbfb52f6
1226 F20110109_AADDQZ lebrun_d_Page_035.txt
34fc0c7c9142b49e9df9b780a86298c6
c1978c3bb6dbef2ff74cfa25275b8a2e0feeb593
2492 F20110109_AADDSB lebrun_d_Page_101.txt
50a5424e36f0efbf32d06280f11521d8
b77fff6ecc6c2f1bf558da74dd135e16f2fe5fee
1196 F20110109_AADDRN lebrun_d_Page_073.txt
db86ea184d117c123f09f9fb3cce3750
6e7ced9bab4e5cffba4788be29bba736127fd9f5
449 F20110109_AADDSC lebrun_d_Page_104.txt
c5d50b6b081595e0547502086d695d11
1135b848bd88943f73e825d07eccfdff0c0e8217
1387 F20110109_AADDRO lebrun_d_Page_074.txt
d3fcded634d6aedad5dae9876378f4a6
d30ff4868a9abe11e403439d8e3d2a88fcf3bd52
19617 F20110109_AADDSD lebrun_d_Page_001.QC.jpg
82625f5b47e5f9757f3198d5682f5723
baf2a88b47cc5b0c33ffdac9fcc41cfd032dd097
1058 F20110109_AADDRP lebrun_d_Page_077.txt
4e6b4d0b177fc7eb0853d1e7227348e9
befd6da24cb635d0608cda2fb4cf542bb8471b22
5947 F20110109_AADDSE lebrun_d_Page_003.QC.jpg
03d76618d8096c8108997600a4c70dff
993ff21ca74396f59b7f7710af073c6dc09625e3
1393 F20110109_AADDRQ lebrun_d_Page_078.txt
3cdc313c8705b5a62cc765164fdd5593
94fd89ef09cd09faf944332226f21ce11ed88f7f
3300 F20110109_AADDSF lebrun_d_Page_003thm.jpg
9adfd035bb9ca6d3ee951c09def97ad9
9d6bb19fc9b55ce1efaef98e109d0429d8834171
1998 F20110109_AADDRR lebrun_d_Page_079.txt
83ed4bd0c28fd2a7513dce3a6fa06073
6640a1119dc4eba51d32b6d6f95463af717bb62d
44149 F20110109_AADDSG lebrun_d_Page_005thm.jpg
a442f117a858f5983ef8be1537d02e45
9e0f4f2a1f2a7b2fc2f33faaeb9c75f70b90263f
997 F20110109_AADDRS lebrun_d_Page_080.txt
ae98c86615808cbb32e72f11fff8a1d2
f32081a298e07000ec88a8b53f9a1543785b8f25
45504 F20110109_AADDSH lebrun_d_Page_010thm.jpg
64b2827ddec3c085dbd092f6898d9306
4fb87d1fa15f139537636e2454d6114daeb394fb
1488 F20110109_AADDRT lebrun_d_Page_081.txt
e729efecef83b650c460e3e40ecaeed9
e3a1562720f12520d95b0faa1fbd747757c61280
58305 F20110109_AADDSI lebrun_d_Page_011.QC.jpg
67dfcfac96b2bbe56dac29001d1100b1
227394b36c80146cb5bb427b596a79939ffa5302
1091 F20110109_AADDRU lebrun_d_Page_082.txt
d981539835920d31d7400adc6f5035bf
6c316ff65913092607e3910dbb8e2f9d6f3eae0d
18783 F20110109_AADDSJ lebrun_d_Page_011thm.jpg
45bcd67b3d40a76f462d024ec011f2dc
eef866d9481808e10f765b5ddfaf672000484b2e
1153 F20110109_AADDRV lebrun_d_Page_090.txt
a25a479d070a665d4acb1d9faf1f8176
9245a82ec6cc3cd2f12c17cb9d7276e65fbdeb1e
7498 F20110109_AADDSK lebrun_d_Page_012thm.jpg
2bd5b40b0ee317846f51a774729a491d
23de58b5fcd4edff8ca130133ed13398b06758f2
951 F20110109_AADDRW lebrun_d_Page_091.txt
5fbc53374979b6b00fa070e2477fc9a2
d0ce798941ea9c7daff931a7d372a2b3a8880f7d
43512 F20110109_AADDTA lebrun_d_Page_031thm.jpg
ece3e1b918ab0744ba2291a8cc818d02
d1a204a6de9fb482a21de733adfecbf0a8d3c0c5
92859 F20110109_AADDSL lebrun_d_Page_013.QC.jpg
9eb634d17e93265502701b01c61fc6ae
ced043d298a8f9aa26dba8b51297f10a2280d432
1172 F20110109_AADDRX lebrun_d_Page_093.txt
b0b5fcf774d8993411f1b88ac74fe2dd
9e3f164faa93b8892db95606cf3579f900a9f6fd
29988 F20110109_AADDTB lebrun_d_Page_032.QC.jpg
75e95a167a36701cce318ee085893f4e
bca54cc4e486a11bd528ad37355468e28536f3ca
47105 F20110109_AADDSM lebrun_d_Page_013thm.jpg
8e3cba338a1fd3fbb857639e1b30284f
97d7f4e2ce325b9edad411d0eef8c2042d22b2e7
F20110109_AADDRY lebrun_d_Page_096.txt
e356c0bb0c189c7544b0ae5e2bb11cd1
76d64feffbda1f872068aa7a84fb170002a1826b
78965 F20110109_AADDSN lebrun_d_Page_014.QC.jpg
8f45d761c0a84a25e16dca43cc18859a
ab009622378bc87d61d897e48c048f07cb566f65
968 F20110109_AADDRZ lebrun_d_Page_098.txt
e996811ce64b881dabc67562e30f743f
03841a7b99bf490e0f8f6af51dd04544203d3a28
21095 F20110109_AADDTC lebrun_d_Page_033thm.jpg
41d83bb26d871ba3d6978f0b83f424ef
936d464a78611fbb37339f36b601b13c6465d8e9
59749 F20110109_AADDSO lebrun_d_Page_015.QC.jpg
10cb71eadcf6bc2b98cf46eb13336685
5b19add4aea4abe83f8634c7fc170c45f692dc80
25879 F20110109_AADDTD lebrun_d_Page_034thm.jpg
26a59f0fb08e8a468626e418f5c07e94
0760849caa8f9bf94520f6824dd749238fb047f7
77534 F20110109_AADDSP lebrun_d_Page_017.QC.jpg
216e98ae226efd3be460120dcf429452
0d0d84145c51eb80fc742d4b32eeabf653382444
44943 F20110109_AADDTE lebrun_d_Page_035thm.jpg
85f2790c0ae1bab40c444bc2a36a8041
8785b1a9461bbb33fa4d6503f2c276d1cbd0adce
24452 F20110109_AADDSQ lebrun_d_Page_017thm.jpg
19c54693bdf7afc54e4792af299172d6
7e7991ba6eac9f55dc92f94ff35dd43fc3efa251
65810 F20110109_AADDTF lebrun_d_Page_036.QC.jpg
b2b6021eddab943ecf11d445b5ef4fbc
f65fb19950a1f999c49431b8891892597622b581
79864 F20110109_AADDSR lebrun_d_Page_018.QC.jpg
6b31815fba9766943b79af6933e7d429
6bce851f1f221d9822eaca1e637f9576eb3d7a3d
83189 F20110109_AADDTG lebrun_d_Page_037.QC.jpg
578a2d2605bf875cadb288626e30b36e
a4805842e92f55226c8c9c8e98658b283c14f130
11575 F20110109_AADDSS lebrun_d_Page_021.QC.jpg
0a4ed4dd62b17d1a8cd9202ae027e545
109690b984c6fb5fd8d963c26eb2073888ab6159
44010 F20110109_AADDTH lebrun_d_Page_037thm.jpg
6cfa5b14f24aeee66e1b420f635b933d
425cd4a464178b90a6195899f6874100dc746d37
68826 F20110109_AADDST lebrun_d_Page_022.QC.jpg
8e710758a7c4acf0706c9775f662c007
017212efbfdb24d371cf8950649a430b6e0afc06
45321 F20110109_AADDTI lebrun_d_Page_038thm.jpg
5e2ca750c154a575a3203e42ffd52a76
ce3e91b0ba6db6221bfa3789dcad48c1a8e56047
22146 F20110109_AADDSU lebrun_d_Page_022thm.jpg
95f9cbb506fab73743f632f7c15f60fe
f7d0ec87fc51472e500e7e49822a4ef7dbb44c42
44312 F20110109_AADDTJ lebrun_d_Page_039thm.jpg
a2df9ca71e52a4aac4665fd388364f3b
93533ee4b26325fb8b0f5a7edffaa507e5795862
74285 F20110109_AADDSV lebrun_d_Page_023.QC.jpg
4bc0c22e505a31a97c96853c8ecfafe3
b99d629de9fc30977f06007b7ef2e74e2dc6e518
61741 F20110109_AADDTK lebrun_d_Page_040.QC.jpg
3569837e3c679680d62aa9b765753173
ba28b44f4c6c377c8629ddfa7360e2436c8e0987
79691 F20110109_AADDSW lebrun_d_Page_025.QC.jpg
701c81b713ddda13862a89d9d3d116ab
6e890781022cd98302eea98a67fedbc9660e019d
20659 F20110109_AADDTL lebrun_d_Page_042thm.jpg
ea65ce5e638423fa70171367798d5f68
07ab7030010c9fc1a2e2edd3dc53b45d2874a6ec
85984 F20110109_AADDSX lebrun_d_Page_026.QC.jpg
a328fe111106dd8aa3cb7b53516e8f1a
d4c6fa39318c2083cb0aa12eece4d914befc9562
40763 F20110109_AADDUA lebrun_d_Page_061thm.jpg
c0dad3e728091b42a44ad630dbfb66ee
39caacc0428f6baea23998154dd56fbe4070fb11
42334 F20110109_AADDTM lebrun_d_Page_043thm.jpg
563516351fd4b02e36a7e979ed888d19
9a8689eac967c81a5abfb8fb5af91634f87ff35f
25892 F20110109_AADDSY lebrun_d_Page_029thm.jpg
14bcbb69f6a93f34f6f66e10262d31ad
fa9e423472e470de4e6e733cd7ad326fa6c1379b
44765 F20110109_AADDUB lebrun_d_Page_065.QC.jpg
5844a4bf7a310bcedfd99a38201c2df6
5dd6fdaf0cae89b012d72a00c138a043931d5d37
23361 F20110109_AADDTN lebrun_d_Page_044thm.jpg
3e7e029ea97f6ff459f2c4c0e3cab6a4
07a2bf3ce39791e6bd59cb908feb2c94416ad8a1
42326 F20110109_AADDSZ lebrun_d_Page_030thm.jpg
fd7b9a5ae11b17b22a6b52b76c911118
95ba4c3cd33abf5320146e1f51ece3e2efc95f7a
14863 F20110109_AADDUC lebrun_d_Page_068thm.jpg
6d47c6c841b0ea22a9da28005f33f88a
56b866f430ea51c4685e40eb4c2004315a47a55c



PAGE 1

ANALYSIS OF NEONATAL HEART RATE VARIABILITY AND CARDIAC ORIENTING RESPONSES By DEVIN N. LEBRUN A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF ENGINEERING UNIVERSITY OF FLORIDA 2003

PAGE 2

Copyright 2003 by Devin N. Lebrun

PAGE 3

Dedicated to my father, Denis Lebrun, for his love and support.

PAGE 4

ACKNOWLEDGMENTS I would like to thank my advisor, Dr. Johannes van Oostrom, for all the advice he has given me on realizing my career goals. I knew what I wanted, but he helped me get from there to here and now onto somewhere great. He endured barrages of scatterbrained, unintelligent questions and always surfaced with the answers I needed. I am grateful to Dr. Charlene Krueger for allowing me to share in her research and express my gratitude to Dr. Krueger, Dr. van Oostrom, and the College of Nursing for providing me with funding throughout my research. I would also like to thank Kelly Spaulding, John Hardcastle, Dr. JS Gravenstein, Dr. Richard Melker, and Dr. John Harris for their support in various areas. Words cannot describe my appreciation for my parents. They have taught me to be determined, self-motivated, and not afraid of failure. I thank Linnea, for her companionship and assistance throughout the past 6 years. Finally, thanks go to all my friends for keeping me from doing something stupid (well, more stupid). iv

PAGE 5

TABLE OF CONTENTS Page ACKNOWLEDGMENTS................................................................................................IV LIST OF TABLES..........................................................................................................VIII LIST OF FIGURES..........................................................................................................IX ABSTRACT......................................................................................................................XI CHAPTER 1 INTRODUCTION........................................................................................................1 2 AUTONOMIC NERVOUS SYSTEM AND HEART RATE VARIABILITY...........4 Autonomic Nervous System.........................................................................................4 Developmental Changes...............................................................................................6 Vagal Tone and Homeostasis.......................................................................................7 Cardiac Orienting Response.........................................................................................8 Respiratory Sinus Arrhythmia......................................................................................8 3 PRELIMINARY RESULTS AND STUDY PROCEDURES....................................10 Preliminary Findings..................................................................................................10 Proposed Study...........................................................................................................11 Recitation Sessions..............................................................................................12 Test Sessions.......................................................................................................13 4 TESTING....................................................................................................................16 5 QRS DETECTION.....................................................................................................21 Noise/Baseline Wander..............................................................................................23 Slope Detection...........................................................................................................27 Outliers.......................................................................................................................32 Direct RRi Analysis.............................................................................................34 Slope Analysis.....................................................................................................36 Differential Analysis...........................................................................................37 Interpolation................................................................................................................39 v

PAGE 6

6 STATISTICAL ANALYSIS......................................................................................44 Time Based.................................................................................................................44 First Order Statistics............................................................................................44 Poincar Plots......................................................................................................45 Quadrant Analysis...............................................................................................48 Frequency Based.........................................................................................................50 Fast Fourier Spectral Analysis.............................................................................51 Lomb Spectral Analysis......................................................................................52 Empirical Mode Decomposition..........................................................................54 Statistical Analysis of Parameters..............................................................................57 7 DISCUSSION AND RESULTS.................................................................................59 Engineering Results....................................................................................................59 QRS Detection Algorithm...................................................................................59 HRV Plots and Statistics.....................................................................................60 HR Plot.........................................................................................................62 Quadrant Plot................................................................................................63 Poincare Plot................................................................................................64 Fourier and Lomb Plots................................................................................65 EMD Plots....................................................................................................67 Clinical Results...........................................................................................................68 8 CONCLUSION...........................................................................................................72 9 FUTURE WORK........................................................................................................74 Record Baseline Adult HRV Data.......................................................................75 Creating the Index of ANS Maturation...............................................................76 APPENDIX A TIME-DOMAIN HRV ANALYSIS...........................................................................77 Week One...................................................................................................................77 Week Two...................................................................................................................78 Week Three.................................................................................................................79 Week Four..................................................................................................................80 Week Five...................................................................................................................81 Week Six.....................................................................................................................82 Week Seven................................................................................................................83 B FREQUENCY-DOMAIN HRV ANALYSIS............................................................84 Week One...................................................................................................................84 Week Two...................................................................................................................85 vi

PAGE 7

Week Three.................................................................................................................85 Week Four..................................................................................................................86 Week Five...................................................................................................................86 Week Six.....................................................................................................................87 Week Seven................................................................................................................87 LIST OF REFERENCES...................................................................................................88 BIOGRAPHICAL SKETCH.............................................................................................92 vii

PAGE 8

LIST OF TABLES Table page 3-1 Organization of Recitation and Test Sessions............................................................11 7-1 Shows the performance of the QRS detection algorithm...........................................60 7-2 Results of the repeated measures ANOVA with independent variable (SV power)..69 7-3 Results of the repeated measures ANOVA with independent variable (HF power)..70 viii

PAGE 9

LIST OF FIGURES Figure page 2-1 The divisions of the central nervous system..............................................................4 3-1 Time-line for the windowing of test data.................................................................14 4-1 Example of text file generated from Data Acquisition Program..............................18 4-2 The setup for the test sessions..................................................................................19 5-1 QRS detection algorithm..........................................................................................23 5-2 Magnitude and phase response of the 60Hz notch filter used to filter the raw ECG signal........................................................................................................................25 5-3 Magnitude and phase response of the bandpass filter used to filter the raw ECG signal........................................................................................................................26 5-4 Magnitude and phase response of the differentiator function..................................27 5-6 Example of detected QRS complexes and beat intervals.........................................31 5-7 Histogram of RRi distribution from one subject......................................................33 5-8 Standard deviations of the Normal Distribution......................................................34 5-9 Illustrates the bouncing effect and upper and lower bound symmetry problems with using the direct RRi analysis method for tagging outliers.......................................35 5-10 Illustrates an RRi trend that would not be preserved using the slope analysis method for tagging outliers......................................................................................37 5-11 Outliers that have been tagged, using the differential method, are marked by a green dot. Shows the efficiency of the method to preserve trend and remove bouncing effects.......................................................................................................39 5-12 LaGrange Polynomial Interpolation.........................................................................40 5-13 Linear Interpolation..................................................................................................41 5-14 Cubic Spline Interpolation.......................................................................................41 ix

PAGE 10

5-15 Cubic Interpolation...................................................................................................42 6-1 Poincar plot representation of RRi. Since data points overlap, colors, coordinated with the visible spectrum, are used to indicate magnitude at each point.................47 6-2 Defines the quadrants used in Quadrant Analysis....................................................48 6-3 Example of Quadrant Analysis applied to an RRi dataset.......................................49 7-1 HR plots of example RRi data. From top to bottom they represent: the (a) developmental, (b) prestimulus, (c) stimulus, and (d) poststimulus periods............62 7-2 Quadrant Analysis plots of example RRi data. From top to bottom they represent: the (a) developmental, (b) prestimulus, (c) stimulus, and (d) poststimulus periods.63 7-3 Poincar plots of example RRi data. From top to bottom they represent: the (a) developmental, (b) prestimulus, (c) stimulus, and (d) poststimulus periods............64 7-4 Fourier PSD plots of example RRi data. From top to bottom they represent: the (a) developmental, (b) prestimulus, (c) stimulus, and (d) poststimulus periods............65 7-5 Lomb PSD plots of example RRi data. From top to bottom they represent: the (a) developmental, (b) prestimulus, (c) stimulus, and (d) poststimulus periods............66 7-6 Hilbert-Haung spectrum of the COR windows of an example RRi dataset.............68 7-7 Plot of mean SV power with standard deviation bars across sessions.....................69 7-8 Plot of mean HF power with standard deviation bars across sessions.....................70 x

PAGE 11

Abstract of Thesis Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Master of Engineering ANALYSIS OF NEONATAL HEART RATE VARIABILITY AND CARDIAC ORIENTING RESPONSES By Devin N. Lebrun August 2003 Chair: Johannes H. van Oostrom Major Department: Electrical and Computer Engineering We have applied signal-processing methods to evaluate the clinical significance of heart rate variability (HRV) features, assess developmental changes in HRV, evaluate vagal tone, and demonstrate a cardiac orienting response (COR). The study is based on a sample of 28 low-risk premature newborns randomly assigned to one of two groups. Group 1 is exposed to an auditory stimulation beginning at 28 weeks GA and Group 2 begins exposure at 32 weeks GA. An eight-minute ECG recording is taken once a week from 28-34 weeks GA. The recordings were analyzed in the time and frequency domain using MATLAB. Extracted features are used to classify developmental changes in HRV across individuals and subject groups. In addition, we have used a staggered grouping approach to separate cardiac orienting responses from reflexive responses. At the conclusion of this thesis, the sample size (n=5) was not large enough to elicit any developmental trends in GA. However, we were able to extract useful features to xi

PAGE 12

describe developmental changes in HRV and we have acknowledged Empirical Mode Decomposition as an effective measure of a cardiac orienting response. Also, we have identified numerous exogenous interactions and perturbations which can interfere with the baroreceptor and chemoreceptor responses controlling HRV. These interactions were not considered in the original design of the study but will now be considered during final analysis. xii

PAGE 13

CHAPTER 1 INTRODUCTION The premature infant faces many obstacles beyond the control of present day medical science. Since the loss of the in utero environment strips the premature infant of the most advantageous setting for growth and development, it is increasingly important for medical science to reach a greater understanding of the developmental changes during this final trimester. Survival rates of neonates will continue to increase as science begins to elucidate more regarding this crucial period of development. In general, scientific advancements are predicated upon a divide and conquer approach. It is important to embrace the past discoveries as background for future research. Therefore, this study expands upon a preliminary study conducted by the Principal Investigator (PI), Charlene Krueger, Ph.D, ARNP. The findings of the preliminary study indicate that the analysis of heart rate variability (HRV) can provide significant insight into the development of the premature and fetus autonomic nervous system (ANS). In collaboration with Dr. Krueger, the current study moves from the comforts of the uterus to the premature infants continuing development in a Neonatal Intensive Care Unit (NICU) setting. About a third of premature births occur for no apparent reason and with little or no warning. The list below describes some known causes of premature birth [1,2]: Pre-eclampsia, occurring in about 1 in 14 pregnancies, causes around a third of all premature births. Pregnancies involving multiple fetuses are likely to end early. Stressful events can elicit the mother into a premature labor. 1

PAGE 14

2 An infection involving the uterus may trigger an early delivery. The mother's water may break early, starting the delivery process. A shortage of oxygen exchange to the placenta may cause a doctor to advise a caesarian section. Malnutrition or chronic diseases, including high blood pressure, diabetes, kidney disease and hypothyroidism can cause premature births. Regardless of the cause for an early birth, premature infants are commonly highly dependant upon the continuation of breathing support, intravenous nutrition, intravenous antibiotics, and phototherapy. These infants are frequently placed in NICUs, which specialize in caring for the very young and/or very small. The neonates high dependency upon medical interventions demonstrates a need to establish criteria regarding his or her efficiency at regulating homeostasis. Homeostasis is the maintenance of equilibrium in a biological system by means of automatic mechanisms, which counteract influences trending toward disequilibria. It involves constant internal monitoring and regulating of numerous factors, including oxygen, carbon dioxide, nutrients, hormones, and organic and inorganic substances. Due to homeostatic controls, the concentrations of these substances in body fluid remain unchanged, within limits, despite changes in the external environment. Efficient regulation of these parameters is necessary to support life, on all levels. Therefore, the ability to evaluate homeostatic controls would facilitate the decision process leading to the neonates discharge from the unit. The events surrounding the control of homeostasis will be discussed further in Chapter 2. We hypothesize that an index of HRV will provide sufficient insight into the maturity level of the neonates ANS. In turn, this will provide us with information regarding the ability of the neonate to control events of homeostasis. In a study by

PAGE 15

3 Chatow et al, increasing conceptual age, CA, or gestational age, GA, is correlated with neonatal autonomic control. Chatow used a power ratio, discussed in Chapter 2, which was assumed to reflect the sympathovagal balance of the ANS [3]. In a study of the fetal baboon, Stark et al. suggest that Sequential decreases in fetal heart rate, increases in RR interval (RRi) variability and increases in changes in RRi and RRi with age imply an overall maturation in autonomic cardio-regulatory control processes. Increases with gestation of high frequency components of variability are compatible with enhanced parasympathetic modulation of the fetal heart rate [4]. This thesis will attempt to expand upon these studies to examine developmental changes, cardiac orienting responses, and to introduce a maturation index based on time and frequency-domain analysis of neonatal HRV. Current methods of frequency domain analysis tend to center around one method of analysis. Therefore, new methods of analysis will be constructed as a result of combining the most functional signal processing tools described in recent literature. This thesis will explore and compare a variety of spectral analysis methods and focus on those methods deemed most clinically significant. In addition, we will parallel techniques used to examine the frequency spectrum of HRV with time-domain techniques to provide a complete clinical evaluation of each patient and grouping.

PAGE 16

CHAPTER 2 AUTONOMIC NERVOUS SYSTEM AND HEART RATE VARIABILITY Autonomic Nervous System The ANS is part of the human central nervous system (CNS). The CNS can essentially be divided into three closely interacting sections: the motor nervous system, sensory nervous system, and autonomic nervous system. The ANS originates from the brainstem at the base of the spinal cord. It includes the nerves, which innervate the smooth muscles of the heart, internal organs, and glands. The ANS is subdivided into the sympathetic (SNS) and the parasympathetic (PNS) systems, which originate from different areas in the brainstem and spinal cord. CNS Sensory Nervous System SNS PNS Autononmic Nervous System Motor Nervous System Figure 2-1. The divisions of the central nervous system. 4

PAGE 17

5 Both systems work simultaneously in an antagonistic nature, with an increased activity in one, inevitably causing a decrease in the other. However, neither substructure of the ANS is ever completely disabled by the response of the other. While both the branches of the ANS serve to regulate the internal organs by stimulating or inhibiting their activity, these two substructures respond to different biochemical transmitter substances [5]. The actions of the PNS are started by the release of acetylcholine. Parasympathetic nerve fibers slow the heart rate, narrow the bronchioles, and cause the release of insulin and digestive fluids. On the other hand, the actions of the SNS are started by the release of norepinepherine. SNS responses include relaxing the tubes of the bronchioles, speeding up the heart, and slowing the digestive tract muscles, while increasing the release of glucose from the liver [5]. Given the broad spectrum of physiological events controlled by the ANS, it is important to establish criteria to evaluate its tone. Past research has indicated, through the measurement of heart rate variability, an inference can be made regarding the functionality of the ANS [3,6,7]. HRV is defined as beat-to-beat fluctuations in heart rate attributed to electrical impulses generated by the sino-atrial (SA) node. The SA node is commonly referred to as the heart's internal pacemaker, and the vagus nerve, or the Xth cranial nerve, inhibits this pacemaker. The PNS influences the SA and AV nodes via the vagus nerve. This influence is produced by release of the neurotransmitter acetylcholine at the vagus nerve endings, which result in the slowing of activity at the SA node and a slowing of the cardiac impulse passing into the ventricles. The SNS has the opposite effect through the release of norepinepherine at the sympathetic nerve endings, thereby increasing the heart rate [5].

PAGE 18

6 HRV reflects the response of the ANS to exogenous or endogenous perturbations. Numerous factors contribute to HRV including blood pressure, temperature, respiration, state of oxygenation, ventilation, biochemical influences of the acid-base balance, and psychological parameters [3]. Continuous changes in sympathetic and parasympathetic neural impulses induce changes in heart rate and cause oscillations around the mean, defined as HRV. By studying heart rate variability, an opportunity exists to study cardiac dynamic behavior and functionality of the ANS. Since the Autonomic Nervous Systems control of heart rate is dynamic and nonlinear, it is necessary to explore viable parameters, which can be used to access the causes of these beat-to-beat fluctuations. This thesis will address time-domain, linear frequency-domain, and nonlinear frequency-domain characteristics of the HRV. Thus the framework for the study has been set, with an attempt to answer each of the following questions: 1. What clinically significant features can be extracted from HRV? 2. What are the developmental changes in HRV? 3. Can we establish criteria to assess vagal tone? 4. Can the premature infant learn or demonstrate a cardiac orienting response, and does HRV change as learning emerges? Developmental Changes Immediately following birth, the heart rate is extremely sensitive to physiological state of the infant. The developmental changes in HRV are hypothesized to indicate a change in the parasympathetic tone. While early fetal life indicates a strong dependence on the SNS, these two systems trend toward an equal balance in the term infant [3]. This thesis assumes a parallel between the fetus and the premature infant. The developmental changes in heart rate variability can be measured through time and frequency domain statistics. Developmental trends of HRV will be measured across subject groups and

PAGE 19

7 gestational age. Time and frequency domain indexes will be used to access maturation of the neonates ANS, or ability to regulate homeostasis. Vagal Tone and Homeostasis Cardiac vagal tone is a construct that describes the functional relationship between the brainstem and the heart. Cardiac responses to brief low intensity stimuli should be deceleratory, indicating receptiveness to incoming stimuli. Cardiac responses to sustained stimuli should be acceleratory, indicating attempts to shut it out [8]. Vagal tone reflects individual differences in degrees of influence of the vagus on the heart rate. High vagal tone is associated with the ability to attend selectively to stimuli and maintain attention during a task, the ability to maintain homeostatic integrity, and infers healthier ANS functioning over the long-term. Homeostasis is primarily maintained by the PNS and provides a physiological framework for the development of complex behaviors. Therefore infants with a high parasympathetic tone are more efficient regulators of homeostasis than those dominated by sympathetic responses [6]. The PNS is primarily responsible with coordinating the life-support process, which intrinsically provides maximum growth and restoration of the bodys tissues and organs [9]. More effective regulation of homeostatis may be an important factor underlying the relationship between high resting vagal tone and favorable neurobehavioral development [6]. Moreover, vagal stimulation controls life-support processes including sucking, swallowing, thermoregulation, and vocalization [9]. The search for an accurate measurement of parasympathetic and sympathetic tone has only begun to express its value. This thesis is anticipated to contribute to the comprehension of these physiological parameters.

PAGE 20

8 Cardiac Orienting Response The existence of a cardiac orienting response (COR) suggests that HRV includes a neurongenic response. The deceleratory parasympathetic component is provided by the bi-directional vagus nerve, which includes both efferent and afferent fibers. Efferent fibers originating in the brain stem terminate on the SA node. Vagal afferent fibers originate throughout the heart and project to the nucleus tractus solitarius [10]. These fibers serve as a continuous response and feedback mechanism between the heart and the brain, facilitating regulation of cardiac function. Porges et al. suggest that the time course of the response, the effects of neural blockades, and studies with clinical populations support this theory. As noted by Porges the heart rate deceleration associated with the cardiac orienting response is rapid and usually returns rapidly to baseline. Also, the latency characteristics of the cardiac orienting response are similar to the opto-vagal, vaso-vagal, baroreceptor-vagal and chemoreceptor-vagal reflexes [8]. Since short latency heart rate reactivity is mediated by the vagus nerve, the magnitude of the cardiac orienting response may be an index of vagal regulation. Respiratory Sinus Arrhythmia Repiratory sinus arrhythmia (RSA) has long been used to estimate vagal tone because of its association with PNS. RSA results from increases in vagal efference during exhalation, which decelerates heart rate, and decreases in vagal efference during inhalation, which accelerate heart rate. Recently, the accuracy of the correlation between RSA and vagal tone has been questioned because it only partially accounts for the beat-to-beat fluctuations of the heart [10]. This thesis will address RSA as a possible criterion to extract vagal tone, but due to the nature of the study will not pharmacologically prove

PAGE 21

9 the efficiency of this technique. However, since RSA accounts for a significant amount of variability in heart rate, it is impossible to completely ignore this aspect.

PAGE 22

CHAPTER 3 PRELIMINARY RESULTS AND STUDY PROCEDURES Preliminary Findings In a preliminary study by the principal investigator [7], cardiac activities during two consecutive quiet periods of 16 healthy fetuses were measured. The exploratory phase used a repeated measures design to follow the normative changes in fetal HRV between 28-34 weeks. Changes in HRV were explained by normalized power shifts in the Fast Fourier Transform (FFT) frequency spectrum. A multi-group pretest-posttest experimental design determined whether 28-34 week-old fetuses would become familiar with a rhyme, recited by the maternal parent. Developmental changes in the HRV frequency spectrum occurred in a nonlinear fashion and individual differences were apparent. The total power diminished significantly with increasing gestational age. Furthermore, the study provided conclusive evidence that the 28-34 week old fetus was capable of responding to a nursery rhyme their mother recited aloud. The response to the nursery rhyme was described by a shift in the power of HRV frequency spectrum to the high frequency bands, which is synonymous with an increase in parasympathetic activity. This conclusion parallels all hypothesized learning capabilities of the premature infant. The testing sessions for the preliminary study utilized a recording of a rhyme spoken by an unfamiliar female, which insured the fetuses reactions would be elicited by familiarity with the acoustic properties of the rhyme. Once learning was detected, ongoing interactions between learning, movement and HRV were detected. A 10

PAGE 23

11 nonparametric signed test was used to compare differences across time. The preliminary study revealed a significant downward trend in fetal movement and an upward trend in a high (.17-.40 Hz) frequency once a cardiac orienting response or learning was identified. Since the subject size was small, there exists a need for a more expanded and complete study before any conclusions can be drawn. However, the findings of the fetal study provide the framework for this expanded research, conducted in collaboration with the PI of the preliminary study. Proposed Study A convenience-based sample of 28 low-risk premature newborns admitted to the NICU at Shands Teaching Hospital are recruited, in blocks of four, and randomly assigned to one of two groups. The target age for admission into the study is 27-28 weeks post-conceptional age. Group 1 will begin exposure to the nursery rhyme at 28 weeks and those in Group 2 will begin exposure at 32 weeks. Staggering the groups shall assist in determining whether an orienting response was a function of learning or simply a reflexive response to presentation of the rhyme. All subjects will be excluded from analysis if they are subject to one or more of the following: An abnormal head ultrasound A sensorineural hearing loss A confirmed prenatally transmitted virus/bacterial infection Cardiac abnormalities Table 3-1. Organization of Recitation and Test Sessions Group 28 weeks 29 weeks 30 weeks 31 weeks 32 weeks 33 weeks 34 weeks 1 Recite Test Recite Test Recite Test Recite Test Recite Test Recite Test Recite Test 2 Test Only Test Only Test Only Test Only Recite Test Recite Test Recite Test

PAGE 24

12 Recitation Sessions The mother of the infant will complete a recording of a nursery rhyme, identical to that presented in the preliminary study. The nursery rhyme, taking approximately 15 seconds to recite, will be played 3 consecutive times to provide the infant with 45s of auditory stimulation. The maternal recitation sessions will occur once in the morning and again in the evening, each day; Group 1 will begin at 28 weeks and Group 2 at 32 weeks of age. The purpose of the recitation session is to introduce the neonate to the rhyme using a familiar voice. The nursery rhyme will be presented once the premature infant is judged to be in a state of active sleep and at least fifteen minutes following a feeding [11]. Between 28-34 weeks post-conceptional age active sleep accounts for appproximately 70% of the entire sleep-wake cycle [12]. Learning in the normal newborn, can occur during periods of active sleep [13,14] and have occurred in the fetus during quiet periods of their quiet-activity cycle [7]. Criteria established by Thoman and Whitney [15] and Holditch-Davis [12] will be used to detect active sleep. The subject is in an active sleep state when: Eyes closed Respiratory rate under 25 breaths per minute Limited movement of extremities Subjects will be monitored at the same time each week because of potential circadian influences on heart rate patterns and movement [16]. A hospital log of recitation times and the time since the neonates last meal will be kept because variations related to unexpected changes in hospital routines are expected. These recitation sessions will be completed entirely separate from the testing session described below.

PAGE 25

13 Test Sessions Each group will undergo a 480 second ECG recording during each test session. Once more, the criteria established by Thoman and Whitney [15] and Holditch-Davis [12] will be used to detect active sleep before the test session begins. The ECG recording will be divided into two stages. The first stage of this study incorporates an exploratory, two-group pretest-posttest subject set which will be used to methodologically replicate the findings from the preliminary study in the premature newborn during a similar developmental time period (28-34 weeks post-conceptional age). The exploratory design will follow specific changes in HRV, measured by time series analysis, power spectral analyses, and time dependant spectral analysis over 28-34 weeks post-conceptional age. A 240-s window separated from the delivery of the auditory stimulus and commencement of the second phase will be used to represent developmental changes in HRV. It is hypothesized that developmental changes in HRV will occur in a decreasing, nonlinear fashion over 28 to 34 weeks gestational age. Two interactions (gestational age vs. gender and gestational age vs. group assignment) will be included in the mixed general linear model. The mixed model allows for the individual random effects when estimating fixed population effects [17]. During a second period of confirmed active sleep, a two-group pretest-posttest experimental design will be used to document when a cardiac orienting response to a presented nursery rhyme first occurs, and how movement and HRV change once it is detected. Again, a nursery rhyme identical to that presented in the preliminary study will be used. An unfamiliar, English-speaking female performs the recitation of the nursery rhyme used in these test sessions. To avoid additional exposure to this unfamiliar voice, this person will not be in attendance for presentations of the rhyme to the infant. The

PAGE 26

14 nursery rhyme takes approximately 15 seconds to recite, and will be repeated 3 consecutive times to provide the 45s window of auditory stimulation. The testing sessions using the unfamiliar recording of the nursery rhyme will occur once a week for both groups from 28-34 weeks post-conceptional age. Figure 3-1. Time-line for the windowing of test data. This second stage will be separated into three 45s windows, in which statistical analysis of heart rate variability will be individually analyzed. The first window will be used to detect the baseline state of the infant before the presence of any auditory stimulation. The second window, immediately following the first, chronologically corresponds to the recitation of the nursery rhyme. This window will be used to observe time and frequency domain changes in HRV, which may reflect the hypothesized COR. A third consecutive window will be used to examine how the neonate responds after the auditory stimulation has been removed from the environment. The third window will be used to verify that an observed COR response is not merely a result of coincidental changes in HRV. In addition, a count of movements will be recorded onto the hospital log during each test session. Our hypotheses pertaining to a COR suggest, (a) over 28-34

PAGE 27

15 weeks of gestational age, it will take 2-5 weeks of recitation to detect and maintain a cardiac orienting response in each subject and (b) once a cardiac orienting response has been observed, subsequent testing session will demonstrate a trend of movement counts and specific HRV characteristics.

PAGE 28

CHAPTER 4 TESTING The rhymes will be played over a 2.5-inch Radio Shack speaker positioned 20 cm from the infants ear. Both versions of the rhyme (maternal and unfamiliar) will be delivered at 55 dB ( + 5 dB) as measured by a Bruel & Kjaer sound level meter (2239) using A-weighted sound pressure levels [18]. This decibel level is just below the normal levels of conversational speech (58-60 dB) and is significantly lower than the background sound levels reported in the literature (70-80 dB + 20 dB) for NICUs [19,20]. The decision to present both versions of the rhyme (maternal and unfamiliar) at 55 dB ( + 5 dB) was based on preliminary findings in the premature fetus where a sound level of 75 dB was used [7]. This decibel level corrects for a 20-decibel attenuation due to passage of sound through the uterine wall [21], bringing the sound level down to approximately 55 dB or to just below the level of normal conversational speech (58-60 dB). The study will record decibel levels at the infants ear prior to presentation of the maternal and unfamiliar rhyme. In the preliminary study, this sound level (75 dB) did not elicit a cardiac deceleration until a history of recitation of the nursery rhyme had occurred [7]. An RS232 interface from the Agilent Neonatal CMS 2001 neonatal monitor to an IBM compatible computer (Dell Inspiron 8100 Laptop) was constructed to obtain the ECG signal and respiratory rates. The data acquisition program is designed to run in DOS and may be activated from the Windows Desktop of Dell Inspiron 8100 Laptop by double clicking on the Neonatal Data Acquisition icon. The C program, originally written by Hewlett Packard (HP), was modified from an earlier version to in order to 16

PAGE 29

17 record the ECG and respiratory rate. The HP program was initially written to demonstrate that information from their monitors could be transferred and displayed on peripheral devices through RS-232 interfacing. The sample program was only available in a DOS format. Although this format is not optimally user friendly, an extensive user interface was not necessary to transfer the required signals into readable text files. Furthermore, HP uses a complex communication protocol, which makes it desirable to avoid writing an original program. The original program was capable of display the ECG signal on a peripheral monitor, which implies that the ECG information was already being transferred. The program modification involved the following: Requesting the respiratory rate signal and displaying the information on the screen. Creating a user interface to request the patient number. Transferring the requested information to a readable text file. Before data acquisition can begin, the Agilent monitor must be set to a baud rate of 19200 and the transition must be set to a high to low trigger. If the baud rate or transition is not correctly set, the program will time-out after eight seconds. When the data acquisition program is started the user enters the patient number and is asked to confirm this number before the recording process begins. The program passes a text string, which signifies the rhyme state and allows the user to mark the data when the rhyme is presented and upon completion. The three measurements: ECG, respiratory rate, and the rhyme state are recorded in an ASCII text file labeled with the month, date, hour (in 24h format), and minute (MMDDHHmm.txt) the program is initialized. ASCII was the chosen format for data storage to permit the user to utilize other analysis programs (i.e. Microsoft EXCEL). The parameters are stored in the format directly readable as a MATLAB matrix and can later be separated into three separate vectors. Preceding the

PAGE 30

18 matrix is an ASCII timestamp, which is recorded once the program has finished the initialization process (DDHHmmss), and appended by another timestamp when the program is terminated. These time stamps allow the user to verify the length of the recording. The program is designed to work with an ECG/Resp module, which will transfer both the ECG and the respiratory rate signal to the Dell Inspiron 8100 Laptop. However, in the absence of the ECG/Resp module it is possible to use only the singular ECG module. This will not record the respiratory signal, but will still record the ECG signal. The first column recorded in the ASCII text file represents the unsigned 12-bit digital values of the ECG signal. The second column contains the information regarding the rhyme state. The final column contains the respiratory rate. The resulting format of the text file is illustrated in the figure below. Figure 4-1. Example of text file generated from Data Acquisition Program

PAGE 31

19 All columns are recorded at a sampling rate of 500Hz, which was preserved from the HP demo program as the rate information was sent to the screen. This sampling rate, much higher than the 200Hz often described in literature, was preserved to prevent jagged QRS complexes. The column corresponding to the state of the rhyme is initialized to , until the r key is pressed and the value changes to . The value remains at until the d key is pressed, which signifies the end of the presentation and returns the value to . When all desired data has been recorded, the user presses the q key twice to exit the program and return to the Windows desktop. Figure 4-2. The setup for the test sessions.

PAGE 32

20 Figure 4-2 illustrates the equipment setup during the test sessions. A male 25-pin RS232 port in the back of the Agilent monitor is interfaced with the female 9-pin RS232 port Dell Inspiron 8100 Laptop. The RS232 cable provides means of serial communication between the Agilent monitor and the Dell Laptop. The Agilent Neonatal CMS 2001 monitor continually monitors the infants vital signs, with a standard 3-lead ECG setup. The speakers, connected to the CD CopyWriter Live, are placed 20cm from the infants ear. This setup is replicated for every test session, regardless of the subjects group number.

PAGE 33

CHAPTER 5 QRS DETECTION Once all data from the testing session is acquired the focus shifts to the data analysis. The first goal is to perform time-series operations, yielding discernable characteristics of the HRV. The HRV is basically controlled by the effect of the vagal and sympathetic influences of the sino-atrial node. Therefore, it seems the best representation of the HRV variability would be expressed as a function of the time intervals between consecutive P waves. However, due to the inconsistent positioning and suboptimal attachment of the infants ECG leads, P waves characteristically have a low signal-to-noise (SNR) ratio [22]. Therefore, in the absence of cardiac arrithymias, addressed in the exclusion criteria, it seems more reliable to replace the detection of PP intervals (PPi), with RRi. While this may include an extra variable in the analysis of the HRV, it seems trivial compared to the accuracy of the detection of PPi. It is our assumption that variations in the PR intervals will be equal or less than error of detection of the PPi. When considering an error-percentage of P-wave detection greater than the standard deviation of the PR interval and greater than the percent error in R-wave detection, the replacement of PPi with RRi becomes trivial. The occurrence of the following cardiac arrhythmias causes this assumption to fail [23]: ATRIAL FLUTTER the atrial flutter waves, known as F waves, are larger than normal P waves and they have a saw-toothed waveform. Not every atrial flutter wave results in a QRS complex because the AV node acts as a filter. Some flutter waves reach the AV node when it is refractory and thus are not propagated to the ventricles. The ventricular rate is usually regular but slower than the atrial rate. 21

PAGE 34

22 ATRIAL FIBRILLATION Atrial fibrillation occurs when the atria depolarize repeatedly and in an irregular uncontrolled manner. As a result, there is no concerted contraction of the atria. No P-waves are observed in the EKG due to the chaotic atrial depolarization. The chaotic atrial depolarization waves penetrate the AV node in an irregular manner, resulting in irregular ventricular contractions. The QRS complexes have normal shape, due to normal ventricular conduction. However, the RR intervals vary from beat to beat. VENTRICULAR TACHYCARDIA Ventricular tachycardia occurs when electrical impulses originating either from the ventricles cause rapid ventricular depolarization. Since the impulse originates from the ventricles, the QRS complexes are wide and bizarre. Ventricular impulses can be sometimes conducted backwards to the atria, in which case, P-waves may be inverted. Otherwise, regular normal P waves may be present but not associated with QRS complexes. The RR intervals are usually regular. THIRD DEGREE AV BLOCK Atrial rate is usually normal and the ventricular rate is usually low, with the atrial rate always faster than the ventricular rate. P waves are normal with constant P-P intervals, but not associated with the QRS complexes. QRS may be normal or widened depending on where the escape pacemaker is located in the conduction system. The result is unrelated atrial and ventricular activities due to the complete blocking of the atrial impulses to the ventricles. Subjects whose electrocardiogram display any of the arrhythmias listed above are removed from analysis on the basis of violating the exclusion criteria. In conclusion, all ECG time-domain analysis of the HRV will consider RRi rather than the ideal analysis of the PPi for the remainder of the thesis. The first step in the analysis of the HRV is to extract the RRi from the recorded ECG waveforms. The ASCII text formatted matrix addressed in the data acquisition chapter is read into MATLAB. The result is an N by 3 matrix, with N being the number of samples. This matrix is then separated into 3 vectors of length N. Vector one contains the ECG information; vector two contains the respiratory rate; and vector three contains information regarded the state of the rhyme. Only vector one expresses useful information for analysis of RRi. Figure 5-1 describes the transform of the raw ECG signal to a vector containing RRi.

PAGE 35

23 Figure 5-1. QRS detection algorithm Noise/Baseline Wander The difficulty in accurate R-wave or QRS complex detection is directly related to the common noise characteristics existing in a raw ECG signal. Noise sources include, but are not limited to: electrode motion, muscle movement, power-line interference, baseline drift, and T-wave interference. Using a method similar to that described by Jiapu Pan [24] we will attempt to improve the SNR of the raw ECG signal and detect the QRS complexes. This process is split into two linear steps and one nonlinear transformation. The first linear process requires filtering the signal with a bandpass filter, which is split into a low-pass and high-pass filter. A low-pass filter with cutoff frequency of approximately 11Hz is first applied the raw ECG signal to remove the low-frequency baseline drift. In addition, the filtered ECG signal is sent through a high-pass filter to remove muscle movement, powerline interference, and T-wave interference. The cutoff frequency of the high-pass filter is approximately 5Hz. This yields a filtered ECG

PAGE 36

24 signal (fECG) containing frequencies of 5-11Hz, which has been described to preserve maximum information regarding the QRS complexes [24]. Low-pass filter: 212611zzzH High-pass filter: 132161321zzzzH Filtering with the Pan-Tompkins method described above, yielded only minor improvements to the raw ECG signal. Problems with the above technique arise as a result of the electrode placement differing from subject to subject. In some case, the Pam-Tompkins algorithm removed substantial frequency components of the QRS complex, thereby deteriorating the quality of the signal. Therefore, we need to replace the low and high pass filters described above with a notch and bandpass filter. The nature of the ECG signal calls for the use of an IIR filter. We will explore the Butterworth and Chebyshev filters. Both are high order filter designs, which can be realized by using simple first order stages cascaded together to achieve the desired order, passband response, and cut-off frequency. The Butterworth filter yields no passband ripple and a roll-off of -20dB per pole. However, the flatness of the passband response comes at the expense of roll-off. The Chebyshev filter displays a much steeper roll-off, but is characterized by a significant passband ripple. The Chebyshev filter thereby optimizes roll-off at the expense of passband ripple. Due to the nature of the ECG signal the roll-off frequency of the passband is not as important as the gain in the passband. Therefore, we will use Butterworth filters for all filtering of the raw ECG signal.

PAGE 37

25 Since 60Hz powerline interference is common among electrical signals, we need to construct a notch filter to remove this noise. The notch filter used is a 5 th order Butterworth filter with center frequency at 60Hz. The magnitude and phase of this filter is shown in Figure 5-2. Figure 5-2. Magnitude and phase response of the 60Hz notch filter used to filter the raw ECG signal. The phase of this filter is essentially linear over the region of attenuation. After the raw ECG signal is passed through the notch filter, it is passed through an 8 th order bandpass filter serving to remove baseline wander and high frequency noise; thereby improving the SNR of the raw ECG signal. Both the notch and bandpass filters are implemented using Butterworth filters. As illustrated in the Figure 5-3 below, the phase over the passband region is approximately linear. Keeping the phase of the filter linear over the passband prevents phase distortion in the resulting filtered ECG waveform.

PAGE 38

26 Figure 5-3. Magnitude and phase response of the bandpass filter used to filter the raw ECG signal. During the recording phase, we obtained ECG recordings that included significant periods where the signal underwent high and low voltage clipping. The clipping resulted in epochs of flat line, thereby making it impossible to detect subsequent QRS complexes. For HRV analysis, a continuous ECG signal is needed. The undesired clipping resulted in extensive signal losses that required considerable epochs of QRS interpolation. It was later discovered that this problem could be overcome by setting the ECG module to filter the signal before it was relayed to the laptop for text conversion. In addition, incorrect electrode placement proved the monitors default selection of Lead II to be an errant assumption. Lead selection was then delegated to the user before the acquisition program was initialized. After realizing the need to select the filter option and correct lead, we were able to attain signals that needed little or no secondary filtering and was not necessary to bandpass filter the recordings. Powerline interference during the passage

PAGE 39

27 from the monitor to the laptop was not completely alleviated and the 60Hz notch filter, described above, was still applied. When the detection phase demonstrated a need for secondary filtering, the bandpass Butterworth Filter, described above, was applied to the signal to conclude the filtering stage of the detection algorithm. Slope Detection In parallel with the Pan-Tompkins algorithm, after filtering, the fECG signal is sent through a differentiator yielding QRS complex slope information; this method results in a differentiated version (dfECG) of the fECG. In accordance with Pan [24], we will use a 5-point transfer function to approximate the derivative. 21122281zzzzTzH where T is the sampling period. This transfer function is roughly linear in the region between dc and 100Hz. Therefore it provides a quick approximation of the ideal derivative over the region of interest. Figure 5-4. Magnitude and phase response of the differentiator function

PAGE 40

28 Following this step, the dfECG signal is passed through a nonlinear point-by-point squaring operator yielding the sdfECG signal: 2nTxnTy This operation makes all data points positive and supplies a nonlinear amplification of the signal at the higher frequencies. Finally, the sdfECG is smoothed using a ten point averaging function, which produces a signal containing the power information of the ECG waveforms slope, pECG. This reduces the presence of local minima in the sdfECG signal. From this point, the method used for detection of QRS complexes deviates from that used by Pan [24]. Since it is not necessary to perform QRS detection in an online fashion, we have the luxury of determining thresholds using the entire information contained in pECG signal. The threshold limit of the QRS slope detection is set in the following fashion: 2/max22dfECGmeandfECG The selection of the current threshold () allows for detection of poorly defined QRS complexes, while distinguishing the R-waves from occurrences of P and T-waves. The threshold is then adjusted throughout the detection process. The following equations illustrates how the threshold is updated: 2/):():(max22winiidfECGmeanwiniidfECG where threshold parameter () is updated in accordance with the current sample (i) over a user-defined window (win)

PAGE 41

29 where represents is the threshold difference and includes a memory constant () to reduce high frequency noise interference during the update Once this threshold has been set, it is now necessary to find maximum values during each interval that the signal spends above threshold. The detection algorithm searches the pECG until it reaches a y-value above the threshold. An array is constructed, storing the waveform from this point until the y-value falls below the threshold. The x-value corresponding to the maximum y-value inside this array is stored as a possible QRS complex. mnnnxQRS,,2,1,max where y[n] and y[m] are above the threshold, but y[n-1] and y[m+1] fall below the threshold. Figure 5-5. Visualization of QRS peak detection

PAGE 42

30 These maximums, or peaks, define the occurrences of possible QRS complexes. If a QRS complex is detected outside the expected range given below, *)()()(RRimeankQRSRRimean 5. RR(0) = 0.375s (160bpm) 6. 1<<2 7. is the detected QRS complex )(kQRS the algorithm reverts back to the previous QRS complex, adjusts the threshold as follows, threshold threshold is a user definable parameter (default = 1.5) and searches for a new QRS complex. This searchback technique plays a crucial role in preventing high frequency noise contamination, P-waves, and T-waves from being marked as a QRS complex. If a complex is not found within the expected range after 5 searchbacks, a complex is interpolated as an average of the previous two intervals. 2)2()1()( kQRSkQRSkQRS Subsequent detections are stored in a vector of increasing length until the algorithm has operated on the entire pECG signal. In addition, the interpolated detections are stored in a separate vector for further analysis. Units of these vectors are stored according to sample numbers. The vectors are then divided by the sampling rate for unit conversion to the time-domain. Finally, another vector is constructed, which represents a time-based representation of the RRi in the following fashion: 1,,23,12 NQRSNQRSQRSQRSQRSQRSRRi

PAGE 43

31 where N = total number of QRS detections Figure 5-6 illustrates the final results of the QRS detection algorithm where the QRS complexes are marked in red. This figure also displays a linear interpolation of the RRi waveform corresponding to the detected beats. Figure 5-6. Example of detected QRS complexes and beat intervals The detection is offset from the peaks of the QRS complex due to the intrinsic phase delays of the filtering process. As a result, the delays are consistent throughout each signal and fall out after computing the RRi vector, as follows: 111 NQRSNQRSNQRSNQRSNQRSNQRSRRi

PAGE 44

32 Outliers Outliers are ill-defined byproducts of data acquisition systems. There is no rigorous definition of an outlier; it is only generally conceived that selectively eliminating inconvenient data points from analysis creates a more robust dataset. Outliers are acknowledged to skew sample means, but tend to have less affect on the median than the mean. For the purpose of this thesis, it is necessary to form a working definition of an outlier. An outlier is a data point that is an "unusual" observation or an extreme value in the data set. The lack of a universally accepted definition of an outlier inherently causes a deficiency of objective mathematical processes for outlier recognition. Adding to the complication of outlier detection, RRi datasets are time dependant vectors, rather than independent samples. Therefore, it is essential to add to our working definition of an outlier. A time dependant outlier is a data point that is an "unusual" observation or an extreme value in the data set, which is not part of a trend in time. Again, the definition of a trend is subjective and needs to be defined for the purposes of this paper. A trend is two or more consecutive data points, which move in the same general direction within a given statistical range. These definitions will be used to construct an algorithm to remove time dependant outliers from the RRi datasets. False detections and missed detections result in possible outliers in these datasets. The presence of outliers in the RRi dataset adversely effect both time and frequency domain analysis of HRV. Possible skewed time domain representations of HRV include the mean and standard deviation of RRi windows. In addition, these outliers produce ghost powers in the high frequency range of the

PAGE 45

33 spectrum. Not only does this increase the power in the high frequency bands, it skews the power in the low frequency bands. Low frequency bands are of the foremost importance during analysis of HRV in the frequency domain. This will become more evident through discussion in Chapter 6. Now that the importance of removing outliers has been established, we must construct a mathematically sound algorithm for their removal. Three techniques are explored as possible strategies for outlier detection: direct RRi analysis, slope analysis, and differential analysis. For each of the aforementioned techniques, we will use standard deviation as the exclusion criterion. Since literature is not conclusive for the distribution of RRi datasets, the exclusion criterion is based on normal distribution of the data. The figures below illustrate the RRi distribution and the normal distribution. Visually, the normal distribution seems an accurate estimation of the RRi dataset. We will discuss the validity of this assumption further in Chapter 6. Figure 5-7. Histogram of RRi distribution from one subject.

PAGE 46

34 Figure 5-8. Standard deviations of the Normal Distribution In a normal distribution, one standard deviation away from the mean in either direction on the horizontal axis accounts for around 68 percent of the values in the dataset. Two standard deviations away from the mean includes for roughly 95 percent of the data. Finally, three standard deviations account for about 99 percent of the data [25]. Therefore, we will assume data outside of three standard deviations from the mean to be statistically irrelevant and tag them as outliers unless they are part of a trend. Direct RRi Analysis The simplest method of removing outliers would be to examine the RRi dataset directly and remove data that fall three standard deviations above or below the mean. The following equation is used for outlier detection: RRistdRRimeannRRioutliers*3 for n RRilength:1

PAGE 47

35 This method inadequately detects outliers and is not optimal to detect a trend, which moves beyond the bounds of the acceptable range. In addition, the direct RRi analysis method does not account for consecutive false or missed detection that exhibit a bouncing effect around the actual QRS complex. Figure 5-9. Illustrates the bouncing effect and upper and lower bound symmetry problems with using the direct RRi analysis method for tagging outliers. The figure above demonstrates problems, which arise from direct RRi analysis method for tagging outliers. The green lines represent the bounds for acceptable datapoints. The bouncing effect is clearly present in this example. According to the exclusion criteria, one of the datapoints would be tagged as an outlier, while the other point would pass through the algorithm without being tagged. This figure also unveils weaknesses with the computation of the mean. When the heart rate deviates to far from

PAGE 48

36 the mean (marked by the black line), situations exist where the upper and lower bound are not symmetrical about the data over a given window. Given all the abovementioned weaknesses, it is necessary to explore a more robust technique to mark the presences of an outlier. Slope Analysis In order to correct for the bouncing effect described above, slope information could be used to isolate these occurrences and tag outliers. First, compute the derivative of the RRi dataset, as follows: tRRiRRi Then, tagging outliers using the same detection method as in the direct analysis: RRistdRRimeannRRioutliers*3 for n RRilength:1 This algorithm adequately accounts for the removal of the bouncing effect, but we will demonstrate that it is impossible to exclude trends from being tagged. Figure 5-10 is an example of trend which seems to begin abruptly, but is sustained over subsequent intervals. The slope algorithm solves the problems in accordance with the mean and bounce effect, but introduces troubles in excluding trends from being tagged as outliers. Over the course of this trend, only those datapoints whose slope exceeded the bound of acceptance (marked by the green line) would be tagged. No mathematically simple method exists to overcome the flaws in this algorithm. Merely modifying this algorithm to remove the outliers and loop through the process again would serve to completely eliminate the trend. Therefore, for a second time it is necessary to search for a more robust algorithm to tag outliers.

PAGE 49

37 Figure 5-10. Illustrates an RRi trend that would not be preserved using the slope analysis method for tagging outliers. Differential Analysis To this point, three problems in the outlier detection process have been identified: mean calculations that deteriorate the symmetry of the acceptance bound, the bouncing effect, and the exclusion of trends. We introduce the differential analysis method as a means, which circumvents all three of these problems. First, a difference vector is constructed as follows: 1)( nxnxnRRidiff for n, where m:2 RRilengthm

PAGE 50

38 1)1(RRiRRimeanRRidiff The end product is a vector, which is the same length as RRi. The resulting mean of the difference vector is close to zero, which further implies the data is predominately normally distributed. The difference vector also alleviates the first problem mentioned above. Then, a vector is created using the same exclusion criterion discussed in the previous two subsections. However, using this difference vector alone to tag outliers outside the acceptable standard deviation range is not sufficient. A single outlier in a dataset would produce two datapoints in this difference vector that would be tagged as outliers (n and n+1). Being that it is not necessary to perform the outlier removal real-time, we are able to create an additional vector, as follows: 1)(nxnxnRRidiff for n, where 1:1m RRilengthm mRRiRRimeanmRRidiff)( The same exclusion criterion is used to create a vector containing the outliers. As a result of this computation, a single outlier in the dataset would produce two datapoints tagged as outliers. Since taking the difference of RR intervals backwards through time generates this vector, the two resulting datapoints tagged as outliers would be n and n-1. Now, we compare the two difference vectors and note only those values for n, which occur in both. These are the actual RRi datapoints that are separated from the previous and subsequent datapoint by three standard deviations above or below the differential mean. Figure 5-11 is a section of RRi data with trends that would normal be removed using the Direct RRi Analysis method, described above, and a bounce effect.

PAGE 51

39 Figure 5-11. Outliers that have been tagged, using the differential method, are marked by a green dot. Shows the efficiency of the method to preserve trend and remove bouncing effects. In figure above, the points that were tagged as outliers are denoted by green dots. It is now evident that we have constructed a robust method to find and remove outliers, as defined above. Interpolation The RRi vector described above is sufficient for time domain analysis of the HRV. The RRi vector can also be converted to an instantaneous representation of the heart rate (iBPM). However, there is a problem with using this vector to represent the HRV in the frequency domain. When plotting the RRi on a time scale, the x-axis must include a cumulative sum of all the preceding RRi plotted against the RRi or iBPM. This results in

PAGE 52

40 an unequally sampled representation of the HRV. In order to apply a meaningful power spectral representation of the HRV, the data must be evenly sampled at a known frequency. Many papers using spectral analysis fail to address this topic and there are no universal methods to solve this problem. Resolving this issue of uneven sampling is essential for spectral analysis. Without addressing this topic, may forms of spectral analysis applied to the RRi would produce meaningless results. The choice of an interpolation method is very important, as it can significantly compromise the integrity of the data. Therefore, we will explore four methods of interpolating between successive data points of RRi: Linear Interpolation (LI) LaGrange Polynomial Interpolation (LPI) Cubic Spline (CS) Cubic Interpolation (CI) Figure 5-12. LaGrange Polynomial Interpolation

PAGE 53

41 Figure 5-13. Linear Interpolation Figure 5-14. Cubic Spline Interpolation

PAGE 54

42 Figure 5-15. Cubic Interpolation The performance of the interpolation techniques is shown applied to a sample RRi dataset marked by Xs in the figures above. LI is continuous, but the slope changes at the vertex points. LPI has an adjustable order of polynomial and reduces to LI when the order equals one. As seen above, the LPI algorithm does not perform well at the endpoints. CSI produces the smoothest results of all the interpolation methods, but performs poorly if the data is highly non-uniform. CI results in both the interpolated data and its derivative being continuous. The CI technique seems the most realistic interpolation method because it does not force a sinusoidal interpolation. Therefore, we will use CI to interpolate and resample RRi in preparation for spectral analysis. It is necessary to resample the RRi at a frequency where the Nyquist rate is above the frequency range of interest for spectral analysis, discussed further in Chapter 6. Also in

PAGE 55

43 Chapter 6, spectral analysis techniques will be introduced that do not require resampling of the RRi waveform prior examining HRV in the frequency domain.

PAGE 56

CHAPTER 6 STATISTICAL ANALYSIS Time Based The simplest methods used to evaluate HRV are based in the time domain. Time domain statistics can be divided into those derived from RRi or iHR and those derived from differences in RRi, namely RRi [26]. The following time-domain techniques will be used: Mean RRi (mRRi) Median RRi (medRRi) Standard deviation of RR interval (SDRR) Poincar plots Quadrant analysis First Order Statistics Analysis of mRRi and medRRi yields some useful information about HRV. We will use this information to analyze trends of the mRRi and medRRi with gestational age and across genders. Also, the mRRi for the 45-s prestimulus period will be subtracted from each heartbeat of the 45-s prestimulus and stimulus period. The result will be a string of difference scores. The difference scores will be used to confirm that no significant HRV trending has occurred during the prestimulus period, which may mask or falsely demonstrate a COR. During the stimulus period, the difference scores will be used to detect an orienting response. The mRRi is calculated in the following manner: NnRRimRRiNn1)( 44

PAGE 57

45 for N= number of RRi To find the median, we must first sort the RR intervals by ascending values. Then, apply the following formula to find medRRi: )2/(NRRimedRRisorted for even values of N 2)2/1()2/1( NRRiNRRimedRRisortedsorted for odd values of N SDRR is highly dependant upon the length of the recording [26]. When using this value as a comparison across datasets, we must be careful to uniformly define the length of the recording. Therefore, we have kept all window lengths constant from subject-to-subject and session-to-session. The variance (SDRR squared) is mathematically equal to the total power of the spectrum; therefore SDRR will also be used to analyze trends corresponding to increasing gestational age. SDRR is the primary first order statistical analysis method we will use to discuss HRV in the time domain. SDRR is calculated in the following manner: 1)(12NmRRinRRiSDRRNn where N= number of RRi Poincar Plots In addition to numerical analysis in the time domain, the importance of geometric representations, such as Poincar plots are useful tools for HRV analysis [26,27]. The nonlinear Poincar plot produces a figure with each RRi plotted against the previous RRi.

PAGE 58

46 According to Brennan et al., the plot provides summary information as well as detailed beat-to-beat information on the behavior of the heart. The dispersion of points perpendicular to the line of identity reflects the level of short-term variability [27]. In subjects with normal cardiac function, the geometry of Poincar plots tends to be elliptical in nature. No literature pertaining to analysis of Poincar plots using an ellipse were found, but we will explore an ellipse fitting method as a possible representation of short-term and long-variability. MATLAB code written by Marcos Duarte [28] will be used to fit an ellipse to the data. Duarte calculates the major and minor axis of the ellipse using principal component analysis (PCA). The method results in 85.35% of the datapoints lying inside of the ellipse. The datapoints that fall outside the ellipse can be described by increasing the order of the ellipsoid, but it is only necessary to include the first two eigenvalues when analyzing the short and long-term variability of the dataset. PCA is a multivariate procedure, which rotates the data such that maximum variabilities are projected onto the axes. Essentially through computing the eigenvalues of the correlation matrix, a set of correlated variables are transformed into a set of uncorrelated variables, which are ordered by reducing variability. The uncorrelated variables are linear combinations of the original variables, and the last of these variables can be removed with minimum loss of real data. The largest eigenvalue corresponds to the principal component in the direction of greatest variance; the next largest eigenvalue corresponds to the principal component in the perpendicular direction of next greatest variance. PCA can be viewed as a rotation of the existing axes to new positions in the space defined by the original variables. In this new rotation, there will be no correlation

PAGE 59

47 between the new variables defined by the rotation. The first new variable contains the maximum amount of variation. The second new variable contains the maximum amount of variation unexplained by, and orthogonal to, the first. The result is an ellipsoid in multidimensional space and reduces to an ellipse for the two dimensional Poincar Plot. When a Gaussian random vector has covariance matrix that is not diagonal, the axes of the resulting ellipse are perpendicular to each other, but are not parallel to the coordinate axes. As shown in the Figure 6-1, the principal components of the RR interval Poincar representation result in an ellipse whose axes are rotated versions of the coordinate axes. Therefore, as discussed in Chapter 5, the assumption that a Normal distribution can describe RRi datasets is valid. The minor axis of the ellipse serves as an interpretation of short-term variability; and the major axis serves as an index on long-term variability. Figure 6-1. Poincar plot representation of RRi. Since data points overlap, colors, coordinated with the visible spectrum, are used to indicate magnitude at each point.

PAGE 60

48 Quadrant Analysis Quadrant analysis is a graphical tool used to represent the number of sustained increases, sustained decreases, and alternating changes in RRi. Two vectors are created as follows: 1 nRRinRRinRRi nRRinRRinRR 11 Each difference value was then plotted against the previous difference, resulting in the RRi (n) versus RRi (n+1) plot shown below [4]. The plot displays values distributed over an area defined by four quadrants. Each quadrant indicates the direction of two consecutive changes in interval length. Figure 6-2. Defines the quadrants used in Quadrant Analysis.

PAGE 61

49 We will use this analysis tool to examine sustained increases or decreases of RRi and alternating RRi. The number of points in quadrants A and D measure high frequency variation, or quick changes in RRi. Low frequency heart rate variation is characterized by many sequential increases and decreases in interval length, which is indicated by points in quadrants B and C. If changes in RRi occurred randomly and independently, all quadrants would have an approximately equal number of data points. In addition, if the heart rate tended to increase quickly and decrease slowly, more points would fall in quadrants A and B than in C and D and visa versa. Figure 6-3 is an example of a quadrant analysis plot extract from real RRi data. Figure 6-3. Example of Quadrant Analysis applied to an RRi dataset.

PAGE 62

50 Frequency Based While analytical methods were first restricted to the time domain, they have quickly shifted to the frequency domain. Spectral analysis is now a common representation of HRV. Transforming HRV to the frequency domain provides a means of separating the parasympathetic and sympathetic responses of the ANS. Heart rate fluctuations in the premature will be examined in the .00033-1Hz, while the frequencies below this will be discarded to reduce the influence of slow trends or artifacts. Research has suggested that the power spectral decomposition of the ECG yields frequency bands essential to analysis of the ANS. These frequency bands are very inconsistent throughout the literature referenced for this subject. This thesis will follow the frequency divisions described by Stein et al. [29]: The ultra low frequency band (ULFB) <.0033Hz The very low frequency band (VLFB) 0.0033 0.04Hz The low frequency band (LFB) 0.04 0.15Hz The high frequency band (HFB) 0.15 0.4Hz The total power (TP) .0033 1Hz The low frequency band of the power spectrum is composed of two peaks. The mid-range peak (.09 .15Hz) corresponds to the Mayer waves, which have been related to the baroreceptor reflex [3]. In addition, there exist a second peak between .02 .09Hz, which has been linked to thermoregulatory fluctuations and variations in vasomotor tone. The high-frequency band reflects the parasympathetic response, while the low frequency band is influenced by the PNS and SNS [3]. For each frequency band listed above, we will extract the normalized power over the described range. Also, the power ratio of the

PAGE 63

51 high and low frequency bands has been assumed to reflect the sympathovagal balance [3]. This power ratio is observed as follows: HFPLFPLFPSVratio The SV ratio should provide us with insight into the ANS balance and should trend downward in accordance with increasing gestational age. Lastly, a range just above the high-frequency band contains information regarding the RSA, which is not represented by a well-defined peak, but is usually centered about the .5-.6Hz range. We will explore three methods of power spectral analysis in the following subsections: Fast Fourier Spectral Analysis Lomb-Scargle Spectral Analysis Empirical Mode Decomposition Fast Fourier Spectral Analysis The simplest and most commonly exploited technique for spectral analysis is the Fast Fourier Transform (FFT). The FFT is commonly used because it is computationally simple algorithm for spectral analysis. Fourier analysis decomposes a time-series into global sinusoidal components with fixed amplitudes. As a result, Fourier spectral analysis is limited to systems that are linear and strictly stationary. However, a high number of variables affect the heart rate, which suggests that an RRi signal is highly nonlinear and non-stationary. Most research pertaining to HRV utilizes the FFT to describe to the spectral components of the RRi signal. We will utilize the FFT spectral analysis method to obtain results, which can be compared with past and future research. The FFT is computed using the internal FFT function provided by MATLAB. Since the FFT method is commonly used in signal processing, we will not discuss this

PAGE 64

52 method in detail, rather we will explain how we computed the transform. All RRi windows for the RRi dataset (developmental, prestimulus, stimulus, and poststimulus) are analyzed separately based on a N-point FFT, where N is equal to the total number of point in the intervals. Often FFT windows are padded with zeros or with repeated copies of the window. Neither of these techniques adds any information to the resulting FFT. Padding only serves to increase the resolution of the transformation. Setting N to the length to the window abolishes the need for padding. Each of the aforementioned windows is extracted from the original recording based on cumulative time, and the resulting windows are unevenly sampled signals. The cubic spline interpolation technique, described in Chapter 5, is applied before computing the FFT and resampling the interpolated signal at 4Hz. We chose 4Hz to prevent aliasing above of the Nyquist frequency (2Hz) from falling into the frequency bands of interest. Finally, the resampled signal is passed through the FFT algorithm and the DC component of the FFT transformation vector is removed by dropping the first term. Results from this algorithm can be found in the following chapter. Lomb Spectral Analysis In order to avoid the consequence of interpolation, we explore the Lomb method to estimate the power spectral density (PSD) of unevenly sampled signals. Laguna et al. concluded that for PSD of these unevenly sampled signals the Lomb method is more suitable than FFT estimates requiring linear or cubic interpolation [30]. The Lomb estimate avoids the low pass effect of resampling and avoids aliasing up to the mean Nyquist frequency: mRRNf*21

PAGE 65

53 The Lomb method is based on the minimization of the squared differences between the projection of the signal onto the basis function and the signal under study [30]. ninntbictx)( Minimizing the variance of n (mean squared error), is synonomous to the minimization of Nnnintbictx12 which results in [30] Nnnintbtxkic1*1 Nnnitbk12 In addition, the signal power at index i of the transformation as described by Lomb [31]: 221*)(*icicktbtxiciPNnninx where kicic Setting the basis function to that of the Fourier transform nitfjnietb2 Then, NekNntfjni122 And fcfPiPXx2 Lomb presented a simplification that permits one to solve for each coefficient independently, known as the Lomb normalized periodogram. First, we solve for the mean and variance [31]:

PAGE 66

54 22SDRRmRR x Then, we generate the Lomb normalized periodogram using the angular frequency thi NniniNnininNniniNnininiXttxtxttxtxP122112212sinsincoscos21 where i is defined as, NnniNnniiitt112cos2sinarctan21 MATLAB code written by David Glover is used to compute this Lomb normalized periodogram [32]. We apply this method of spectral analysis to the identical windows used to compute the Fourier PSD; and we will compare the results in the following chapter. Empirical Mode Decomposition Approximations are the key to understanding complex natural phenomena and often there is a significant advantage to these methods. However, science is continuing to develop ways to describing complex system and behaviors. Why approximate the RRi signal as a linear, stationary signal when tools exist to analyze nonstationary, nonlinear systems? The answer is rhetorical because the only reason describes the simplicity of the computation of the FFT. Given all the assumptions needed to compute the FFT, it is necessary to search for a more reliable spectral transformation. We have already observed the Lomb method, which still assumes the signal is linear and stationary. Now,

PAGE 67

55 using a technique developed by Haung et al. [33], this thesis will compare the effectiveness of the Empirical Mode Decomposition (EMD) as an analysis tool of HRV. The goal of EMD is to decompose a time series into superposition of Intrinsic Mode Functions (IMF). EMD uses a shifting process to produce IMFs, which enables complicated data to be reduced into a form with defined instantaneous frequencies. The IMFs has the appearance of a generalized Fourier analysis, but with variable amplitudes and frequencies. Haung et al. describes EMD as the first local and adaptive method in frequency time analysis. Furthermore, Haung suggests that the advantage of EMD and Hilbert spectral analysis is that it gives the best-fit local sine or cosine form to the local data. The frequency resolution for any point is uniformly defined by the local phase method and is especially effective in extracting low frequencies oscillations [33]. Since the HRV is characterized by low frequency oscillations, this method should produce informative frequency domain representation of HRV. Computation of the EMD algorithm is done in MATLAB using modified code originally written by G. Rilling [34]. The EMD algorithm [33] coded by Rilling begins with the sifting process. First, all local extrema of the original signal are identified. The local maxima are then connected via a cubic spline and the procedure is repeated for the local minima, respectively forming the upper and lower envelopes. Then, we find the first component of the sifting process [33], h. 1 1010)(mhhtRRitXh where is the mean of the envelope 1m The shift process is repeated k times, as follows, until is an IMF (). kh1 1c

PAGE 68

56 kkkkhcmhh111111 Then, a residual is found 11ctXr Finally the IMF procedure is repeated n times until the residual, is a trend or constant. nr 11 nnnchr It is then necessary to determine if all the calculated IMF are actually orthogonal to one another, as the theory suggests. Huang suggests using the follow equation as a check for orthogonality [33]: 022gfgffgccccIO for all IMFs In order to visualize the result of EMD, we must find the Hilbert transforms of the IMF components and represent the RRi signal in terms of time, frequency, and amplitude. We compute the Hilbert transforms using the Cauchy principal value integral to find the magnitude, and phase, tan tn and instantaneous frequencies, n [33]. tttXtYttYtXtaetatiYtXtZtdtttctYjjjjjjjtijjjjjjj arctan122 Finally, we can express the RRi dataset as follows

PAGE 69

57 ttitatRRijnjjexp1 Cleary, the equation above represents the generalized Fourier expansion, but the amplitude and instantaneous frequency are variable with respect to time [33]. At this point, we will use the EMD method to provide us with a visual representation of the power spectrum changes over a prestimulus stimulus poststimulus window. In accordance with the description of a COR, detailed in Chapter 2, we should see significant shift in the magnitude of frequency spectrum over this window. Statistical Analysis of Parameters Statistical analysis of the parameters extracted above is performed using a technique for longitudinal data analysis, called mixed general liner models (MixMod) or random regression models [35]. The MixMod model allows the concurrent identification of both group and individual patterns in longitudinal sets with covariates that vary over time, inconsistently timed data, irregularly timed data, and randomly missed values. Given the vulnerable state of the infants in the NICU, we must perform test sessions in accordance with the health status of the infant. These stability considerations can result in the testing schedules varying from subject to subject and session to session. Fortunately, problems arising from this effect are alleviated using the MixMod analysis. In addition, subjects are often transferred or released before reaching 34 weeks GA. The MixMod model allows us to include the data from these subjects, thereby reducing the amount of data, which would otherwise be lost. Strengths of the MixMod model include the acceptance of multiple characteristics and the exploration of patterns across subjects and groups. Using multiple characteristics, MixMod leaves the statistical correlation and inclusive of parameters up to the investigator [35]. All MixMod analysis is performed by

PAGE 70

58 Dr. Kruger using readily available SAS Proc Mixed software. The results of the MixMod analysis can be found in the Clinical Results section of Chapter 7.

PAGE 71

CHAPTER 7 DISCUSSION AND RESULTS Engineering Results We will explore the results of the signal-processing phase of this thesis in accordance with ECG recordings from one subject. The example subject is not intended to prove the clinical significance of the hypothesis described in Chapter 3. For clinical evaluation of aforementioned hypotheses, please refer to the Clinical Results subsection of this chapter. QRS Detection Algorithm The following table is a representation of the performance of the QRS detection algorithm, detailed in Chapter 5. The percent of QRS complexes is computed, as follows, missedectionsectionsectedQRS###detdetdet where the interpolated detections are tagged as missed detections and are stored in a separate array to determine the QRS detection ratio defined above. The interpolations are a result of no detected QRS complexes falling within the user defined expectancy range, described in Chapter 5. Table 7-1 clearly illustrates that on average the algorithm only misses 0.85% of the QRS complexes in the signal. This detection algorithm is a robust system, which can effectively detected QRS complexes in the presence of noise and muscle artifact and produce reliable RRi datasets. Beyond the detection process, we improve quality of the 59

PAGE 72

60 RRi signal through the implementation of the outliers removal algorithm designed in Chapter 5. Table 7-1. Shows the performance of the QRS detection algorithm HRV Plots and Statistics Scientific findings are based on equally important qualitative and quantitative measures. Quantitative methods observe distinguishing characteristics and elemental properties resulting in numerical comparative analyses and statistical analyses. For each recording, an excel file is created, which contains all the pertinent quantitative measures for each window:

PAGE 73

61 mRR medRR SDRR Number of sustained increase and decreases in HR derived through Quadrant Analysis The length, width, and (W/L) ratio of the PCA fit elliptical representation of the Poincar plot ULF, VLF, LF, and HF power of the Lomb PSD TP and of Lomb PSD ratioSV ULF, VLF, LF, and HF power of the Fourier PSD TP and of Fourier PSD ratioSV These quantitative measures are used for MixMod analysis in the Clinical Results section. Qualitative measures are those associated with interpretative approaches, from the investigators point of view, rather than discrete observations. While qualitative methodologies can lead to biased results, there is no doubt that they should be included in medical research. The power of visualization and quick interpretation of data possessed by humans cannot be replaced by quantitative measures. Therefore, for each recording, six plots are created for qualitative analysis: HR with outlier removal Quadrant analysis Poincar Fourier PSD Lomb PSD Hilbert-Haung spectrum of EMD

PAGE 74

62 HR Plot Figure 7-1. HR plots of example RRi data. From top to bottom they represent: the (a) developmental, (b) prestimulus, (c) stimulus, and (d) poststimulus periods. The HR plot shown above illustrates the original RRi (blue) waveform displayed as instantaneous heart rates and the resulting RRi (red) after outliers have been removed. The figure above is divided into four plots representing the four windows of analysis. Plot (b) illustrates the removal of an outlier, which causes a high frequency spike in the signal. Plot (a) demonstrated the algorithm efficiency of preserving trends in RR intervals. Figure 7-1 is used to visualize cardiac accelerations and decelerations, quantitatively described through Quadrant Analysis.

PAGE 75

63 Quadrant Plot The Quadrant plot shown in Figure 7-2, below, is organized in the same manner as Figure 7-1 with each plot representing a particular analysis window. Quadrant Analysis is a more useful tool for quantitative analysis, but the visual representation can provide insight into the amount of high and low frequency variability. The excel file described above contains some useful information regarding this plot that will later be used for MixMod analysis. Figure 7-2. Quadrant Analysis plots of example RRi data. From top to bottom they represent: the (a) developmental, (b) prestimulus, (c) stimulus, and (d) poststimulus periods.

PAGE 76

64 Poincare Plot The Poincare plot is a valuable tool for both quantitative and qualitative analysis. The Poincar method of RRi interpretation is observed to prove normal cardiac function and analytically evaluated by the fitted ellipse to described short and long-term HRV. Figure 7-3. Poincar plots of example RRi data. From top to bottom they represent: the (a) developmental, (b) prestimulus, (c) stimulus, and (d) poststimulus periods. Again the magnitude corresponding of a given point is represented by the basic colors of the visible light spectrum with red being the next to least frequent (black is the least) and violet being the most frequent. Plot (a) represents the dataset with the most long-term variability or low-frequency changes inferred by the length of the major axis of the ellipse. This is expected because plot (a) is a 240s window, while plots (a,b,c) are 45s

PAGE 77

65 windows. We can also see that during the stimulus period the major and minor axis of the ellipse become significantly longer, perhaps signifying an increase the LF power, HF power, and total power of the HRV spectrum. We will next look at the Fourier and Lomb PSD to confirm this hypothesis. Fourier and Lomb Plots The Fourier and Lomb PSD plots essentially describe the same information; therefore we will observe them together. Figure 7-4. Fourier PSD plots of example RRi data. From top to bottom they represent: the (a) developmental, (b) prestimulus, (c) stimulus, and (d) poststimulus periods.

PAGE 78

66 Figure 7-5. Lomb PSD plots of example RRi data. From top to bottom they represent: the (a) developmental, (b) prestimulus, (c) stimulus, and (d) poststimulus periods. First, we will address the observational hypothesis from the Poincar Plot section. Inspecting the Fourier and Lomb stimulus plots, we note that the area under the curve increases from the prestimulus period to the stimulus period, which means there is an increase in total power. Also, power increases in the LF and HF bands correspond to increases in the length of the major and minor axes of the ellipse, respectively. In conclusion, through qualitative analysis, we have shown that the Poincar plot is a time domain representation of the PSD. Now we will compare the PSD representations of the Fourier and Lomb Spectrum. The Fourier and Lomb plots are comparable over the developmental window,

PAGE 79

67 but the Fourier plot is choppier than the Lomb plot over the windows corresponding to the COR. The choppy Fourier plot is a result of the low frequency resolution of the N-point Fourier transformation. Therefore, the frequency resolution is lower for the short windows analyzed in plots b, c, and d. However, the Lomb computation allows the user to set the frequency resolution. For the purpose of obtaining a smooth transformation, we have set the Lomb PSD resolution to 0.001Hz. To increase the frequency resolution of the FFT, we must pad the data before performing the Fourier transform. For reasons discussed in Chapter 6, we have refrained from doing so. The apparent equivalent Lomb and Fourier PSD plots for the developmental window suggest that the Lomb method is at least as successful as the Fourier for PSD estimation. For clinical interpretation, we will use the Lomb Spectrum when computing the cumulative power in the frequency bands of interest. EMD Plots At this point, we use the Hilbert-Haung Spectrum (HHS) only as a qualitative analysis tool. If the clinical findings indicate the usefulness of this time-frequency domain representation of HRV, we will expand to use EMD as a quantitative analysis tool. Figure 7-6 is the HHS of the example RRi dataset used in this section. Black lines separate the prestimulus, stimulus, and poststimulus periods. We can relate the HHS power shift of the stimulus period to those described by the Poincar and PSD assessments of HRV. All four representations are in agreement that during the stimulus period the LF power, HF power, and TP increase. However, the HHS shows us that this increase occurs just after the rhyme begins only to tail-off significantly shortly thereafter. Initially the EMD seems to be a robust method to describe the HRV characteristics of the COR with respect to time. However, since this method has not often been used for

PAGE 80

68 clinical interpretation, the inclusion of the method in the analysis of the hypothesized COR will be left up to the PI. Figure 7-6. Hilbert-Haung spectrum of the COR windows of an example RRi dataset. Clinical Results At the time of submission, we only have four complete datasets (recordings for the 28-34 weeks GA) and one incomplete dataset (recordings for 28-33 weeks GA) that can be used for clinical interpretation. The current sample size (n=5) is well below the minimum sample size (n=28) described in the design of this study [7]. Therefore, we will not be able to perform MixMod analysis of the data at this time. Instead, we will use a repeated measures ANOVA analysis of the PSD. Repeated measure ANOVA allows us to compare the variance caused by the independent variables to a more accurate error term, which includes the variance effects of individual subjects. Thus, this method

PAGE 81

69 requires fewer participants to yield an adequate interpretation of the data than the MixMod method. Early analysis based on the Poincar plots, Quadrant Analysis, and EMD will not be performed due to the lack of comparable literature. We will exploit the aforementioned methods upon completion of the study, so the PSD results can be used to confirm the clinical significance of these models. The repeated measure ANOVA model returned only two independent variables (HF power and SV power), which passed the equal variance test. We will first examine results of the repeated measures ANOVA analysis of the SV power defined in Chapter 6. Table 7-2. Results of the repeated measures ANOVA with independent variable (SV power). Session n 1 5 0.807 0.0857 2 5 0.729 0.089 3 5 0.846 0.121 4 5 0.878 0.0501 5 5 0.742 0.239 6 5 0.866 0.0836 7 4 0.613 0.186 SV Power00.20.40.60.811.21234567SessionMagnitude Figure 7-7. Plot of mean SV power with standard deviation bars across sessions

PAGE 82

70 Now we examine repeated measures ANOVA analysis of the HF power, also defined in Chapter 6. Table 7-3. Results of the repeated measures ANOVA with independent variable (HF power). Session n 1 5 68.2 25.925 2 5 133.6 144.921 3 5 48.46 26.837 4 5 41.32 23.877 5 5 178.1 172.465 6 5 47.22 58.952 7 4 150.75 126.818 HF Power-500501001502002503003504001234567SessionMagintude Figure 7-8. Plot of mean HF power with standard deviation bars across sessions While Figure 7-7 shows a possible non-linear trend with respect to GA, Figure 7-8 does not elicit any significant trend in time. First, the small sample size doesnt allow for the development of a trend. Secondly, there exist many exogenous variables, which conceal a developing trend corresponding to GA. These variables include, but are not limited to:

PAGE 83

71 Drug interventions and interactions Respiratory support interventions Physiological discrepancies in ANS maturation Changes in NICU environment Incorrect evaluations in GA Patient logs will be analyzed following the completion of the study for drug interventions. The PI will attempt to parallel the interventions with the recordings that contribute most to the standard deviation of each variable. It is known that certain drugs inhibit or even block parasympathetic and sympathetic responses. Furthermore, many of our subjects are on multiple drugs whose interactions with respect to the ANS are not known. Many subjects intermittently require respiratory interventions such as continuous positive airway pressure (CPAP). A closed-loop study showed that adequate application of long-term CPAP therapy produced substantial alterations in the major physiological mechanisms that influence HRV and blood pressure variability [36]. Based upon preliminary results, see Figures 7-7 and 7-8, final results need to include an analysis of how drug and respiratory interventions hinder the ability to identify trends in the ANS landscape. The significance of all exogenous variables mentioned above will emerge as the number of subject in the dataset increases. Unfortunately, these results are not included in this thesis, but we refer you to further publications by the PI and the author of this paper. For a complete time and frequency domain analysis of a randomly selected subject, please refer to Appendix A and B, respectively.

PAGE 84

CHAPTER 8 CONCLUSION We have successfully modified code written by HP to communicate with the Agilent/HP NICU monitor. This program records the ECG, respiratory rate, and rhyme state in a text file at a 500 Hz sampling rate. The text file can be read into MATLAB for QRS detection and HRV analysis. In MATLAB, we have created an algorithm to perform accurate detection of QRS complexes. The average percentage of missed detections is approximately 0.8%, which we feel represents a near optimal detection algorithm. For those QRS complexes that are undetectable, we have constructed an interpolation method to replace the missing data. Using our novel method of tagging outliers, we are able to remove outliers, but preserve trends across time. We have found that the most useful measurement of HRV in the time-domain is through the analysis of Poincar plots. Fitting an ellipse to the Poincar representation of the RRi is usefully in relating time-domain analysis to the frequency domain. Currently, most spectral representations described in literature use the Fourier transform. We have found that the Lomb normalized periodogram is a better representation of the power spectrum of RRi waveforms. The Lomb algorithm avoids the negative effects of interpolation and resampling required for FFT computation. Also, Empirical Mode Decomposition is a favorable method for measuring changes in the frequency spectrum across time. The EMD spectrum preserves more information regarding shifts in the frequency spectrum during the stimulus period than either the Lomb or Fourier spectrum. EMD has been proven to be a useful analysis tool to depict a COR. 72

PAGE 85

73 We have discussed the importance of both qualitative and quantitative measures of developmental changes in HRV and in identifying a COR; we have extracted numerous features using time and frequency domain HRV analyses, which will be used to analyze and test the hypotheses of the clinical study still in progress. We feel we have completed the design phase of the study and need only to increase our sample size. As the sample size increases, the clinical significance of each feature will be evaluated and we will move to interpret our findings and test our hypotheses.

PAGE 86

CHAPTER 9 FUTURE WORK A need exists in the NICU and beyond to determine the state of development of the ANS. Before an infant is discharged from the NICU, it is important to determine their ability to perform certain life support tasks. In Chapter 2, these life-supporting tasks were linked to the maturity of the ANS, specifically the PNS. The list below is compiled from the 1998 suggested guidelines for neonatal discharge established by the American Academy of Pediatrics (AAP) [37]. In the judgment of the responsible physician there has been: A sustained pattern of weight gain of sufficient duration. Adequate maintenance of normal body temperature with the infant fully clothed in an open bed with normal ambient temperature (24C to 25C). Competent suckle feeding, breast or bottle, without cardiorespiratory compromise. Physiologically mature and stable cardiorespiratory function of sufficient duration. Appropriate immunizations have been administered. Appropriate metabolic screening has been performed. Hematologic status has been assessed and appropriate therapy instituted as indicated. Nutritional risks have been assessed and therapy and dietary modification instituted as indicated. Sensorineural assessments, hearing and funduscopy, have been completed as indicated. Review of hospital course has been completed, unresolved medical problems identified, and plans for treatment instituted as indicated. 74

PAGE 87

75 The AAP has established these guidelines, which include subjective observations of cardiorespiratory function and sensorineural assessments. It is our hypothesis that cardiac function, as well as, neural assessments can be objectively realized using an index of ANS maturation. We will create this index of ANS functionality based on HRV comparisons with mature young adults. Below, we suggest a pilot study, which shall provide evidence that an index of maturation can be realized. Record Baseline Adult HRV Data In order to establish a baseline for a mature ANS with a synonymously high vagal tone, it is important to analyze the HRV of the young, healthy adult. ECG recordings will be obtained for a sample of 28 male and female adults meeting the following criteria: Between the ages of 18-30 No history of congestive heart failure, myocardial infarction, coronary artery disease, head trauma, or diabetes Normal sinus ECG rhythm The length of the recordings should be equal to those used in the study of the fetus. This is important we comparing both time and frequency domain features of HRV and the COR. Each subject should be oriented in the supine position to ensure consistency with ECG recordings of the neonates. Before recording, all subjects should remain in this position without auditory and visual simulation, until they are verified to be in REM or paradoxical sleep. Consistency with the neonatal recording described in the paper is a significant issue for accurate interpretation of HRV data and the adult period of REM sleep corresponds to the neonatal state of active sleep [38]. Siegel suggests that the phasic motor activation that characterizes REM sleep in the adult is also present in the

PAGE 88

76 neonate. The amplitude of the twitching, described during REM sleep, is more intense in neonates than in the adult, but a developmental continuity can be observed between "active sleep" in the neonate and REM sleep in the adult. Both active sleep in the neonate and REM sleep in the adult can be defined by purely behavioral criteria [38]. Therefore, the recordings should be taken in the presence of a medical professional qualified to identify phases of paradoxical sleep. An auditory stimulation that is widely familiar and contains no political or nationalistic propaganda should be used to prevent to prevent undesirable neural responses due to stress. The ECG recordings during the auditory stimulation can later be used to compare the COR of the neonate and adult. Creating the Index of ANS Maturation We suggest using the EMD method, detailed in Chapter 6, to compare the COR of the HRV groupings described in the previous subsection with the mature adult. The EMD data provides a time-based representation of the COR. The examiner should compare differences in the COR of the adult and neonate to determine an efficient method for relating these changes. We suggest exploring the use of power-ratios, neural networks, windowing functions, etc. as possible criteria to represent the EMD data in the form of an index. In addition, the development data can be compared using PCA and a neural network to classify the Lomb spectrum into different level of ANS maturation.

PAGE 89

APPENDIX A TIME-DOMAIN HRV ANALYSIS Week One 77

PAGE 90

78 Week Two

PAGE 91

79 Week Three

PAGE 92

80 Week Four

PAGE 93

81 Week Five

PAGE 94

82 Week Six

PAGE 95

83 Week Seven

PAGE 96

APPENDIX B FREQUENCY-DOMAIN HRV ANALYSIS Week One 84

PAGE 97

85 Week Two Week Three

PAGE 98

86 Week Four Week Five

PAGE 99

87 Week Six Week Seven

PAGE 100

LIST OF REFERENCES 1. BLISS, Causes of premature birth (2003), http://www.bliss.org.uk/support/causes.asp Accessed 4/4/2003. 2. Thomas, K., Causes of premature birth (2001), http://www.kerri.thomas.btinternet.co.uk/causes.htm Accessed 4/4/2003. 3. Chatow, U., Davidson, S., Reichman, B. & Akselrod, S. (1995). Development and maturation of theautonomic nervous system in premature and term infants using spectral analysis of heart rate fluctuations. Pediatric Research, 37, 294-302. 4. Stark, R., Myers, M., Daniel, S.,Garland, M., & Kim, Y. (1999). Gestational age related changes in cardiac dynamics of the fetal baboon. Early Human Development, 53, 219-237. 5. Boon, R., Learning Discoveries Psychological Services (2002), http://home.iprimus.com.au/rboon/HeartRhythmsandHRV.htm Accessed 1/12/2003. 6. Groome, L., Mooney, D., Holland, S., Smith, L., Atterbury, J., & Loizou, P. (1999). High Vagal Tone is Associated with More Effiecient Regulation of Homeostasis in Low-Risk Human Fetuses. Dev. Psychophysiology, 36, 25-34. 7. Krueger, C. (2002). Fetal heart rate variability and learning in the 28-34 week fetus. (Doctoral dissertation, University of North Carolina at Chapel Hill, 2001). Dissertation Abstracts International, 62, 11B. 8. Porges, S. (1995). Orienting in a defensive world: Mammalian modifications of our evolutionary heritage. A Polyvagal Theory. Psychophysiology., 32, 301-318. 9. Porges, S., Doussard-Roosevelt, J., Stifter, C., McClenny, B., & Riniolo, T. (1999). Sleep state vagal regulation of heart period patterns in the human newborn: An extension of the polyvagal theory. Psychophysiology, 36, 14-21. 10. Beauchaine, T. (2001). Vagal tone, development, and Grays motivational theory: Toward an integrated model of autonomic nervous system functioning in psychopathology. Development and Psychopathology, 13, 183-214. 11. Gagnon, R. (1992). Fetal behavioral states: Biological alteration. Seminars in Perinatology, 16, 234-238. 88

PAGE 101

89 12. Holditch-Davis, D. (1990). The development of sleeping and waking states in high-risk preterm infants. Infant Behavior and Development 13, 513-53 1. 13. Brackbill, Y. (1977). Behavioral state and temporal conditioning of heart rate in infants. In G. Etzel, J. LeBlanc, D. Baer (Eds.), New developments in behavioral research: Theory, method and application. Hillsdale, NJ: Erlbaum. 14. Kisilevsky, B. & Muir, D. (1984). Neonatal habituation and dishabituation to tactile stimulation during sleep. Developmental Psychology, 20, 367-373. 15. Thoman, E.B., & Whitney, M.P. (1990). Behavioral states in infants. Individual differences and individual analyses. In LJ. Colombo & J. Fagen (Eds.) Individual differences in infancy: Reliability, stability, and prediction, (pp. 113-136). Hillsdale, HJ: Erlbaum. 16. Arduini, D., Rizzo, G. & Romanini, C. (1995). Fetal behavioral states and behavioral transitions in normal and compromised fetuses. In J.-P. Lecanuet, W. Fifer, N. Krasnegor and W. Smotherman (Eds.), Fetal development: A psychobiological perspective, (pp. 83-100), Hillsdale, NJ: Erlbaum. 17. Faircloth, D., & Helms, R. (1986). A mixed model with linear covariance structure: A sensitivity analysis of the maximum likelihood estimates. Journal of Statistical Computer Simulations, 25, 205-206. 18. Gray, L. & Philbin, M. (2000). Measuring sound in hospital nurseries. Journal of Perinatology, 20(8), S100-S104. 19. Gottfried, A. & Hodgman, J. (1984). How intensive is newborn intensive care? Pediatrics, 74(8), 292-294. 20. Morris, B. Philbin, M. & Bose, C. (2000). Physiological effects of sound on the newborn. Journal of Perinatology, 20(8), S55-S60. 21. Gerhardt, K. (1989). Characteristics of the fetal sheep sound environment. Seminars in Perinatology, 13, 362-370. 22. Laguna, P., Caminal, P., Raimon, J., & Rix, H. (1991). Evaluation of HRV by PP and RR Interval Analysis Using a New Time Delay Estimate. IEEE Trans. Biomedical Eng., 63, 63-66. 23. American Heart Association (2002). Handbook of emergency cardiovascular care. AHA, 2002: Dallas, TX. 24. Pan, J., & Tompkins, W. (1985). Evaluation of HRV by PP and RR Interval Analysis Using a New Time Delay Estimate. IEEE Trans. Biomedical Eng., 32 n.3, 230-236.

PAGE 102

90 25. Niles, R., Stats lessons (2003), http://www.robertniles.com/stats/stdev.shtml Accessed 5/4/2003. 26. Malik, M. (1996). Heart rate variability: Standards of measurement, physiological interpretation, and clinical use. European Heart Journal, 17, 354-381. 27. Brennan, M., Palaniswami, M., & Kamen, P. (2002). Poincar plot interpretation using physiological model of HRV based on a network of oscillators. Am. J. Physiol. Heart. Circ. Physiol., 283, H1873-H1886. 28. Duarte, M. Software (1999), http://www.usp.br/eef/lob/md/software Accessed 6/4/2003. 29. Stein, P.K., & Kleiger, R.E. (1999). Insights from the study of heart rate variability. Annu. Rev. Med., 50, 249-261. 30. Laguna, P., Moody, G.B., & Mark, R.G. (1998). Power spectral density of unevenly sampled data by least-squares analysis: Performance and application to heart rate signals. IEEE Trans. Biomedical Eng., 45 n.6, 698-715. 31. Lomb N.R. (1976). Least-squares frequency analysis of unequally spaced data. Astrophysical and Space Science, 39, 447. 32. Glover, D.M. Sequence analysis II: Optimal filtering and spectral analysis (2000), http://w3eos.whoi.edu/12.747/notes/lect07/l07s05.html Accessed 5/20/2003. 33. Huang, N., Shen, Z., Long, S., Wu, M., Shih, H., Zheng, Q., Yen, N., Tung, C., & Liu, H. (1998). The empirical mode decomposition and the Hilbert spectrum for nonlinear and non-stationary time series analysis. Proc. R. Soc. Lond. A., 454, 903-995. 34. Rilling, G., Flandrin, P., & Gonalvs, P. Empirical mode decomposition (2002), http://www.ens-lyon.fr/~flandrin/emd.html Accessed March 25 th 2003. 35. Holditch-Davis, D., Edwards, L.J., &Helms, R.W. (1998). Modeling development of sleep-wake behaviors: I. Using the Mixed General Linear Model. Physiology & Behavior. 63 n.3, 6311-318. 36. Belozeroff, V., Berry R.B., Sassoon C.S.H., & Khoo M.C.K. (2002). Effects of CPAP therapy on cardiovascular variability in obstructive sleep apnea: A closed-loop analysis. Am. J. Physiol. Heart. Circ. Physiol., 282, H110-H121. 37. Lemons, J.A., & The American Academy of Pediatrics Committee on Fetus and Newborn (1998). Hospital discharge of the high-risk neonate: Proposed guidelines (RE9812). Pediatrics, 102 n.2, 411-417.

PAGE 103

91 38. Siegel, J (1999). The evolution of REM sleep. Handbook of Behavioral State Control. Lydic, R and Baghdoyan (Eds.), (pp. 87-100), Boca Raton, FL: CRC Press.

PAGE 104

BIOGRAPHICAL SKETCH Devin Lebrun received his bachelors degree in electrical engineering with a focus in communications from the University of Florida in 2001. He received his Master of Engineering in electrical engineering from the University of Florida in August 2003. His research interests include biomedical imaging, digital signal processing, and neural networks. 92


Permanent Link: http://ufdc.ufl.edu/UFE0001215/00001

Material Information

Title: Analysis of neonatal heart rate variability and cardiac orienting responses
Physical Description: Mixed Material
Language: English
Creator: Lebrun, Devin N. ( Dissertant )
van Oostrom, Johannes H. ( Thesis advisor )
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2003
Copyright Date: 2003

Subjects

Subjects / Keywords: Electrical and Computer Engineering thesis, M.E
Heart rate monitoring   ( lcsh )
Dissertations, Academic -- UF -- Electrical and Computer Engineering
Electrocardiography -- Data processing   ( lcsh )
Signal processing -- Digital techniques   ( lcsh )
Genre: bibliography   ( marcgt )
non-fiction   ( marcgt )
theses   ( marcgt )

Notes

Abstract: We have applied signal-processing methods to evaluate the clinical significance of heart rate variability (HRV) features, assess developmental changes in HRV, evaluate vagal tone, and demonstrate a cardiac orienting response (COR). The study is based on a sample of 28 low-risk premature newborns randomly assigned to one of two groups. Group 1 is exposed to an auditory stimulation beginning at 28 weeks GA and Group 2 begins exposure at 32 weeks GA. An eight-minute ECG recording is taken once a week from 28-34 weeks GA. The recordings were analyzed in the time and frequency domain using MATLAB. Extracted features are used to classify developmental changes in HRV across individuals and subject groups. In addition, we have used a staggered grouping approach to separate cardiac orienting responses from reflexive responses. At the conclusion of this thesis, the sample size (n=5) was not large enough to elicit any developmental trends in GA. However, we were able to extract useful features to describe developmental changes in HRV and we have acknowledged Empirical Mode Decomposition as an effective measure of a cardiac orienting response. Also, we have identified numerous exogenous interactions and perturbations which can interfere with the baroreceptor and chemoreceptor responses controlling HRV. These interactions were not considered in the original design of the study but will now be considered during final analysis.
Subject: density, ecg, heart, neonate, power, premature, spectral, stationarity, variability
General Note: Title from title page of source document.
General Note: Includes vita.
Thesis: Thesis (M.E.)--University of Florida, 2003.
Bibliography: Includes bibliographical references.
General Note: Text (Electronic thesis) in PDF format.

Record Information

Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
Resource Identifier: aleph - 002950135
System ID: UFE0001215:00001

Permanent Link: http://ufdc.ufl.edu/UFE0001215/00001

Material Information

Title: Analysis of neonatal heart rate variability and cardiac orienting responses
Physical Description: Mixed Material
Language: English
Creator: Lebrun, Devin N. ( Dissertant )
van Oostrom, Johannes H. ( Thesis advisor )
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2003
Copyright Date: 2003

Subjects

Subjects / Keywords: Electrical and Computer Engineering thesis, M.E
Heart rate monitoring   ( lcsh )
Dissertations, Academic -- UF -- Electrical and Computer Engineering
Electrocardiography -- Data processing   ( lcsh )
Signal processing -- Digital techniques   ( lcsh )
Genre: bibliography   ( marcgt )
non-fiction   ( marcgt )
theses   ( marcgt )

Notes

Abstract: We have applied signal-processing methods to evaluate the clinical significance of heart rate variability (HRV) features, assess developmental changes in HRV, evaluate vagal tone, and demonstrate a cardiac orienting response (COR). The study is based on a sample of 28 low-risk premature newborns randomly assigned to one of two groups. Group 1 is exposed to an auditory stimulation beginning at 28 weeks GA and Group 2 begins exposure at 32 weeks GA. An eight-minute ECG recording is taken once a week from 28-34 weeks GA. The recordings were analyzed in the time and frequency domain using MATLAB. Extracted features are used to classify developmental changes in HRV across individuals and subject groups. In addition, we have used a staggered grouping approach to separate cardiac orienting responses from reflexive responses. At the conclusion of this thesis, the sample size (n=5) was not large enough to elicit any developmental trends in GA. However, we were able to extract useful features to describe developmental changes in HRV and we have acknowledged Empirical Mode Decomposition as an effective measure of a cardiac orienting response. Also, we have identified numerous exogenous interactions and perturbations which can interfere with the baroreceptor and chemoreceptor responses controlling HRV. These interactions were not considered in the original design of the study but will now be considered during final analysis.
Subject: density, ecg, heart, neonate, power, premature, spectral, stationarity, variability
General Note: Title from title page of source document.
General Note: Includes vita.
Thesis: Thesis (M.E.)--University of Florida, 2003.
Bibliography: Includes bibliographical references.
General Note: Text (Electronic thesis) in PDF format.

Record Information

Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
Resource Identifier: aleph - 002950135
System ID: UFE0001215:00001


This item has the following downloads:


Full Text












ANALYSIS OF NEONATAL HEART RATE VARIABILITY AND CARDIAC
ORIENTING RESPONSES















By

DEVIN N. LEBRUN


A THESIS PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
MASTER OF ENGINEERING

UNIVERSITY OF FLORIDA


2003

































Copyright 2003

by

Devin N. Lebrun

































Dedicated to my father, Denis Lebrun, for his love and support.















ACKNOWLEDGMENTS

I would like to thank my advisor, Dr. Johannes van Oostrom, for all the advice he

has given me on realizing my career goals. I knew what I wanted, but he helped me get

from there to here and now onto somewhere great. He endured barrages of

scatterbrained, unintelligent questions and always surfaced with the answers I needed.

I am grateful to Dr. Charlene Krueger for allowing me to share in her research and

express my gratitude to Dr. Krueger, Dr. van Oostrom, and the College of Nursing for

providing me with funding throughout my research. I would also like to thank Kelly

Spaulding, John Hardcastle, Dr. JS Gravenstein, Dr. Richard Melker, and Dr. John Harris

for their support in various areas.

Words cannot describe my appreciation for my parents. They have taught me to be

determined, self-motivated, and not afraid of failure. I thank Linnea, for her

companionship and assistance throughout the past 6 years. Finally, thanks go to all my

friends for keeping me from doing something stupid (well, more stupid).
















TABLE OF CONTENTS
Page

A C K N O W L E D G M E N T S ................................................................................................ IV

LIST OF TABLES ...... ........................... ............. ........... ............... VIII

LIST OF FIGURES ......... ....................... ............ ........ ........... IX

ABSTRACT .............. .................... .......... .............. XI

CHAPTER

1 IN T R O D U C T IO N ............................................................................. .............. ...

2 AUTONOMIC NERVOUS SYSTEM AND HEART RATE VARIABILITY ..........4

A utonom ic N ervous System ...........................................................................4
D evelopm ental C changes ................................................................ ...................... 6
V agal Tone and H om eostasis ............................................... ............................ 7
C ardiac O renting R response ............................................................................. 8
Respiratory Sinus Arrhythm ia ............................................................. ........ .....8

3 PRELIMINARY RESULTS AND STUDY PROCEDURES............................... 10

P relim inary F finding s ......................................................... .. ...... .. ...... ............10
Proposed Study ............................................................... .............. ........ 11
Recitation Sessions .................. .......................... .. .. .. .............. ... 12
T e st S e ssio n s ................................................................................................. 1 3

4 T E S T IN G .......................................................................................16

5 QRS DETECTION .............................................................................. 21

N oise/B aselin e W an d er ..............................................................................................2 3
S lo p e D ete ctio n ..................................................................................................... 2 7
O u tlie rs ..............................................................................3 2
D direct RRi Analysis ............................................................................................ 34
Slope Analysis ........................................................................ ......... .......... ........ 36
D differential A naly sis ........................................................................... 37
Interp olation ........... .. ......... ............. .. .. ....................................... 39



v









6 STA TISTIC A L A N A L Y SIS ............................................................. ....................44

T im e B a se d ....................................................................................................4 4
First O rder Statistics ............................................................................ 44
P o in car P lots ...............................................................4 5
Q u adrant A naly sis ............................. .... ................ .. ......... .... ............4 8
Frequency Based............................................ ........ 50
Fast Fourier Spectral A nalysis......................................... .......................... 51
L om b Spectral A naly sis ........................................ .........................................52
Empirical Mode Decomposition...................... ..... .......................... 54
Statistical A analysis of Param eters ........................................ .......................... 57

7 DISCU SSION AND RESULTS .......................................... ............................. 59

E engineering R results .............................................. ... ............ .............. 59
QRS D election Algorithm ...........................................................................59
H RV Plots and Statistics .................................. ......................................60
HR Plot .................. ..... ................... 62
Q u a d ran t P lo t........................................................................................... 6 3
P oin care P lot ............................................................64
F ourier and L om b P lots......... .............................................. ..... .......... 65
EMD Plots ................ ......... .......... ........67
Clinical Results ............. ...... ..... .. ......................................... 68

8 C O N C L U SIO N ......... ......................................................................... ......... ........72

9 FU TU R E W O R K ........................ ............................ .. .... ................ 74

Record Baseline A dult HRV D ata.................................... ........................ 75
Creating the Index of ANS Maturation .............. ....................................76

APPENDIX

A TIME-DOMAIN HRV ANALYSIS .............................. 77

W e e k O n e .......................................................................... 7 7
W e e k T w o ........................................................................ 7 8
W eek T h re e ...................................................................... 7 9
W eek F ou r ............................................................................ 80
W eek Five .................................... .................................. .......... 81
W e e k S ix ......................................................................... 8 2
W eek S ev en ........................................................................... 83

B FREQUENCY-DOMAIN HRV ANALYSIS ................................... ............... 84

W e e k O n e .......................................................................... 8 4
W e e k T w o ........................................................................ 8 5










W e e k T h re e ............................................................................................................ 8 5
W eek Four ................................................................86
W e e k F iv e .............................................................................................................. 8 6
W e e k S ix ................................................................................................................ 8 7
W eek Seven ................................................................87

LIST OF REFEREN CE S ................................................................. .. ......... 88

BIO GR A PH ICA L SK ETCH ........................................................................................ 92
















LIST OF TABLES


Table p

3-1 Organization of R ecitation and Test Sessions ...........................................................11

7-1 Shows the performance of the QRS detection algorithm ...........................................60

7-2 Results of the repeated measures ANOVA with independent variable (SV power)..69

7-3 Results of the repeated measures ANOVA with independent variable (HF power)..70
















LIST OF FIGURES


Figure p

2-1 The divisions of the central nervous system ........................................ ..................4

3-1 Time-line for the windowing of test data. .......... ..... ...........................14

4-1 Example of text file generated from Data Acquisition Program............................18

4-2 The setup for the test sessions. ........................................ ............................ 19

5-1 Q R S detection algorithm ............................ ............. .................................... 23

5-2 Magnitude and phase response of the 60Hz notch filter used to filter the raw ECG
signal. ..................................................................25

5-3 Magnitude and phase response of the bandpass filter used to filter the raw ECG
signal. ..................................................................26

5-4 Magnitude and phase response of the differentiator function.............................27

5-6 Example of detected QRS complexes and beat intervals...................................31

5-7 Histogram of RRi distribution from one subject................................. ...............33

5-8 Standard deviations of the Normal Distribution .......................................... 34

5-9 Illustrates the bouncing effect and upper and lower bound symmetry problems with
using the direct RRi analysis method for tagging outliers. .....................................35

5-10 Illustrates an RRi trend that would not be preserved using the slope analysis
m ethod for tagging outliers. ...... ........................... ......................................37

5-11 Outliers that have been tagged, using the differential method, are marked by a
green dot. Shows the efficiency of the method to preserve trend and remove
bouncing effects. ......................................................................39

5-12 LaGrange Polynomial Interpolation...................... .... .......................... 40

5-13 L near Interpolation ...... ...................................................................... .. .... ... ....4 1

5-14 Cubic Spline Interpolation ............................................... ............................ 41









5-15 C ub ic Interp olation .......................................................................... ................... 42

6-1 Poincare plot representation of RRi. Since data points overlap, colors, coordinated
with the visible spectrum, are used to indicate magnitude at each point ................47

6-2 Defines the quadrants used in Quadrant Analysis.................................. ............... 48

6-3 Example of Quadrant Analysis applied to an RRi dataset .................................49

7-1 HR plots of example RRi data. From top to bottom they represent: the (a)
developmental, (b) prestimulus, (c) stimulus, and (d) poststimulus periods............62

7-2 Quadrant Analysis plots of example RRi data. From top to bottom they represent:
the (a) developmental, (b) prestimulus, (c) stimulus, and (d) poststimulus periods.63

7-3 Poincare plots of example RRi data. From top to bottom they represent: the (a)
developmental, (b) prestimulus, (c) stimulus, and (d) poststimulus periods............64

7-4 Fourier PSD plots of example RRi data. From top to bottom they represent: the (a)
developmental, (b) prestimulus, (c) stimulus, and (d) poststimulus periods............65

7-5 Lomb PSD plots of example RRi data. From top to bottom they represent: the (a)
developmental, (b) prestimulus, (c) stimulus, and (d) poststimulus periods............66

7-6 Hilbert-Haung spectrum of the COR windows of an example RRi dataset.............68

7-7 Plot of mean SV power with standard deviation bars across sessions ...................69

7-8 Plot of mean HF power with standard deviation bars across sessions....... ........ 70















Abstract of Thesis Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Master of Engineering

ANALYSIS OF NEONATAL HEART RATE VARIABILITY AND CARDIAC
ORIENTING RESPONSES


By

Devin N. Lebrun

August 2003

Chair: Johannes H. van Oostrom
Major Department: Electrical and Computer Engineering

We have applied signal-processing methods to evaluate the clinical significance of

heart rate variability (HRV) features, assess developmental changes in HRV, evaluate

vagal tone, and demonstrate a cardiac orienting response (COR).

The study is based on a sample of 28 low-risk premature newborns randomly

assigned to one of two groups. Group 1 is exposed to an auditory stimulation beginning

at 28 weeks GA and Group 2 begins exposure at 32 weeks GA. An eight-minute ECG

recording is taken once a week from 28-34 weeks GA. The recordings were analyzed in

the time and frequency domain using MATLAB. Extracted features are used to classify

developmental changes in HRV across individuals and subject groups. In addition, we

have used a staggered grouping approach to separate cardiac orienting responses from

reflexive responses.

At the conclusion of this thesis, the sample size (n=5) was not large enough to elicit

any developmental trends in GA. However, we were able to extract useful features to









describe developmental changes in HRV and we have acknowledged Empirical Mode

Decomposition as an effective measure of a cardiac orienting response. Also, we have

identified numerous exogenous interactions and perturbations which can interfere with

the baroreceptor and chemoreceptor responses controlling HRV. These interactions were

not considered in the original design of the study but will now be considered during final

analysis.














CHAPTER 1
INTRODUCTION

The premature infant faces many obstacles beyond the control of present day

medical science. Since the loss of the "in utero environment" strips the premature infant

of the most advantageous setting for growth and development, it is increasingly important

for medical science to reach a greater understanding of the developmental changes during

this final trimester. Survival rates of neonates will continue to increase as science begins

to elucidate more regarding this crucial period of development. In general, scientific

advancements are predicated upon a divide and conquer approach. It is important to

embrace the past discoveries as background for future research. Therefore, this study

expands upon a preliminary study conducted by the Principal Investigator (PI), Charlene

Krueger, Ph.D, ARNP. The findings of the preliminary study indicate that the analysis of

heart rate variability (HRV) can provide significant insight into the development of the

premature and fetus' autonomic nervous system (ANS). In collaboration with Dr.

Krueger, the current study moves from the comforts of the uterus to the premature

infant's continuing development in a Neonatal Intensive Care Unit (NICU) setting.

About a third of premature births occur for no apparent reason and with little or no

warning. The list below describes some known causes of premature birth [1,2]:

* Pre-eclampsia, occurring in about 1 in 14 pregnancies, causes around a third of all
premature births.

* Pregnancies involving multiple fetuses are likely to end early.

* Stressful events can elicit the mother into a premature labor.









* An infection involving the uterus may trigger an early delivery.

* The mother's water may break early, starting the delivery process.

* A shortage of oxygen exchange to the placenta may cause a doctor to advise a
caesarian section.

* Malnutrition or chronic diseases, including high blood pressure, diabetes, kidney
disease and hypothyroidism can cause premature births.

Regardless of the cause for an early birth, premature infants are commonly highly

dependant upon the continuation of breathing support, intravenous nutrition, intravenous

antibiotics, and phototherapy. These infants are frequently placed in NICUs, which

specialize in caring for the very young and/or very small. The neonate's high

dependency upon medical interventions demonstrates a need to establish criteria

regarding his or her efficiency at regulating homeostasis. Homeostasis is the

maintenance of equilibrium in a biological system by means of automatic mechanisms,

which counteract influences trending toward disequilibria. It involves constant internal

monitoring and regulating of numerous factors, including oxygen, carbon dioxide,

nutrients, hormones, and organic and inorganic substances. Due to homeostatic controls,

the concentrations of these substances in body fluid remain unchanged, within limits,

despite changes in the external environment. Efficient regulation of these parameters is

necessary to support life, on all levels. Therefore, the ability to evaluate homeostatic

controls would facilitate the decision process leading to the neonate's discharge from the

unit. The events surrounding the control of homeostasis will be discussed further in

Chapter 2.

We hypothesize that an index of HRV will provide sufficient insight into the

maturity level of the neonate's ANS. In turn, this will provide us with information

regarding the ability of the neonate to control events of homeostasis. In a study by









Chatow et al, increasing conceptual age, CA, or gestational age, GA, is correlated with

neonatal autonomic control. Chatow used a power ratio, discussed in Chapter 2, which

was assumed to reflect the sympathovagal balance of the ANS [3]. In a study of the fetal

baboon, Stark et al. suggest that "Sequential decreases in fetal heart rate, increases in RR

interval (RRi) variability and increases in changes in RRi and ARRi with age imply an

overall maturation in autonomic cardio-regulatory control processes. Increases with

gestation of high frequency components of variability are compatible with enhanced

parasympathetic modulation of the fetal heart rate" [4]. This thesis will attempt to

expand upon these studies to examine developmental changes, cardiac orienting

responses, and to introduce a maturation index based on time and frequency-domain

analysis of neonatal HRV. Current methods of frequency domain analysis tend to center

around one method of analysis. Therefore, new methods of analysis will be constructed

as a result of combining the most functional signal processing tools described in recent

literature. This thesis will explore and compare a variety of spectral analysis methods and

focus on those methods deemed most clinically significant. In addition, we will parallel

techniques used to examine the frequency spectrum of HRV with time-domain

techniques to provide a complete clinical evaluation of each patient and grouping.














CHAPTER 2
AUTONOMIC NERVOUS SYSTEM AND HEART RATE VARIABILITY

Autonomic Nervous System

The ANS is part of the human central nervous system (CNS). The CNS can

essentially be divided into three closely interacting sections: the motor nervous system,

sensory nervous system, and autonomic nervous system. The ANS originates from the

brainstem at the base of the spinal cord. It includes the nerves, which innervate the

smooth muscles of the heart, internal organs, and glands. The ANS is subdivided into the

sympathetic (SNS) and the parasympathetic (PNS) systems, which originate from

different areas in the brainstem and spinal cord.


Sensory
Nervous Syste


Figure 2-1. The divisions of the central nervous system.









Both systems work simultaneously in an antagonistic nature, with an increased

activity in one, inevitably causing a decrease in the other. However, neither substructure

of the ANS is ever completely disabled by the response of the other. While both the

branches of the ANS serve to regulate the internal organs by stimulating or inhibiting

their activity, these two substructures respond to different biochemical transmitter

substances [5]. The actions of the PNS are started by the release of acetylcholine.

Parasympathetic nerve fibers slow the heart rate, narrow the bronchioles, and cause the

release of insulin and digestive fluids. On the other hand, the actions of the SNS are

started by the release of norepinepherine. SNS responses include relaxing the tubes of

the bronchioles, speeding up the heart, and slowing the digestive tract muscles, while

increasing the release of glucose from the liver [5]. Given the broad spectrum of

physiological events controlled by the ANS, it is important to establish criteria to

evaluate its tone. Past research has indicated, through the measurement of heart rate

variability, an inference can be made regarding the functionality of the ANS [3,6,7].

HRV is defined as beat-to-beat fluctuations in heart rate attributed to electrical

impulses generated by the sino-atrial (SA) node. The SA node is commonly referred to

as the heart's internal pacemaker, and the vagus nerve, or the Xth cranial nerve, inhibits

this pacemaker. The PNS influences the SA and AV nodes via the vagus nerve. This

influence is produced by release of the neurotransmitter acetylcholine at the vagus nerve

endings, which result in the slowing of activity at the SA node and a slowing of the

cardiac impulse passing into the ventricles. The SNS has the opposite effect through the

release of norepinepherine at the sympathetic nerve endings, thereby increasing the heart

rate [5].









HRV reflects the response of the ANS to exogenous or endogenous perturbations.

Numerous factors contribute to HRV including blood pressure, temperature, respiration,

state of oxygenation, ventilation, biochemical influences of the acid-base balance, and

psychological parameters [3]. Continuous changes in sympathetic and parasympathetic

neural impulses induce changes in heart rate and cause oscillations around the mean,

defined as HRV. By studying heart rate variability, an opportunity exists to study cardiac

dynamic behavior and functionality of the ANS. Since the Autonomic Nervous System's

control of heart rate is dynamic and nonlinear, it is necessary to explore viable

parameters, which can be used to access the causes of these beat-to-beat fluctuations.

This thesis will address time-domain, linear frequency-domain, and nonlinear frequency-

domain characteristics of the HRV. Thus the framework for the study has been set, with

an attempt to answer each of the following questions:

1. What clinically significant features can be extracted from HRV?
2. What are the developmental changes in HRV?
3. Can we establish criteria to assess vagal tone?
4. Can the premature infant learn or demonstrate a cardiac orienting response, and
does HRV change as learning emerges?

Developmental Changes

Immediately following birth, the heart rate is extremely sensitive to physiological

state of the infant. The developmental changes in HRV are hypothesized to indicate a

change in the parasympathetic tone. While early fetal life indicates a strong dependence

on the SNS, these two systems trend toward an equal balance in the term infant [3]. This

thesis assumes a parallel between the fetus and the premature infant. The developmental

changes in heart rate variability can be measured through time and frequency domain

statistics. Developmental trends of HRV will be measured across subject groups and









gestational age. Time and frequency domain indexes will be used to access maturation of

the neonate's ANS, or ability to regulate homeostasis.

Vagal Tone and Homeostasis

Cardiac vagal tone is a construct that describes the functional relationship between

the brainstem and the heart. Cardiac responses to brief low intensity stimuli should be

deceleratory, indicating receptiveness to incoming stimuli. Cardiac responses to

sustained stimuli should be acceleratory, indicating attempts to shut it out [8]. Vagal tone

reflects individual differences in degrees of influence of the vagus on the heart rate. High

vagal tone is associated with the ability to attend selectively to stimuli and maintain

attention during a task, the ability to maintain homeostatic integrity, and infers healthier

ANS functioning over the long-term.

Homeostasis is primarily maintained by the PNS and provides a physiological

framework for the development of complex behaviors. Therefore infants with a high

parasympathetic tone are more efficient regulators of homeostasis than those dominated

by sympathetic responses [6]. The PNS is primarily responsible with coordinating the

life-support process, which intrinsically provides maximum growth and restoration of the

body's tissues and organs [9]. More effective regulation of homeostatis may be an

important factor underlying the relationship between high resting vagal tone and

favorable neurobehavioral development [6]. Moreover, vagal stimulation controls life-

support processes including sucking, swallowing, thermoregulation, and vocalization [9].

The search for an accurate measurement of parasympathetic and sympathetic tone has

only begun to express its value. This thesis is anticipated to contribute to the

comprehension of these physiological parameters.









Cardiac Orienting Response

The existence of a cardiac orienting response (COR) suggests that HRV includes a

neurongenic response. The deceleratory parasympathetic component is provided by the

bi-directional vagus nerve, which includes both efferent and afferent fibers. Efferent

fibers originating in the brain stem terminate on the SA node. Vagal afferent fibers

originate throughout the heart and project to the nucleus tractus solitarius [10]. These

fibers serve as a continuous response and feedback mechanism between the heart and the

brain, facilitating regulation of cardiac function. Porges et al. suggest that the time

course of the response, the effects of neural blockades, and studies with clinical

populations support this theory. As noted by Porges the heart rate deceleration associated

with the cardiac orienting response is rapid and usually returns rapidly to baseline. Also,

the latency characteristics of the cardiac orienting response are similar to the opto-vagal,

vaso-vagal, baroreceptor-vagal and chemoreceptor-vagal reflexes [8]. Since short latency

heart rate reactivity is mediated by the vagus nerve, the magnitude of the cardiac

orienting response may be an index of vagal regulation.

Respiratory Sinus Arrhythmia

Repiratory sinus arrhythmia (RSA) has long been used to estimate vagal tone

because of its association with PNS. RSA results from increases in vagal efference

during exhalation, which decelerates heart rate, and decreases in vagal efference during

inhalation, which accelerate heart rate. Recently, the accuracy of the correlation between

RSA and vagal tone has been questioned because it only partially accounts for the beat-

to-beat fluctuations of the heart [10]. This thesis will address RSA as a possible criterion

to extract vagal tone, but due to the nature of the study will not pharmacologically prove






9


the efficiency of this technique. However, since RSA accounts for a significant amount

of variability in heart rate, it is impossible to completely ignore this aspect.















CHAPTER 3
PRELIMINARY RESULTS AND STUDY PROCEDURES

Preliminary Findings

In a preliminary study by the principal investigator [7], cardiac activities during

two consecutive quiet periods of 16 healthy fetuses were measured. The exploratory

phase used a repeated measures design to follow the normative changes in fetal HRV

between 28-34 weeks. Changes in HRV were explained by normalized power shifts in the

Fast Fourier Transform (FFT) frequency spectrum. A multi-group pretest-posttest

experimental design determined whether 28-34 week-old fetuses would become familiar

with a rhyme, recited by the maternal parent. Developmental changes in the HRV

frequency spectrum occurred in a nonlinear fashion and individual differences were

apparent. The total power diminished significantly with increasing gestational age.

Furthermore, the study provided conclusive evidence that the 28-34 week old fetus was

capable of responding to a nursery rhyme their mother recited aloud. The response to the

nursery rhyme was described by a shift in the power of HRV frequency spectrum to the

high frequency bands, which is synonymous with an increase in parasympathetic activity.

This conclusion parallels all hypothesized learning capabilities of the premature infant.

The testing sessions for the preliminary study utilized a recording of a rhyme

spoken by an unfamiliar female, which insured the fetuses' reactions would be elicited by

familiarity with the acoustic properties of the rhyme. Once learning was detected,

ongoing interactions between learning, movement and HRV were detected. A









nonparametric signed test was used to compare differences across time. The preliminary

study revealed a significant downward trend in fetal movement and an upward trend in a

high (.17-.40 Hz) frequency once a cardiac orienting response or learning was identified.

Since the subject size was small, there exists a need for a more expanded and complete

study before any conclusions can be drawn. However, the findings of the fetal study

provide the framework for this expanded research, conducted in collaboration with the PI

of the preliminary study.

Proposed Study

A convenience-based sample of 28 low-risk premature newborns admitted to the

NICU at Shand's Teaching Hospital are recruited, in blocks of four, and randomly

assigned to one of two groups. The target age for admission into the study is 27-28

weeks post-conceptional age. Group 1 will begin exposure to the nursery rhyme at 28

weeks and those in Group 2 will begin exposure at 32 weeks. Staggering the groups shall

assist in determining whether an orienting response was a function of learning or simply a

reflexive response to presentation of the rhyme. All subjects will be excluded from

analysis if they are subject to one or more of the following:

* An abnormal head ultrasound
* A sensorineural hearing loss
* A confirmed prenatally transmitted virus/bacterial infection
* Cardiac abnormalities

Table 3-1. Organization of Recitation and Test Sessions

Group 28 weeks 29 weeks 30 weeks 31 weeks 32 weeks 33 weeks 34 weeks
1 Recit Recit Recit Recit Recit Recit Reie
____Test Test Test Test Test Test Test
2 Test Test Test Test Recit Recite Recit-
Only Onlnl y nly Only Test -Test Test









Recitation Sessions

The mother of the infant will complete a recording of a nursery rhyme, identical to

that presented in the preliminary study. The nursery rhyme, taking approximately 15

seconds to recite, will be played 3 consecutive times to provide the infant with 45s of

auditory stimulation. The maternal recitation sessions will occur once in the morning and

again in the evening, each day; Group 1 will begin at 28 weeks and Group 2 at 32 weeks

of age. The purpose of the recitation session is to introduce the neonate to the rhyme

using a familiar voice.

The nursery rhyme will be presented once the premature infant is judged to be in a

state of active sleep and at least fifteen minutes following a feeding [11]. Between 28-34

weeks post-conceptional age active sleep accounts for approximately 70% of the entire

sleep-wake cycle [12]. Learning in the normal newborn, can occur during periods of

active sleep [13,14] and have occurred in the fetus during quiet periods of their quiet-

activity cycle [7]. Criteria established by Thoman and Whitney [15] and Holditch-Davis

[12] will be used to detect active sleep. The subject is in an active sleep state when:

* Eyes closed
* Respiratory rate under 25 breaths per minute
* Limited movement of extremities

Subjects will be monitored at the same time each week because of potential

circadian influences on heart rate patterns and movement [16]. A hospital log of

recitation times and the time since the neonate's last meal will be kept because variations

related to unexpected changes in hospital routines are expected. These recitation sessions

will be completed entirely separate from the testing session described below.









Test Sessions

Each group will undergo a 480 second ECG recording during each test session.

Once more, the criteria established by Thoman and Whitney [15] and Holditch-Davis [12]

will be used to detect active sleep before the test session begins. The ECG recording will

be divided into two stages. The first stage of this study incorporates an exploratory, two-

group pretest-posttest subject set which will be used to methodologically replicate the

findings from the preliminary study in the premature newborn during a similar

developmental time period (28-34 weeks post-conceptional age). The exploratory design

will follow specific changes in HRV, measured by time series analysis, power spectral

analyses, and time dependant spectral analysis over 28-34 weeks post-conceptional age.

A 240-s window separated from the delivery of the auditory stimulus and commencement

of the second phase will be used to represent developmental changes in HRV. It is

hypothesized that developmental changes in HRV will occur in a decreasing, nonlinear

fashion over 28 to 34 weeks gestational age. Two interactions (gestational age vs. gender

and gestational age vs. group assignment) will be included in the mixed general linear

model. The mixed model allows for the individual random effects when estimating fixed

population effects [17].

During a second period of confirmed active sleep, a two-group pretest-posttest

experimental design will be used to document when a cardiac orienting response to a

presented nursery rhyme first occurs, and how movement and HRV change once it is

detected. Again, a nursery rhyme identical to that presented in the preliminary study will

be used. An unfamiliar, English-speaking female performs the recitation of the nursery

rhyme used in these test sessions. To avoid additional exposure to this unfamiliar voice,

this person will not be in attendance for presentations of the rhyme to the infant. The









nursery rhyme takes approximately 15 seconds to recite, and will be repeated 3

consecutive times to provide the 45s window of auditory stimulation. The testing

sessions using the unfamiliar recording of the nursery rhyme will occur once a week for

both groups from 28-34 weeks post-conceptional age.








Variable 240 s 45 s 45 s 45 s


Washout Developmental Prestimulus Stimulus Poststimulus
Analysis




Figure 3-1. Time-line for the windowing of test data.

This second stage will be separated into three 45s windows, in which statistical

analysis of heart rate variability will be individually analyzed. The first window will be

used to detect the baseline state of the infant before the presence of any auditory

stimulation. The second window, immediately following the first, chronologically

corresponds to the recitation of the nursery rhyme. This window will be used to observe

time and frequency domain changes in HRV, which may reflect the hypothesized COR.

A third consecutive window will be used to examine how the neonate responds after the

auditory stimulation has been removed from the environment. The third window will be

used to verify that an observed COR response is not merely a result of coincidental

changes in HRV. In addition, a count of movements will be recorded onto the hospital

log during each test session. Our hypotheses pertaining to a COR suggest, (a) over 28-34






15


weeks of gestational age, it will take 2-5 weeks of recitation to detect and maintain a

cardiac orienting response in each subject and (b) once a cardiac orienting response has

been observed, subsequent testing session will demonstrate a trend of movement counts

and specific HRV characteristics.














CHAPTER 4
TESTING

The rhymes will be played over a 2.5-inch Radio Shack speaker positioned 20 cm

from the infant's ear. Both versions of the rhyme (maternal and unfamiliar) will be

delivered at 55 dB (+ 5 dB) as measured by a Bruel & Kjaer sound level meter (2239)

using A-weighted sound pressure levels [18]. This decibel level is just below the normal

levels of conversational speech (58-60 dB) and is significantly lower than the background

sound levels reported in the literature (70-80 dB + 20 dB) for NICUs [19,20]. The

decision to present both versions of the rhyme (maternal and unfamiliar) at 55 dB (+ 5

dB) was based on preliminary findings in the premature fetus where a sound level of 75

dB was used [7]. This decibel level corrects for a 20-decibel attenuation due to passage

of sound through the uterine wall [21], bringing the sound level down to approximately

55 dB or to just below the level of normal conversational speech (58-60 dB). The study

will record decibel levels at the infant's ear prior to presentation of the maternal and

unfamiliar rhyme. In the preliminary study, this sound level (75 dB) did not elicit a

cardiac deceleration until a history of recitation of the nursery rhyme had occurred [7].

An RS232 interface from the Agilent Neonatal CMS 2001 neonatal monitor to an

IBM compatible computer (Dell Inspiron 8100 Laptop) was constructed to obtain the

ECG signal and respiratory rates. The data acquisition program is designed to run in

DOS and may be activated from the Windows Desktop of Dell Inspiron 8100 Laptop by

double clicking on the "Neonatal Data Acquisition" icon. The C program, originally

written by Hewlett Packard (HP), was modified from an earlier version to in order to









record the ECG and respiratory rate. The HP program was initially written to

demonstrate that information from their monitors could be transferred and displayed on

peripheral devices through RS-232 interfacing. The sample program was only available

in a DOS format. Although this format is not optimally user friendly, an extensive user

interface was not necessary to transfer the required signals into readable text files.

Furthermore, HP uses a complex communication protocol, which makes it desirable to

avoid writing an original program. The original program was capable of display the ECG

signal on a peripheral monitor, which implies that the ECG information was already

being transferred. The program modification involved the following:

* Requesting the respiratory rate signal and displaying the information on the screen.
* Creating a user interface to request the patient number.
* Transferring the requested information to a readable text file.


Before data acquisition can begin, the Agilent monitor must be set to a baud rate of

19200 and the transition must be set to a high to low trigger. If the baud rate or transition

is not correctly set, the program will time-out after eight seconds. When the data

acquisition program is started the user enters the patient number and is asked to confirm

this number before the recording process begins. The program passes a text string, which

signifies the rhyme state and allows the user to mark the data when the rhyme is

presented and upon completion. The three measurements: ECG, respiratory rate, and the

rhyme state are recorded in an ASCII text file labeled with the month, date, hour (in 24h

format), and minute (MMDDHHmm.txt) the program is initialized. ASCII was the

chosen format for data storage to permit the user to utilize other analysis programs (i.e.

Microsoft EXCEL). The parameters are stored in the format directly readable as a

MATLAB matrix and can later be separated into three separate vectors. Preceding the











matrix is an ASCII timestamp, which is recorded once the program has finished the


initialization process (DDHHmmss), and appended by another timestamp when the

program is terminated. These time stamps allow the user to verify the length of the

recording.

The program is designed to work with an ECG/Resp module, which will transfer


both the ECG and the respiratory rate signal to the Dell Inspiron 8100 Laptop. However,

in the absence of the ECG/Resp module it is possible to use only the singular ECG


module. This will not record the respiratory signal, but will still record the ECG signal.

The first column recorded in the ASCII text file represents the unsigned 12-bit digital


values of the ECG signal. The second column contains the information regarding the

rhyme state. The final column contains the respiratory rate. The resulting format of the


text file is illustrated in the figure below.




06101635
1985,0,51,
1996,0, 51,
1996,0, 51,
1997,0,51,
1997,0, 51,


2027, 1,58,
2028, 1 58,
2028, 1 58,
2028, 1, 58,
2049,1, 58,


2079,0,61,
2078,0, 61,
2089,0, 61,
2078,0, 61,
2078,0, 61,
06102452



Figure 4-1. Example of text file generated from Data Acquisition Program









All columns are recorded at a sampling rate of 500Hz, which was preserved from the HP

demo program as the rate information was sent to the screen. This sampling rate, much

higher than the 200Hz often described in literature, was preserved to prevent jagged QRS

complexes. The column corresponding to the state of the rhyme is initialized to '0', until

the 'r' key is pressed and the value changes to '1'. The value remains at '1' until the 'd'

key is pressed, which signifies the end of the presentation and returns the value to '0'.

When all desired data has been recorded, the user presses the 'q' key twice to exit the

program and return to the Windows desktop.


Figure 4-2. The setup for the test sessions.









Figure 4-2 illustrates the equipment setup during the test sessions. A male 25-pin

RS232 port in the back of the Agilent monitor is interfaced with the female 9-pin RS232

port Dell Inspiron 8100 Laptop. The RS232 cable provides means of serial

communication between the Agilent monitor and the Dell Laptop. The Agilent Neonatal

CMS 2001 monitor continually monitors the infant's vital signs, with a standard 3-lead

ECG setup. The speakers, connected to the CD CopyWriter Live, are placed 20cm from

the infant's ear. This setup is replicated for every test session, regardless of the subject's

group number.














CHAPTER 5
QRS DETECTION

Once all data from the testing session is acquired the focus shifts to the data

analysis. The first goal is to perform time-series operations, yielding discernable

characteristics of the HRV. The HRV is basically controlled by the effect of the vagal

and sympathetic influences of the sino-atrial node. Therefore, it seems the best

representation of the HRV variability would be expressed as a function of the time

intervals between consecutive P waves. However, due to the inconsistent positioning and

suboptimal attachment of the infants ECG leads, P waves characteristically have a low

signal-to-noise (SNR) ratio [22]. Therefore, in the absence of cardiac arrithymias,

addressed in the exclusion criteria, it seems more reliable to replace the detection of PP

intervals (PPi), with RRi. While this may include an extra variable in the analysis of the

HRV, it seems trivial compared to the accuracy of the detection of PPi. It is our

assumption that variations in the PR intervals will be equal or less than error of detection

of the PPi. When considering an error-percentage of P-wave detection greater than the

standard deviation of the PR interval and greater than the percent error in R-wave

detection, the replacement of PPi with RRi becomes trivial. The occurrence of the

following cardiac arrhythmias causes this assumption to fail [23]:

* ATRIAL FLUTTER the atrial flutter waves, known as F waves, are larger than
normal P waves and they have a saw-toothed waveform. Not every atrial flutter
wave results in a QRS complex because the AV node acts as a filter. Some flutter
waves reach the AV node when it is refractory and thus are not propagated to the
ventricles. The ventricular rate is usually regular but slower than the atrial rate.









* ATRIAL FIBRILLATION Atrial fibrillation occurs when the atria depolarize
repeatedly and in an irregular uncontrolled manner. As a result, there is no
concerted contraction of the atria. No P-waves are observed in the EKG due to the
chaotic atrial depolarization. The chaotic atrial depolarization waves penetrate the
AV node in an irregular manner, resulting in irregular ventricular contractions. The
QRS complexes have normal shape, due to normal ventricular conduction.
However, the RR intervals vary from beat to beat.

* VENTRICULAR TACHYCARDIA Ventricular tachycardia occurs when
electrical impulses originating either from the ventricles cause rapid ventricular
depolarization. Since the impulse originates from the ventricles, the QRS
complexes are wide and bizarre. Ventricular impulses can be sometimes conducted
backwards to the atria, in which case, P-waves may be inverted. Otherwise, regular
normal P waves may be present but not associated with QRS complexes. The RR
intervals are usually regular.

* THIRD DEGREE AV BLOCK Atrial rate is usually normal and the ventricular
rate is usually low, with the atrial rate always faster than the ventricular rate. P
waves are normal with constant P-P intervals, but not associated with the QRS
complexes. QRS may be normal or widened depending on where the escape
pacemaker is located in the conduction system. The result is unrelated atrial and
ventricular activities due to the complete blocking of the atrial impulses to the
ventricles.

Subjects whose electrocardiogram display any of the arrhythmias listed above are

removed from analysis on the basis of violating the exclusion criteria. In conclusion, all

ECG time-domain analysis of the HRV will consider RRi rather than the ideal analysis of

the PPi for the remainder of the thesis.

The first step in the analysis of the HRV is to extract the RRi from the recorded

ECG waveforms. The ASCII text formatted matrix addressed in the data acquisition

chapter is read into MATLAB. The result is an N by 3 matrix, with N being the number

of samples. This matrix is then separated into 3 vectors of length N. Vector one contains

the ECG information; vector two contains the respiratory rate; and vector three contains

information regarded the state of the rhyme. Only vector one expresses useful

information for analysis of RRi. Figure 5-1 describes the transform of the raw ECG


signal to a vector containing RRi.






























Figure 5-1. QRS detection algorithm

Noise/Baseline Wander

The difficulty in accurate R-wave or QRS complex detection is directly related to

the common noise characteristics existing in a raw ECG signal. Noise sources include,

but are not limited to: electrode motion, muscle movement, power-line interference,

baseline drift, and T-wave interference. Using a method similar to that described by

Jiapu Pan [24] we will attempt to improve the SNR of the raw ECG signal and detect the

QRS complexes. This process is split into two linear steps and one nonlinear

transformation. The first linear process requires filtering the signal with a bandpass

filter, which is split into a low-pass and high-pass filter. A low-pass filter with cutoff

frequency of approximately 11Hz is first applied the raw ECG signal to remove the low-

frequency baseline drift. In addition, the filtered ECG signal is sent through a high-pass

filter to remove muscle movement, powerline interference, and T-wave interference. The

cutoff frequency of the high-pass filter is approximately 5Hz. This yields a filtered ECG









signal (fECG) containing frequencies of 5-1 1Hz, which has been described to preserve

maximum information regarding the QRS complexes [24].

Low-pass filter:

Hz) Z-6 2
H ( 1)2

High-pass filter:

H(z) (- 132z-16+ -32)
1( z1)

Filtering with the Pan-Tompkins method described above, yielded only minor

improvements to the raw ECG signal. Problems with the above technique arise as a

result of the electrode placement differing from subject to subject. In some case, the

Pam-Tompkins algorithm removed substantial frequency components of the QRS

complex, thereby deteriorating the quality of the signal. Therefore, we need to replace

the low and high pass filters described above with a notch and bandpass filter. The

nature of the ECG signal calls for the use of an IIR filter. We will explore the

Butterworth and Chebyshev filters. Both are high order filter designs, which can be

realized by using simple first order stages cascaded together to achieve the desired order,

passband response, and cut-off frequency. The Butterworth filter yields no passband

ripple and a roll-off of -20dB per pole. However, the flatness of the passband response

comes at the expense of roll-off. The Chebyshev filter displays a much steeper roll-off,

but is characterized by a significant passband ripple. The Chebyshev filter thereby

optimizes roll-off at the expense of passband ripple. Due to the nature of the ECG signal

the roll-off frequency of the passband is not as important as the gain in the passband.

Therefore, we will use Butterworth filters for all filtering of the raw ECG signal.











Since 60Hz powerline interference is common among electrical signals, we need to

construct a notch filter to remove this noise. The notch filter used is a 5th order


Butterworth filter with center frequency at 60Hz. The magnitude and phase of this filter


is shown in Figure 5-2.


Notch Filter
50






.100

-150 -
0 10 20 30 40 50 60 70 80 90 100
Frequency (Hz)


.200 -
S -400


-ao
-1000 ---_
-1200 I i i
0 10 20 30 40 50 60 70 80 90 100
Frequency (Hz)



Figure 5-2. Magnitude and phase response of the 60Hz notch filter used to filter the raw
ECG signal.

The phase of this filter is essentially linear over the region of attenuation. After the raw


ECG signal is passed through the notch filter, it is passed through an 8th order bandpass


filter serving to remove baseline wander and high frequency noise; thereby improving the


SNR of the raw ECG signal. Both the notch and bandpass filters are implemented using


Butterworth filters. As illustrated in the Figure 5-3 below, the phase over the passband


region is approximately linear. Keeping the phase of the filter linear over the passband


prevents phase distortion in the resulting filtered ECG waveform.












Bandpass Filter

i--------- ---I----------- ------------- *
20




-40
II I I I I i
0 10 20 30 40 50 60 70 80 90 100
Frequency (Hz)

1500

S1000

500 .



-500
.500 --- I -- I -- I----------------1 --- I -
0 10 20 30 40 50 60 70 80 90 100
Frequency (Hz)



Figure 5-3. Magnitude and phase response of the bandpass filter used to filter the raw
ECG signal.

During the recording phase, we obtained ECG recordings that included significant


periods where the signal underwent high and low voltage clipping. The clipping resulted


in epochs of flat line, thereby making it impossible to detect subsequent QRS complexes.


For HRV analysis, a continuous ECG signal is needed. The undesired clipping resulted


in extensive signal losses that required considerable epochs of QRS interpolation. It was


later discovered that this problem could be overcome by setting the ECG module to filter


the signal before it was relayed to the laptop for text conversion. In addition, incorrect


electrode placement proved the monitors default selection of Lead II to be an errant


assumption. Lead selection was then delegated to the user before the acquisition program


was initialized. After realizing the need to select the "filter" option and correct lead, we


were able to attain signals that needed little or no secondary filtering and was not


necessary to bandpass filter the recordings. Powerline interference during the passage











from the monitor to the laptop was not completely alleviated and the 60Hz notch filter,


described above, was still applied. When the detection phase demonstrated a need for


secondary filtering, the bandpass Butterworth Filter, described above, was applied to the


signal to conclude the filtering stage of the detection algorithm.


Slope Detection

In parallel with the Pan-Tompkins algorithm, after filtering, the fECG signal is sent


through a differentiator yielding QRS complex slope information; this method results in a


differentiated version (dfECG) of the fECG. In accordance with Pan [24], we will use a


5-point transfer function to approximate the derivative.


H(z) =(1/8T)(-z2 -2z1 +2' +z2)


where T is the sampling period.


This transfer function is roughly linear in the region between dc and 100Hz. Therefore it


provides a quick approximation of the ideal derivative over the region of interest.


Differentiator
20 ---------F-- --I--------
10

0 /
-10


-30

0 10 20 30 40 50 60 70 80 90 100
Frequency (Hz)

-50

-100





-2500
0 10 20 30 40 50 60 70 60 90 100
Frequency (Hz)



Figure 5-4. Magnitude and phase response of the differentiator function









Following this step, the dfECG signal is passed through a nonlinear point-by-point

squaring operator yielding the sdfECG signal:

y(nT)= [x(nT)]2

This operation makes all data points positive and supplies a nonlinear amplification of the

signal at the higher frequencies. Finally, the sdfECG is smoothed using a ten point

averaging function, which produces a signal containing the power information of the

ECG waveform's slope, pECG. This reduces the presence of local minima in the sdfECG

signal. From this point, the method used for detection of QRS complexes deviates from

that used by Pan [24]. Since it is not necessary to perform QRS detection in an online

fashion, we have the luxury of determining thresholds using the entire information

contained in pECG signal. The threshold limit of the QRS slope detection is set in the

following fashion:

S= [max(dECG 2)+mean(dCG 2 )]/ 2

The selection of the current threshold (y) allows for detection of poorly defined QRS

complexes, while distinguishing the R-waves from occurrences of P and T-waves. The

threshold is then adjusted throughout the detection process. The following equations

illustrates how the threshold is updated:

77 = [max(dfCG (i: i + win))+mean(dfECG (i : i +win))]/2

where threshold parameter (rl) is updated in accordance with the current sample (i) over a

user-defined window (win)

S6=-y


;i y(g*3)









where 6 represents is the threshold difference and includes a memory constant (E) to

reduce high frequency noise interference during the update

Once this threshold has been set, it is now necessary to find maximum values

during each interval that the signal spends above threshold. The detection algorithm

searches the pECG until it reaches a y-value above the threshold. An array is

constructed, storing the waveform from this point until the y-value falls below the

threshold. The x-value corresponding to the maximum y-value inside this array is stored

as a possible QRS complex.

QRS = max(x[n,n + 1,n + 2, .,m]

where y[n] and y[m] are above the threshold, but y[n-1] and y[m+l] fall below the

threshold.


Figure 5-5. Visualization of QRS peak detection









These maximums, or peaks, define the occurrences of possible QRS complexes. If a

QRS complex is detected outside the expected range given below,

mean(RRi) < QRS(k) < (mean(RRi) A)
A

5. RR() = 0.375s (160bpm)
6. 1 7. QRS(k) is the detected QRS complex


the algorithm reverts back to the previous QRS complex, adjusts the threshold as follows,


/ threshold

Threshold is a user definable parameter (default = 1.5)

and searches for a new QRS complex. This searchback technique plays a crucial role in

preventing high frequency noise contamination, P-waves, and T-waves from being

marked as a QRS complex.

If a complex is not found within the expected range after 5 searchbacks, a

complex is interpolated as an average of the previous two intervals.

QRS(k) = [QRS(k 1) + QRS(k 2)/2

Subsequent detections are stored in a vector of increasing length until the algorithm has

operated on the entire pECG signal. In addition, the interpolated detections are stored in

a separate vector for further analysis. Units of these vectors are stored according to

sample numbers. The vectors are then divided by the sampling rate for unit conversion to

the time-domain. Finally, another vector is constructed, which represents a time-based

representation of the RRi in the following fashion:

RRi = {QRS(2)- QRSl), QRS3) QRS2),..., QRS(N) QRS(N 1)










where N = total number of QRS detections

Figure 5-6 illustrates the final results of the QRS detection algorithm where the

QRS complexes are marked in red. This figure also displays a linear interpolation of the

RRi waveform corresponding to the detected beats.




QRS Complexes Marked in Red
3500

3000

S2500-

S2000--

1500

1000 --
100 100.5 101 101.5 102 102.5 103 103.5 104
Time(s)
Beat Intervals


0.365


D 0 -355/
i =\ / /

0.35
0.35 5------------
100 100.5 101 101.5 102 102.5 103 103.5 104
Time (s)


Figure 5-6. Example of detected QRS complexes and beat intervals

The detection is offset from the peaks of the QRS complex due to the intrinsic phase

delays of the filtering process. As a result, the delays are consistent throughout each

signal and fall out after computing the RRi vector, as follows:

RRi= QRS(N)- ] [QRS(N -1)- rj}
= {QRS(N)-- QRS(N-1)+ ri
= {QRS(N) QRS(N 1)}









Outliers

Outliers are ill-defined byproducts of data acquisition systems. There is no

rigorous definition of an "outlier"; it is only generally conceived that selectively

eliminating inconvenient data points from analysis creates a more robust dataset.

Outliers are acknowledged to skew sample means, but tend to have less affect on the

median than the mean. For the purpose of this thesis, it is necessary to form a working

definition of an outlier.

An outlier is a data point that is an "unusual" observation or an extreme value in the
data set.

The lack of a universally accepted definition of an outlier inherently causes a deficiency

of objective mathematical processes for outlier recognition. Adding to the complication

of outlier detection, RRi datasets are time dependant vectors, rather than independent

samples. Therefore, it is essential to add to our working definition of an outlier.

A time dependant outlier is a data point that is an "unusual" observation or an
extreme value in the data set, which is not part of a trend in time.

Again, the definition of a "trend" is subjective and needs to be defined for the purposes of

this paper.

A trend is two or more consecutive data points, which move in the same general
direction within a given statistical range.

These definitions will be used to construct an algorithm to remove time dependant

outliers from the RRi datasets. False detections and missed detections result in possible

outliers in these datasets. The presence of outliers in the RRi dataset adversely effect

both time and frequency domain analysis of HRV. Possible skewed time domain

representations of HRV include the mean and standard deviation of RRi windows. In

addition, these outliers produce "ghost powers" in the high frequency range of the










spectrum. Not only does this increase the power in the high frequency bands, it skews

the power in the low frequency bands. Low frequency bands are of the foremost

importance during analysis of HRV in the frequency domain. This will become more

evident through discussion in Chapter 6.

Now that the importance of removing outliers has been established, we must

construct a mathematically sound algorithm for their removal. Three techniques are

explored as possible strategies for outlier detection: direct RRi analysis, slope analysis,

and differential analysis. For each of the aforementioned techniques, we will use standard

deviation as the exclusion criterion. Since literature is not conclusive for the distribution

of RRi datasets, the exclusion criterion is based on normal distribution of the data. The

figures below illustrate the RRi distribution and the normal distribution. Visually, the

normal distribution seems an accurate estimation of the RRi dataset. We will discuss the

validity of this assumption further in Chapter 6.


Distibuton of RRi
160 I



120

100

80

60






0.28 0.3 032 0.34 0,36 0.38


Figure 5-7. Histogram of RRi distribution from one subject.


0.4 0.42




























Figure 5-8. Standard deviations of the Normal Distribution

In a normal distribution, one standard deviation away from the mean in either

direction on the horizontal axis accounts for around 68 percent of the values in the

dataset. Two standard deviations away from the mean includes for roughly 95 percent of

the data. Finally, three standard deviations account for about 99 percent of the data [25].

Therefore, we will assume data outside of three standard deviations from the mean to be

statistically irrelevant and tag them as outliers unless they are part of a trend.

Direct RRi Analysis

The simplest method of removing outliers would be to examine the RRi dataset

directly and remove data that fall three standard deviations above or below the mean.

The following equation is used for outlier detection:

outliers = RRi(n)> mean(RRi)+ 3 std(RRij


for n = 1: length(RRi)











This method inadequately detects outliers and is not optimal to detect a trend, which

moves beyond the bounds of the acceptable range. In addition, the direct RRi analysis

method does not account for consecutive false or missed detection that exhibit a bouncing

effect around the actual QRS complex.






0.48

0.46

0.44

0.42

0.4

t 0.38

S0.36-

0.34

0.32

0.3-

0.28





Figure 5-9. Illustrates the bouncing effect and upper and lower bound symmetry
problems with using the direct RRi analysis method for tagging outliers.

The figure above demonstrates problems, which arise from direct RRi analysis

method for tagging outliers. The green lines represent the bounds for acceptable

datapoints. The bouncing effect is clearly present in this example. According to the

exclusion criteria, one of the datapoints would be tagged as an outlier, while the other

point would pass through the algorithm without being tagged. This figure also unveils

weaknesses with the computation of the mean. When the heart rate deviates to far from









the mean (marked by the black line), situations exist where the upper and lower bound

are not symmetrical about the data over a given window. Given all the abovementioned

weaknesses, it is necessary to explore a more robust technique to mark the presence of

an outlier.

Slope Analysis

In order to correct for the bouncing effect described above, slope information could

be used to isolate these occurrences and tag outliers. First, compute the derivative of the

RRi dataset, as follows:

aRRi = R t)

Then, tagging outliers using the same detection method as in the direct analysis:

outliers = 8RRi(n) > mean(8RRi)+ 3 std(RRi)

for n = 1: length(ORRi)

This algorithm adequately accounts for the removal of the bouncing effect, but we will

demonstrate that it is impossible to exclude trends from being tagged. Figure 5-10 is an

example of trend which seems to begin abruptly, but is sustained over subsequent

intervals. The slope algorithm solves the problems in accordance with the mean and

bounce effect, but introduces troubles in excluding trends from being tagged as outliers.

Over the course of this trend, only those datapoints whose slope exceeded the bound of

acceptance (marked by the green line) would be tagged. No mathematically simple

method exists to overcome the flaws in this algorithm. Merely modifying this algorithm

to remove the outliers and loop through the process again would serve to completely

eliminate the trend. Therefore, for a second time it is necessary to search for a more

robust algorithm to tag outliers.














RRi Plot
0.6


0.5


0.4


0.3


RRi Derivative Plot
0.15

0.1 -

0.05

0

-0.05

-0.1




Figure 5-10. Illustrates an RRi trend that would not be preserved using the slope analysis
method for tagging outliers.

Differential Analysis

To this point, three problems in the outlier detection process have been identified:

mean calculations that deteriorate the symmetry of the acceptance bound, the bouncing

effect, and the exclusion of trends. We introduce the differential analysis method as a

means, which circumvents all three of these problems. First, a difference vector is

constructed as follows:


RRidff(n) x(n) x(n 1)


for n =2 : m, where m = length(RRi)









RRid (1) = mean{RRi) RRi(l)

The end product is a vector, which is the same length as RRi. The resulting mean of the

difference vector is close to zero, which further implies the data is predominately

normally distributed. The difference vector also alleviates the first problem mentioned

above. Then, a vector is created using the same exclusion criterion discussed in the

previous two subsections. However, using this difference vector alone to tag outliers

outside the acceptable standard deviation range is not sufficient. A single outlier in a

dataset would produce two datapoints in this difference vector that would be tagged as

outliers (n and n+1). Being that it is not necessary to perform the outlier removal real-

time, we are able to create an additional vector, as follows:

RRi^ff (n)= x(n) x(n +1

for n = 1: m -1, where m = length(RRi)


RRidff (m) = mean(RRi)- RRi(m)

The same exclusion criterion is used to create a vector containing the outliers. As a

result of this computation, a single outlier in the dataset would produce two datapoints

tagged as outliers. Since taking the difference of RR intervals backwards through time

generates this vector, the two resulting datapoints tagged as outliers would be n and n-1.

Now, we compare the two difference vectors and note only those values for n, which

occur in both. These are the actual RRi datapoints that are separated from the previous

and subsequent datapoint by three standard deviations above or below the differential

mean. Figure 5-11 is a section of RRi data with trends that would normal be removed

using the Direct RRi Analysis method, described above, and a bounce effect.










Differential Analysis Outlier Removal


Figure 5-11. Outliers that have been tagged, using the differential method, are marked by
a green dot. Shows the efficiency of the method to preserve trend and remove
bouncing effects.

In figure above, the points that were tagged as outliers are denoted by green dots. It is

now evident that we have constructed a robust method to find and remove outliers, as

defined above.

Interpolation

The RRi vector described above is sufficient for time domain analysis of the HRV.

The RRi vector can also be converted to an instantaneous representation of the heart rate

(iBPM). However, there is a problem with using this vector to represent the HRV in the

frequency domain. When plotting the RRi on a time scale, the x-axis must include a

cumulative sum of all the preceding RRi plotted against the RRi or iBPM. This results in










an unequally sampled representation of the HRV. In order to apply a meaningful power

spectral representation of the HRV, the data must be evenly sampled at a known

frequency. Many papers using spectral analysis fail to address this topic and there are no

universal methods to solve this problem.

Resolving this issue of uneven sampling is essential for spectral analysis. Without

addressing this topic, may forms of spectral analysis applied to the RRi would produce

meaningless results. The choice of an interpolation method is very important, as it can

significantly compromise the integrity of the data. Therefore, we will explore four

methods of interpolating between successive data points of RRi:

* Linear Interpolation (LI)

* LaGrange Polynomial Interpolation (LPI)

* Cubic Spline (CS)

* Cubic Interpolation (CI)

LaGrange Polynormi Irtapola
0 38
0,375.





//


0.345[ -
J4- r /


TIme(s)


Figure 5-12. LaGrange Polynomial Interpolation









41




Linear Interpolation


--------------L- ------- -- L
1 2 3 4
Times)


5 6 7


Figure 5-13. Linear Interpolation


Cubic Spine
I I~ -


S 1 2 3 4
Times)


5 6 7


Figure 5-14. Cubic Spline Interpolation


*) 37


0365





S0 25S





n?F


n.,O


'A'







42



Cubic Interpolation
0 38

3.375 -

(137

0A





0,35 X
0 35 -4


0 34




0 1 2 3 4 5 6 7
Time(s)



Figure 5-15. Cubic Interpolation

The performance of the interpolation techniques is shown applied to a sample RRi

dataset marked by X's in the figures above. LI is continuous, but the slope changes at

the vertex points. LPI has an adjustable order of polynomial and reduces to LI when the

order equals one. As seen above, the LPI algorithm does not perform well at the

endpoints. CSI produces the smoothest results of all the interpolation methods, but

performs poorly if the data is highly non-uniform. CI results in both the interpolated data

and its derivative being continuous. The CI technique seems the most realistic

interpolation method because it does not force a sinusoidal interpolation. Therefore, we

will use CI to interpolate and resample RRi in preparation for spectral analysis. It is

necessary to resample the RRi at a frequency where the Nyquist rate is above the

frequency range of interest for spectral analysis, discussed further in Chapter 6. Also in






43


Chapter 6, spectral analysis techniques will be introduced that do not require resampling

of the RRi waveform prior examining HRV in the frequency domain.















CHAPTER 6
STATISTICAL ANALYSIS

Time Based

The simplest methods used to evaluate HRV are based in the time domain. Time

domain statistics can be divided into those derived from RRi or iHR and those derived

from differences in RRi, namely ARRi [26]. The following time-domain techniques will

be used:

* Mean RRi (mRRi)
* Median RRi (medRRi)
* Standard deviation of RR interval (SDRR)
* Poincare plots
* Quadrant analysis


First Order Statistics

Analysis of mRRi and medRRi yields some useful information about HRV. We

will use this information to analyze trends of the mRRi and medRRi with gestational age

and across genders. Also, the mRRi for the 45-s prestimulus period will be subtracted

from each heartbeat of the 45-s prestimulus and stimulus period. The result will be a

string of difference scores. The difference scores will be used to confirm that no

significant HRV trending has occurred during the prestimulus period, which may mask or

falsely demonstrate a COR. During the stimulus period, the difference scores will be

used to detect an orienting response. The mRRi is calculated in the following manner:

N
ZRRi(n)
mRRi = "
N









for N= number of RRi

To find the median, we must first sort the RR intervals by ascending values. Then, apply

the following formula to find medRRi:

medRRi = RRisted (N / 2)

for even values ofN


medRRi = RRsorted ([N -1]/ 2) + RRsorted ([N + 1]/2)
2

for odd values ofN

SDRR is highly dependant upon the length of the recording [26]. When using this

value as a comparison across datasets, we must be careful to uniformly define the length

of the recording. Therefore, we have kept all window lengths constant from subject-to-

subject and session-to-session. The variance (SDRR squared) is mathematically equal to

the total power of the spectrum; therefore SDRR will also be used to analyze trends

corresponding to increasing gestational age. SDRR is the primary first order statistical

analysis method we will use to discuss HRV in the time domain. SDRR is calculated in

the following manner:


Z(RRi(n) -mRRi)2
SDRR = -"
(N 1)

where N= number of RRi

Poincare Plots

In addition to numerical analysis in the time domain, the importance of geometric

representations, such as Poincare plots are useful tools for HRV analysis [26,27]. The

nonlinear Poincare plot produces a figure with each RRi plotted against the previous RRi.









According to Brennan et al., the plot provides summary information as well as detailed

beat-to-beat information on the behavior of the heart. The dispersion of points

perpendicular to the line of identity reflects the level of short-term variability [27].

In subjects with normal cardiac function, the geometry of Poincare plots tends to be

elliptical in nature. No literature pertaining to analysis of Poincare plots using an ellipse

were found, but we will explore an ellipse fitting method as a possible representation of

short-term and long-variability. MATLAB code written by Marcos Duarte [28] will be

used to fit an ellipse to the data. Duarte calculates the major and minor axis of the ellipse

using principal component analysis (PCA). The method results in 85.35% of the

datapoints lying inside of the ellipse. The datapoints that fall outside the ellipse can be

described by increasing the order of the ellipsoid, but it is only necessary to include the

first two eigenvalues when analyzing the short and long-term variability of the dataset.

PCA is a multivariate procedure, which rotates the data such that maximum

variabilities are projected onto the axes. Essentially through computing the eigenvalues

of the correlation matrix, a set of correlated variables are transformed into a set of

uncorrelated variables, which are ordered by reducing variability. The uncorrelated

variables are linear combinations of the original variables, and the last of these variables

can be removed with minimum loss of real data. The largest eigenvalue corresponds to

the principal component in the direction of greatest variance; the next largest eigenvalue

corresponds to the principal component in the perpendicular direction of next greatest

variance.

PCA can be viewed as a rotation of the existing axes to new positions in the space

defined by the original variables. In this new rotation, there will be no correlation







47


between the new variables defined by the rotation. The first new variable contains the

maximum amount of variation. The second new variable contains the maximum amount

of variation unexplained by, and orthogonal to, the first. The result is an ellipsoid in

multidimensional space and reduces to an ellipse for the two dimensional Poincare Plot.

When a Gaussian random vector has covariance matrix that is not diagonal, the

axes of the resulting ellipse are perpendicular to each other, but are not parallel to the

coordinate axes. As shown in the Figure 6-1, the principal components of the RR interval

Poincare representation result in an ellipse whose axes are rotated versions of the

coordinate axes. Therefore, as discussed in Chapter 5, the assumption that a Normal

distribution can describe RRi datasets is valid. The minor axis of the ellipse serves as an

interpretation of short-term variability; and the major axis serves as an index on long-

term variability.


Poincare Plot
0.45

0.44

0.43 *
0-2**
0.42


039,




0-38-

0.37

0.36
0,36 0.37 0.38 0.39 04 0.41 0.42 0.43 0,44 0,45
RR(n)(s)

Figure 6-1. Poincare plot representation of RRi. Since data points overlap, colors,
coordinated with the visible spectrum, are used to indicate magnitude at each
point.









Quadrant Analysis

Quadrant analysis is a graphical tool used to represent the number of sustained

increases, sustained decreases, and alternating changes in RRi. Two vectors are created

as follows:

ARRi(n) = RRi(n) RRi(n 1)

ARR(n +1)= RRi(n +1) -RRi(n)

Each difference value was then plotted against the previous difference, resulting in

the ARRi (n) versus ARRi (n+1) plot shown below [4]. The plot displays values

distributed over an area defined by four quadrants. Each quadrant indicates the direction

of two consecutive changes in interval length.








(-,+) (+,+)



A B ARR+1


C D



(-A -) (+, -)



ARR,


Figure 6-2. Defines the quadrants used in Quadrant Analysis.







49


We will use this analysis tool to examine sustained increases or decreases of RRi


and alternating RRi. The number of points in quadrants A and D measure high frequency


variation, or quick changes in RRi. Low frequency heart rate variation is characterized


by many sequential increases and decreases in interval length, which is indicated by


points in quadrants B and C. If changes in RRi occurred randomly and independently, all


quadrants would have an approximately equal number of data points. In addition, if the


heart rate tended to increase quickly and decrease slowly, more points would fall in


quadrants A and B than in C and D and visa versa. Figure 6-3 is an example of a


quadrant analysis plot extract from real RRi data.


Quadrant Analysis
0.3



0.2



0.1



0
-n


-0.1



-0.2



-0.3
-0.3 -0.2 -0.1 0 0.1 0.2 0.3
deltRR(n)



Figure 6-3. Example of Quadrant Analysis applied to an RRi dataset.









Frequency Based

While analytical methods were first restricted to the time domain, they have

quickly shifted to the frequency domain. Spectral analysis is now a common

representation of HRV. Transforming HRV to the frequency domain provides a means of

separating the parasympathetic and sympathetic responses of the ANS. Heart rate

fluctuations in the premature will be examined in the .00033-1Hz, while the frequencies

below this will be discarded to reduce the influence of slow trends or artifacts. Research

has suggested that the power spectral decomposition of the ECG yields frequency bands

essential to analysis of the ANS. These frequency bands are very inconsistent throughout

the literature referenced for this subject. This thesis will follow the frequency divisions

described by Stein et al. [29]:

* The ultra low frequency band (ULFB) <.0033Hz

* The very low frequency band (VLFB) 0.0033 0.04Hz

* The low frequency band (LFB) 0.04 0.15Hz

* The high frequency band (HFB) 0.15 0.4Hz

* The total power (TP) .0033 1Hz

The low frequency band of the power spectrum is composed of two peaks. The

mid-range peak (.09 .15Hz) corresponds to the Mayer waves, which have been related

to the baroreceptor reflex [3]. In addition, there exist a second peak between .02 .09Hz,

which has been linked to thermoregulatory fluctuations and variations in vasomotor tone.

The high-frequency band reflects the parasympathetic response, while the low frequency

band is influenced by the PNS and SNS [3]. For each frequency band listed above, we

will extract the normalized power over the described range. Also, the power ratio of the









high and low frequency bands has been assumed to reflect the sympathovagal balance

[3]. This power ratio is observed as follows:

LFP
SV
rto (LFP + HFP)


The SV ratio should provide us with insight into the ANS balance and should trend

downward in accordance with increasing gestational age. Lastly, a range just above the

high-frequency band contains information regarding the RSA, which is not represented

by a well-defined peak, but is usually centered about the .5-.6Hz range. We will explore

three methods of power spectral analysis in the following subsections:

* Fast Fourier Spectral Analysis
* Lomb-Scargle Spectral Analysis
* Empirical Mode Decomposition


Fast Fourier Spectral Analysis

The simplest and most commonly exploited technique for spectral analysis is the

Fast Fourier Transform (FFT). The FFT is commonly used because it is computationally

simple algorithm for spectral analysis. Fourier analysis decomposes a time-series into

global sinusoidal components with fixed amplitudes. As a result, Fourier spectral

analysis is limited to systems that are linear and strictly stationary. However, a high

number of variables affect the heart rate, which suggests that an RRi signal is highly

nonlinear and non-stationary. Most research pertaining to HRV utilizes the FFT to

describe to the spectral components of the RRi signal. We will utilize the FFT spectral

analysis method to obtain results, which can be compared with past and future research.

The FFT is computed using the internal FFT function provided by MATLAB.

Since the FFT method is commonly used in signal processing, we will not discuss this









method in detail, rather we will explain how we computed the transform. All RRi

windows for the RRi dataset (developmental, prestimulus, stimulus, and poststimulus) are

analyzed separately based on a N-point FFT, where N is equal to the total number of

point in the intervals. Often FFT windows are padded with zeros or with repeated copies

of the window. Neither of these techniques adds any information to the resulting FFT.

Padding only serves to increase the resolution of the transformation. Setting N to the

length to the window abolishes the need for padding. Each of the aforementioned

windows is extracted from the original recording based on cumulative time, and the

resulting windows are unevenly sampled signals. The cubic spline interpolation

technique, described in Chapter 5, is applied before computing the FFT and resampling

the interpolated signal at 4Hz. We chose 4Hz to prevent aliasing above of the Nyquist

frequency (2Hz) from falling into the frequency bands of interest. Finally, the resampled

signal is passed through the FFT algorithm and the DC component of the FFT

transformation vector is removed by dropping the first term. Results from this algorithm

can be found in the following chapter.

Lomb Spectral Analysis

In order to avoid the consequence of interpolation, we explore the Lomb method to

estimate the power spectral density (PSD) of unevenly sampled signals. Laguna et al.

concluded that for PSD of these unevenly sampled signals the Lomb method is more

suitable than FFT estimates requiring linear or cubic interpolation [30]. The Lomb

estimate avoids the low pass effect of resampling and avoids aliasing up to the mean

Nyquist frequency:

1
Nf =(2 mRR)









The Lomb method is based on the minimization of the squared differences between

the projection of the signal onto the basis function and the signal under study [30].

X(t )+ 6, = c(i)b,(t )

Minimizing the variance of sE (mean squared error), is synonomous to the minimization

of

N
L x (t,) c (i)b,(t, 2
n=1

which results in [30]

1N
cO(i-= )x(t,)b:(t,) k= b,(tn2
kn n=l

In addition, the signal power at index i of the transformation as described by Lomb

[31]:

N
PJ (i) =c(i)f x (t )b (t= k c(i) 2 22 where 4(i)= c(i)k
n=l

Setting the basis function to that of the Fourier transform

b, (t )= e 2'tn

Then,


k = e2ftn 2 = N
n=1

And

P(i) = PA)= 2f)

Lomb presented a simplification that permits one to solve for each coefficient

independently, known as the Lomb normalized periodogram. First, we solve for the

mean and variance [31]:









x= mRR
o 2 = SDRR)2

Then, we generate the Lomb normalized periodogram using the ith angular frequency

IN -2 )coN 2)sin(ot
(x(t) x)cos(o (t r,)) Y (x(t- x)sin (a), -r)
P(, = 2cos2 ( )) sin2, (n -))

n=l n=l

where r, is defined as,



Y1 sin(2 at)
r =- arctan 1-I
20, zcos(2atn)
Sn=l

MATLAB code written by David Glover is used to compute this Lomb normalized

periodogram [32]. We apply this method of spectral analysis to the identical windows

used to compute the Fourier PSD; and we will compare the results in the following

chapter.

Empirical Mode Decomposition

Approximations are the key to understanding complex natural phenomena and

often there is a significant advantage to these methods. However, science is continuing to

develop ways to describing complex system and behaviors. Why approximate the RRi

signal as a linear, stationary signal when tools exist to analyze nonstationary, nonlinear

systems? The answer is rhetorical because the only reason describes the simplicity of the

computation of the FFT. Given all the assumptions needed to compute the FFT, it is

necessary to search for a more reliable spectral transformation. We have already

observed the Lomb method, which still assumes the signal is linear and stationary. Now,









using a technique developed by Haung et al. [33], this thesis will compare the

effectiveness of the Empirical Mode Decomposition (EMD) as an analysis tool of HRV.

The goal of EMD is to decompose a time series into superposition of Intrinsic

Mode Functions (IMF). EMD uses a shifting process to produce IMFs, which enables

complicated data to be reduced into a form with defined instantaneous frequencies. The

IMFs has the appearance of a generalized Fourier analysis, but with variable amplitudes

and frequencies. Haung et al. describes EMD as the first local and adaptive method in

frequency time analysis. Furthermore, Haung suggests that the advantage of EMD and

Hilbert spectral analysis is that it gives the best-fit local sine or cosine form to the local

data. The frequency resolution for any point is uniformly defined by the local phase

method and is especially effective in extracting low frequencies oscillations [33]. Since

the HRV is characterized by low frequency oscillations, this method should produce

informative frequency domain representation of HRV.

Computation of the EMD algorithm is done in MATLAB using modified code

originally written by G. Rilling [34]. The EMD algorithm [33] coded by Rilling begins

with the sifting process. First, all local extrema of the original signal are identified. The

local maxima are then connected via a cubic spline and the procedure is repeated for the

local minima, respectively forming the upper and lower envelopes. Then, we find the

first component of the sifting process [33], h1.

h, = X(t) = RRi(t)
h, = h0 m,

where m, is the mean of the envelope

The shift process is repeated k times, as follows, until h,, is an IMF (c,).









hik = hl(k-) k
C, = hlk

Then, a residual is found

r = X(t) -c,

Finally the IMF procedure is repeated n times until the residual, r, is a trend or constant.

rn = h, ) c(n+ )

It is then necessary to determine if all the calculated IMF are actually orthogonal to one

another, as the theory suggests. Huang suggests using the follow equation as a check for

orthogonality [33]:


C +C2
10fg -c c2+c2 0 for all IMFs
f g

In order to visualize the result of EMD, we must find the Hilbert transforms of the

IMF components and represent the RRi signal in terms of time, frequency, and amplitude.

We compute the Hilbert transforms using the Cauchy principal value integral to find the

magnitude, a (t), and phase, (t), and instantaneous frequencies, CO [33].



j t td

Z (t) = X (t)+ iY, (t)= a (t)e' (t)

a, (t) = 2X (t) + Y, (t)

Yc (t)
,j (t)= arctan YJ


ast

Finally, we can express the RRi dataset as follows










RRi(t) = a,(t)exp if o, (t\t
j=1

Cleary, the equation above represents the generalized Fourier expansion, but the

amplitude and instantaneous frequency are variable with respect to time [33]. At this

point, we will use the EMD method to provide us with a visual representation of the

power spectrum changes over a prestimulus -* stimulus -* poststimulus window. In

accordance with the description of a COR, detailed in Chapter 2, we should see

significant shift in the magnitude of frequency spectrum over this window.

Statistical Analysis of Parameters

Statistical analysis of the parameters extracted above is performed using a

technique for longitudinal data analysis, called mixed general liner models (MixMod) or

random regression models [35]. The MixMod model allows the concurrent identification

of both group and individual patterns in longitudinal sets with covariates that vary over

time, inconsistently timed data, irregularly timed data, and randomly missed values.

Given the vulnerable state of the infants in the NICU, we must perform test sessions in

accordance with the health status of the infant. These stability considerations can result

in the testing schedules varying from subject to subject and session to session.

Fortunately, problems arising from this effect are alleviated using the MixMod analysis.

In addition, subjects are often transferred or released before reaching 34 weeks GA. The

MixMod model allows us to include the data from these subjects, thereby reducing the

amount of data, which would otherwise be lost. Strengths of the MixMod model include

the acceptance of multiple characteristics and the exploration of patterns across subjects

and groups. Using multiple characteristics, MixMod leaves the statistical correlation and

inclusive of parameters up to the investigator [35]. All MixMod analysis is performed by






58


Dr. Kruger using readily available SAS Proc Mixed software. The results of the MixMod

analysis can be found in the Clinical Results section of Chapter 7.















CHAPTER 7
DISCUSSION AND RESULTS

Engineering Results

We will explore the results of the signal-processing phase of this thesis in

accordance with ECG recordings from one subject. The example subject is not intended

to prove the clinical significance of the hypothesis described in Chapter 3. For clinical

evaluation of aforementioned hypotheses, please refer to the "Clinical Results"

subsection of this chapter.

QRS Detection Algorithm

The following table is a representation of the performance of the QRS detection

algorithm, detailed in Chapter 5. The percent of QRS complexes is computed, as follows,

D det sections
QR detected (# det ec on
Sdetectnons missed

where the interpolated detections are tagged as missed detections and are stored in a

separate array to determine the QRS detection ratio defined above. The interpolations are

a result of no detected QRS complexes falling within the user defined expectancy range,

described in Chapter 5.

Table 7-1 clearly illustrates that on average the algorithm only misses 0.85% of the

QRS complexes in the signal. This detection algorithm is a robust system, which can

effectively detected QRS complexes in the presence of noise and muscle artifact and

produce reliable RRi datasets. Beyond the detection process, we improve quality of the










RRi signal through the implementation of the outliers removal algorithm designed in

Chapter 5.

Table 7-1. Shows the performance of the QRS detection algorithm


Patient


Recording Number QRS detections (%)


3 12130842 99.25
3 12200930 98.8
3 12130842 98.65
3 12060959 93.73
3 12020916 100
3 11221027 99.94
3 11151124 95.11
3 11120913 100
5 2191432 99.11
5 1311022 97.94
5 2071118 99.86
5 2160627 99.36
5 1241004 98.09
5 1211028 99.83
6 3110940 99.88
6 2281027 99.77
6 2211035 99.78
6 2160711 99.94
6 2050923 99.94
6 2071032 99.24
7 3281006 99.67
7 3210950 99.91
7 3141005 100
7 2281002 99.86
7 3071003 99.61
7 2190951 99.92
7 2211017 99.63
Average: 99.14148148


HRV Plots and Statistics

Scientific findings are based on equally important qualitative and quantitative

measures. Quantitative methods observe distinguishing characteristics and elemental

properties resulting in numerical comparative analyses and statistical analyses. For each

recording, an excel file is created, which contains all the pertinent quantitative measures


for each window:









* mRR

* medRR

* SDRR

* Number of sustained increase and decreases in HR derived through Quadrant
Analysis

* The length, width, and (W/L) ratio of the PCA fit elliptical representation of the
Poincare plot

* ULF, VLF, LF, and HF power of the Lomb PSD

* TP and SVrao of Lomb PSD

* ULF, VLF, LF, and HF power of the Fourier PSD

* TP and SVroao of Fourier PSD

These quantitative measures are used for MixMod analysis in the "Clinical Results"

section.

Qualitative measures are those associated with interpretative approaches, from the

investigators point of view, rather than discrete observations. While qualitative

methodologies can lead to biased results, there is no doubt that they should be included in

medical research. The power of visualization and quick interpretation of data possessed

by humans cannot be replaced by quantitative measures. Therefore, for each recording,

six plots are created for qualitative analysis:

* HR with outlier removal

* Quadrant analysis

* Poincare

* Fourier PSD

* Lomb PSD

* Hilbert-Haung spectrum of EMD







62


HR Plot



160

A 150


0 50 100 150 200
160

B 2 150-

140

0 5 10 15 20 25 30 35 40 45
160 ..-

) 150-
C)a IL
140

0 5 10 15 20 25 30 35 40


D) i 150-
140

0 5 10 15 20 25 30 35 40 45
Time (s)


Figure 7-1. HR plots of example RRi data. From top to bottom they represent: the (a)
developmental, (b) prestimulus, (c) stimulus, and (d) poststimulus periods.




The HR plot shown above illustrates the original RRi (blue) waveform displayed

as instantaneous heart rates and the resulting RRi (red) after outliers have been removed.

The figure above is divided into four plots representing the four windows of analysis.

Plot (b) illustrates the removal of an outlier, which causes a high frequency spike in the

signal. Plot (a) demonstrated the algorithm efficiency of preserving trends in RR

intervals. Figure 7-1 is used to visualize cardiac accelerations and decelerations,

quantitatively described through Quadrant Analysis.








63



Quadrant Plot


The Quadrant plot shown in Figure 7-2, below, is organized in the same manner as


Figure 7-1 with each plot representing a particular analysis window. Quadrant Analysis


is a more useful tool for quantitative analysis, but the visual representation can provide


insight into the amount of high and low frequency variability. The excel file described


above contains some useful information regarding this plot that will later be used for


MixMod analysis.


0.





-0.

0.
s,

A) s

a,



-0.

0.
B) -


-0.

0.


C) S
-

-D -0.


Quadrant Analysis
05 I, I, ,I,-,-,
05



0
05 L L L L L---
-0.05 -0.04 -0.03 -0.02 -0.01 0 0.01 0.02 0.03 0.04 0.C
05


0


05 L L -
-0.05 -0.04 -0.03 -0.02 -0.01 0 0.01 0.02 0.03 0.04 0.C
05





05
0.05 -0.04 -0.03 -0.02 -0.01 0 0.01 0.02 0.03 0.04 0.C
-0.05 -0.04 -0.03 -0.02 -0.01 0 0.01 0.02 0.03 0.04 0.C


0.05
-)

D) S 0
0
"o -0.05------


0
deltRR(n)


Figure 7-2. Quadrant Analysis plots of example RRi data. From top to bottom they
represent: the (a) developmental, (b) prestimulus, (c) stimulus, and (d)
poststimulus periods.











Poincare Plot

The Poincare plot is a valuable tool for both quantitative and qualitative analysis.


The Poincare method of RRi interpretation is observed to prove normal cardiac function


and analytically evaluated by the fitted ellipse to described short and long-term HRV.


0.46
u 0.44
0.42
A) +
S0.4
of 0.38
0.36
0.
0.46
0.44
B) 0.42
S 0.4
rO
of 0.38
0.36
0.
0.46
0.44
C) 0.42
S 0.4
0.38
0.36
0.
0.46
0.44
D) 0.42
S 0.4
of 0.38
0.36
0.


Poincare Plot

+-



36 0.37 0.38 0.39 0.4 0.41 0.42 0.43 0.44 0.45 0.'






36 0.37 0.38 0.39 0.4 0.41 0.42 0.43 0.44 0.45 0.,


36



36


0.37 0.38 0.39 0.4 0.41
RR(n)(s)


0.42 0.43 0.44 0.45 0.46


Figure 7-3. Poincare plots of example RRi data. From top to bottom they represent: the
(a) developmental, (b) prestimulus, (c) stimulus, and (d) poststimulus periods.



Again the magnitude corresponding of a given point is represented by the basic


colors of the visible light spectrum with red being the next to least frequent (black is the


least) and violet being the most frequent. Plot (a) represents the dataset with the most


long-term variability or low-frequency changes inferred by the length of the major axis of


the ellipse. This is expected because plot (a) is a 240s window, while plots (a,b,c) are 45s










windows. We can also see that during the stimulus period the major and minor axis of

the ellipse become significantly longer, perhaps signifying an increase the LF power, HF

power, and total power of the HRV spectrum. We will next look at the Fourier and Lomb

PSD to confirm this hypothesis.

Fourier and Lomb Plots

The Fourier and Lomb PSD plots essentially describe the same information;

therefore we will observe them together.


0.03

A) 0.02
0.01
0

0.06
B) 0.04
0.02
0

0.06
C) 0.04
0.02


FFT Power Spectrum

--



0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1





0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
I-


0.(
D) 0.(
0.(


0.5
Frequency (Hz)


Figure 7-4. Fourier PSD plots of example RRi data. From top to bottom they represent:
the (a) developmental, (b) prestimulus, (c) stimulus, and (d) poststimulus
periods.












Lomb Power Spectrum
I I I


100-

A) 50


0 1 I -/L I I I I I I I I
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0

0

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
5


10-

5

0
0 0.1 0.2


0.4 0.5 0.6 0.7 0.8 0.9 1


t 0.5
Frequency (Hz)


Figure 7-5. Lomb PSD plots of example RRi data. From top to bottom they represent:
the (a) developmental, (b) prestimulus, (c) stimulus, and (d) poststimulus
periods.



First, we will address the observational hypothesis from the Poincare Plot section.


Inspecting the Fourier and Lomb stimulus plots, we note that the area under the curve


increases from the prestimulus period to the stimulus period, which means there is an


increase in total power. Also, power increases in the LF and HF bands correspond to


increases in the length of the major and minor axes of the ellipse, respectively. In


conclusion, through qualitative analysis, we have shown that the Poincare plot is a time


domain representation of the PSD.


Now we will compare the PSD representations of the Fourier and Lomb


Spectrum. The Fourier and Lomb plots are comparable over the developmental window,


1









but the Fourier plot is choppier than the Lomb plot over the windows corresponding to

the COR. The choppy Fourier plot is a result of the low frequency resolution of the N-

point Fourier transformation. Therefore, the frequency resolution is lower for the short

windows analyzed in plots b, c, and d. However, the Lomb computation allows the user

to set the frequency resolution. For the purpose of obtaining a smooth transformation, we

have set the Lomb PSD resolution to 0.001Hz. To increase the frequency resolution of

the FFT, we must pad the data before performing the Fourier transform. For reasons

discussed in Chapter 6, we have refrained from doing so. The apparent equivalent Lomb

and Fourier PSD plots for the developmental window suggest that the Lomb method is at

least as successful as the Fourier for PSD estimation. For clinical interpretation, we will

use the Lomb Spectrum when computing the cumulative power in the frequency bands of

interest.

EMD Plots

At this point, we use the Hilbert-Haung Spectrum (HHS) only as a qualitative

analysis tool. If the clinical findings indicate the usefulness of this time-frequency

domain representation of HRV, we will expand to use EMD as a quantitative analysis

tool. Figure 7-6 is the HHS of the example RRi dataset used in this section. Black lines

separate the prestimulus, stimulus, and poststimulus periods. We can relate the HHS

power shift of the stimulus period to those described by the Poincare and PSD

assessments of HRV. All four representations are in agreement that during the stimulus

period the LF power, HF power, and TP increase. However, the HHS shows us that this

increase occurs just after the rhyme begins only to tail-off significantly shortly thereafter.

Initially the EMD seems to be a robust method to describe the HRV characteristics of the

COR with respect to time. However, since this method has not often been used for









clinical interpretation, the inclusion of the method in the analysis of the hypothesized

COR will be left up to the PI.


6b
Time


Figure 7-6. Hilbert-Haung spectrum of the COR windows of an example RRi dataset.

Clinical Results

At the time of submission, we only have four complete datasets (recordings for the

28-34 weeks GA) and one incomplete dataset (recordings for 28-33 weeks GA) that can

be used for clinical interpretation. The current sample size (n=5) is well below the

minimum sample size (n=28) described in the design of this study [7]. Therefore, we will

not be able to perform MixMod analysis of the data at this time. Instead, we will use a

repeated measures ANOVA analysis of the PSD. Repeated measure ANOVA allows us

to compare the variance caused by the independent variables to a more accurate error

term, which includes the variance effects of individual subjects. Thus, this method










requires fewer participants to yield an adequate interpretation of the data than the

MixMod method.

Early analysis based on the Poincare plots, Quadrant Analysis, and EMD will not

be performed due to the lack of comparable literature. We will exploit the

aforementioned methods upon completion of the study, so the PSD results can be used to

confirm the clinical significance of these models. The repeated measure ANOVA model

returned only two independent variables (HF power and SV power), which passed the

equal variance test. We will first examine results of the repeated measures ANOVA

analysis of the SV power defined in Chapter 6.

Table 7-2. Results of the repeated measures ANOVA with independent variable (SV
power).


Session n p a
1 5 0.807 0.0857
2 5 0.729 0.089
3 5 0.846 0.121
4 5 0.878 0.0501
5 5 0.742 0.239
6 5 0.866 0.0836
7 4 0.613 0.186


SV Power

1.2


S0.8 --\-

0.6
| 0.4
0.2
0 ------------------------
1 2 3 4 5 6 7
Session

Figure 7-7. Plot of mean SV power with standard deviation bars across sessions










Now we examine repeated measures ANOVA analysis of the HF power, also defined in

Chapter 6.

Table 7-3. Results of the repeated measures ANOVA with independent variable (HF
power).


Session n p a
1 5 68.2 25.925
2 5 133.6 144.921
3 5 48.46 26.837
4 5 41.32 23.877
5 5 178.1 172.465
6 5 47.22 58.952
7 4 150.75 126.818


HF Power


400
350
300
250
200
150
100
50
0
-50


S7
1) 13 A r_ n


Session


Figure 7-8. Plot of mean HF power with standard deviation bars across sessions

While Figure 7-7 shows a possible non-linear trend with respect to GA, Figure 7-

8 does not elicit any significant trend in time. First, the small sample size doesn't allow

for the development of a trend. Secondly, there exist many exogenous variables, which

conceal a developing trend corresponding to GA. These variables include, but are not

limited to:


\J r-









* Drug interventions and interactions

* Respiratory support interventions

* Physiological discrepancies in ANS maturation

* Changes in NICU environment

* Incorrect evaluations in GA

Patient logs will be analyzed following the completion of the study for drug

interventions. The PI will attempt to parallel the interventions with the recordings that

contribute most to the standard deviation of each variable. It is known that certain drugs

inhibit or even block parasympathetic and sympathetic responses. Furthermore, many of

our subjects are on multiple drugs whose interactions with respect to the ANS are not

known.

Many subjects intermittently require respiratory interventions such as continuous

positive airway pressure (CPAP). A closed-loop study showed that adequate application

of long-term CPAP therapy produced substantial alterations in the major physiological

mechanisms that influence HRV and blood pressure variability [36]. Based upon

preliminary results, see Figures 7-7 and 7-8, final results need to include an analysis of

how drug and respiratory interventions hinder the ability to identify trends in the ANS

landscape. The significance of all exogenous variables mentioned above will emerge as

the number of subject in the dataset increases. Unfortunately, these results are not

included in this thesis, but we refer you to further publications by the PI and the author of

this paper. For a complete time and frequency domain analysis of a randomly selected

subject, please refer to Appendix A and B, respectively.














CHAPTER 8
CONCLUSION

We have successfully modified code written by HP to communicate with the

Agilent/HP NICU monitor. This program records the ECG, respiratory rate, and rhyme

state in a text file at a 500 Hz sampling rate. The text file can be read into MATLAB for

QRS detection and HRV analysis. In MATLAB, we have created an algorithm to

perform accurate detection of QRS complexes. The average percentage of missed

detections is approximately 0.8%, which we feel represents a near optimal detection

algorithm. For those QRS complexes that are undetectable, we have constructed an

interpolation method to replace the missing data. Using our novel method of tagging

outliers, we are able to remove outliers, but preserve trends across time.

We have found that the most useful measurement of HRV in the time-domain is

through the analysis of Poincare plots. Fitting an ellipse to the Poincare representation of

the RRi is usefully in relating time-domain analysis to the frequency domain. Currently,

most spectral representations described in literature use the Fourier transform. We have

found that the Lomb normalized periodogram is a better representation of the power

spectrum of RRi waveforms. The Lomb algorithm avoids the negative effects of

interpolation and resampling required for FFT computation. Also, Empirical Mode

Decomposition is a favorable method for measuring changes in the frequency spectrum

across time. The EMD spectrum preserves more information regarding shifts in the

frequency spectrum during the stimulus period than either the Lomb or Fourier spectrum.

EMD has been proven to be a useful analysis tool to depict a COR.






73


We have discussed the importance of both qualitative and quantitative measures

of developmental changes in HRV and in identifying a COR; we have extracted

numerous features using time and frequency domain HRV analyses, which will be used

to analyze and test the hypotheses of the clinical study still in progress. We feel we have

completed the design phase of the study and need only to increase our sample size. As

the sample size increases, the clinical significance of each feature will be evaluated and

we will move to interpret our findings and test our hypotheses.














CHAPTER 9
FUTURE WORK

A need exists in the NICU and beyond to determine the state of development of the

ANS. Before an infant is discharged from the NICU, it is important to determine their

ability to perform certain life support tasks. In Chapter 2, these life-supporting tasks

were linked to the maturity of the ANS, specifically the PNS. The list below is compiled

from the 1998 suggested guidelines for neonatal discharge established by the American

Academy of Pediatrics (AAP) [37]. In the judgment of the responsible physician there

has been:

* A sustained pattern of weight gain of sufficient duration.

* Adequate maintenance of normal body temperature with the infant fully clothed in
an open bed with normal ambient temperature (240C to 250C).

* Competent suckle feeding, breast or bottle, without cardiorespiratory compromise.

* Physiologically mature and stable cardiorespiratory function of sufficient duration.

* Appropriate immunizations have been administered.

* Appropriate metabolic screening has been performed.

* Hematologic status has been assessed and appropriate therapy instituted as
indicated.

* Nutritional risks have been assessed and therapy and dietary modification instituted
as indicated.

* Sensorineural assessments, hearing and funduscopy, have been completed as
indicated.

* Review of hospital course has been completed, unresolved medical problems
identified, and plans for treatment instituted as indicated.









The AAP has established these guidelines, which include subjective observations of

cardiorespiratory function and sensorineural assessments. It is our hypothesis that

cardiac function, as well as, neural assessments can be objectively realized using an index

of ANS maturation. We will create this index of ANS functionality based on HRV

comparisons with mature young adults. Below, we suggest a pilot study, which shall

provide evidence that an index of maturation can be realized.

Record Baseline Adult HRV Data

In order to establish a baseline for a mature ANS with a synonymously high vagal

tone, it is important to analyze the HRV of the young, healthy adult. ECG recordings

will be obtained for a sample of 28 male and female adults meeting the following criteria:

* Between the ages of 18-30

* No history of congestive heart failure, myocardial infarction, coronary artery
disease, head trauma, or diabetes

* Normal sinus ECG rhythm


The length of the recordings should be equal to those used in the study of the fetus. This

is important we comparing both time and frequency domain features of HRV and the

COR.

Each subject should be oriented in the supine position to ensure consistency with

ECG recordings of the neonates. Before recording, all subjects should remain in this

position without auditory and visual simulation, until they are verified to be in REM or

paradoxical sleep. Consistency with the neonatal recording described in the paper is a

significant issue for accurate interpretation of HRV data and the adult period of REM

sleep corresponds to the neonatal state of "active sleep" [38]. Siegel suggests that the

phasic motor activation that characterizes REM sleep in the adult is also present in the









neonate. The amplitude of the twitching, described during REM sleep, is more intense in

neonates than in the adult, but a developmental continuity can be observed between

"active sleep" in the neonate and REM sleep in the adult.

Both active sleep in the neonate and REM sleep in the adult can be defined by

purely behavioral criteria [38]. Therefore, the recordings should be taken in the presence

of a medical professional qualified to identify phases of paradoxical sleep. An auditory

stimulation that is widely familiar and contains no political or nationalistic propaganda

should be used to prevent to prevent undesirable neural responses due to stress. The ECG

recordings during the auditory stimulation can later be used to compare the COR of the

neonate and adult.

Creating the Index of ANS Maturation

We suggest using the EMD method, detailed in Chapter 6, to compare the COR of

the HRV groupings described in the previous subsection with the mature adult. The

EMD data provides a time-based representation of the COR. The examiner should

compare differences in the COR of the adult and neonate to determine an efficient

method for relating these changes. We suggest exploring the use of power-ratios, neural

networks, windowing functions, etc. as possible criteria to represent the EMD data in the

form of an index. In addition, the development data can be compared using PCA and a

neural network to classify the Lomb spectrum into different level of ANS maturation.






















APPENDIX A

TIME-DOMAIN HRV ANALYSIS


Week One




155 --


3 25 30 35 40
Time (s)


( 0 4

0.35
0,37 0-38 0.39


0.37 0.38 0.39
0,44
S0.42 -
S 0.4
n 0.38
0 37 038 0.39


Poincare Plot



* *


0.4 0.41 042 0.43 0.44


0.4 0.41 042 0.43 0.44






0.4 0.41 042 0.43 0.44
RR(nXs)


5 10








78



Week Two


0 50 100


150


200


0 5 10 15 20 25 30 35 40 45
150

5 145 -

140',
A A~ /


Time (s)


Poincare Plot
0.46 -
0.44 *
0.42- -
S0.4 -c- -
0.38
0.38 0.39 0.4 0.41 0.42 0.43 0.44 0.45 0.46 0.47

0.46 -
0.44 -
0 44
S0.42 --
o 0.4 -.----
0.38
0.38 039 0.4 0.41 0.42 0.43 0.44 0.45 046 047

0.46 -
0.44- -
0.42 -* t
o4 04
0.38
0.38 039 0.4 0.41 0.42 0.43 0.44 0.45 046 047

S0.46 -
0.44-2
S0.42 t
W 0.4
0.38
0.38 0.39 0.4 0.41 0.42 0.43 0.44 0.45 0.46 0.47
RR(n)(s)








79



Week Three


0 5 10 15 20 25 30 35 40


0 5 10 15 20 25 30 35 40


0 5 10 15 20 25 30 35 40 45
Time (s)


Poincare Plot







0.31 0.32 0.33 0.34 0.35 0.36 0.37 0.38







0.31 0.32 0.33 0.34 0.35 0.36 0.37 0.38







0.31 0.32 0.33 0.34 0.35 0.36 0.37 038


0.31 0.32 0.33 0.34 0.35 0.36 0.37 0.38
RR(n)(s)


0.38
. 0.36
+ 0.34
j 0.32
0.3
0.3
0.38
.. 0.36
+ 0.34
f 0.32
0.3
0.3
0.38
. 0.36
S0.34
r 0.32
0.3
0.3
0.38
. 0.36
0.34
of 0.32
0.3
0.3


-





-








80



Week Four


50 100 150 200


182
180
I 178
- 176
m 174-
172
0
182
180
r 178
0 176
m 174
172
0
182
180
r 178
a 176
m 174-
172
0
182
180
1 178
a0 176
m 174-
172
0





0.38
0.36
S0.34
"f 0.32
0.3
0.3
0.38
S0.36
0.34
0.32
0.3
0.3
0.38
0.36
S0.34
If 0.32
0.3
0.3
0.38
." 0.36
g 0.36-
0.34
"i 0.32
0.3
0.3


5 10 15 20 25 30 35 40


5 10 15 20 25 30 35 40
Time (s)


Poincare Plot







0.31 0.32 0.33 0.34 0.35 0.36 0.37 0.38


0.31 0.32 0.33 0.34 0.35 0.36 0.37 0.38

0.31 0.32 0.33 0.34 0.35 0.36 0.37 0.38


0.31 0.32 0.33 0.34 0.35 0.36 0.37 0.38
RR(n)(s)


5 10 15 20 25 30 35 40








81



Week Five


180

S170
m 160
0 160


170
0..
m 160

150

180

170
0-..
m 160

150

180

170

m 160


Poincare Plot
0.4
0.38- .
+ 0.36 t-. *
S0.34 -
0.32
0.32 0.33 0.34 0.35 0.36 0.37 0.38 0.39 0.4
0.4 p l?.ii
S0.38-
0.36 "
' 0.34-
0.32
0.32 0.33 0.34 0.35 0.36 0.37 0.38 0.39 0.4
0.4 pp;
S0.38-
S0.36 -
O 0.34
0.32 I I
0.32 0.33 0.34 0.35 0.36 0.37 0.38 0.39 0.4
0.4 p

X 0.38

0.34
0.32
0.32 0.33 0.34 0.35 0.36 0.37 0.38 0.39 0.4
RR(n)(s)








82



Week Six




180

S170 r s

160

0 50 100 150 200
180 Tim

170

160

0 5 10 15 20 25 30 35 40
180

S170

160

0 5 10 15 20 25 30 35 40
180

170-

160

0 5 10 15 20 25 30 35 40




Poincare Plot
0.38 -

0.36 *-

" 0.34 -

0.32
0.32 0.33 0.34 0.35 0.36 0.37 0.38
0.38

0.36 -

S0.34

0.32I
0.32 0.33 0.34 0.35 0.36 0.37 0.38
0.38 PP'

0.36-

0.34-

0 .3 2 L L
0.32 033 0.34 0.35 0.36 0.37 038
0.38

0.36

' 0.34

0.32 I
0.3 0.31 0.32 0.33 0.34 0.35 0.36 0.37 0.38 0.39 0.4
RR(n)(s)








83



Week Seven


1I-(U- Ir -' I

2 160o

150

0 50 100 150 200
170- i ,(,

160 55 A

150

0 5 10 15 20 25 30 35 40 45


0 5 10 15 20 25 30 35 40
170 Tima


0 5 10 15 20 25
Time (s)


30 35 40 45


Poincare Plot


--0.42
0.4
S0.38
a 0.36
aO 0.34
0.32
0.:

S0.42
0.4
0.38
" 0.36
n, 0.34
0.32
0.:

-0.42
0.4
0.38
a 0.36
Of 0.34


w 0.42
S0.4
+0.38
- 0.36
rO 0.34
0.32
0.32


0.34 0.36


0.38
RR(n)(s)


0.42 0.44



















APPENDIX B
FREQUENCY-DOMAIN HRV ANALYSIS


Week One


Lomb Power Spectrum
40 II------ I I -------


40
20


0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
20


10


0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
60

40

20 -

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9


Frequency (Hz)


0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9













Week Two



Lomb Power Spectrum
100


50



0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1


30

20

10

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9


20 -

10

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

Frequency (Hz)



Week Three




150



50









S 0.2 0.3 0.4 0.5 0. 0.7 08 0.9
60 r r i I i



20

Cl 02 03 0C4 Cb Cr3 06 0 C.S '


0.1 0.2 03 0.4 0.5 0.6 0.7 08 0,9 1
Frequency (Hz)








86



Week Four



1501





0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 08 09 1
30

20


I0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
40



20

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
30


20 I I
20


10
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Frequency (Hz)


Week Five



150

100

50

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
40 1 1



20__ I

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
30 ,

20 -

10

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
40




0 Ii
20

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Frequency (Hz)








87



Week Six



60 Lqmb Powr Spoectrurm-n


-' I ,


0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
20"





0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
20





0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
20 I I




0. --
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Frequency (Hz)



Week Seven


Lomb Power Spectrum
100




0 ^ ^ I I-- ,4,^ I A- I--.--------------
0 0.1 02 03 0.4 0.5 0.6 0.7 08 09 1
15

10

5

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1




10




20

10
30 i 0I 0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
30 -----------------------------------


Frequency (Hz)















LIST OF REFERENCES


1. BLISS, Causes of premature birth (2003),
http://www.bliss.org.uk/support/causes.asp, Accessed 4/4/2003.

2. Thomas, K., Causes of premature birth (2001),
http://www.kerri.thomas.btintemet.co.uk/causes.htm, Accessed 4/4/2003.

3. Chatow, U., Davidson, S., Reichman, B. & Akselrod, S. (1995). Development and
maturation of theautonomic nervous system in premature and term infants using
spectral analysis of heart rate fluctuations. Pediatric Research, 37, 294-302.

4. Stark, R., Myers, M., Daniel, S.,Garland, M., & Kim, Y. (1999). Gestational age
related changes in cardiac dynamics of the fetal baboon. Early Human
Development, 53, 219-237.

5. Boon, R., Learning Discoveries Psychological Services (2002),
http://home.iprimus.com.au/rboon/HeartRhythmsandHRV.htm, Accessed
1/12/2003.

6. Groome, L., Mooney, D., Holland, S., Smith, L., Atterbury, J., & Loizou, P.
(1999). High Vagal Tone is Associated with More Effiecient Regulation of
Homeostasis in Low-Risk Human Fetuses. Dev. Psychophysiology, 36, 25-34.

7. Krueger, C. (2002). Fetal heart rate variability and learning in the 28-34 week fetus.
(Doctoral dissertation, University of North Carolina at Chapel Hill, 2001).
Dissertation Abstracts International, 62, 11B.

8. Porges, S. (1995). Orienting in a defensive world: Mammalian modifications of
our evolutionary heritage. A Polyvagal Theory. Psychophysiology., 32, 301-318.

9. Porges, S., Doussard-Roosevelt, J., Stifter, C., McClenny, B., & Riniolo, T.
(1999). Sleep state vagal regulation of heart period patterns in the human newborn:
An extension of the polyvagal theory. Psychophysiology, 36, 14-21.

10. Beauchaine, T. (2001). Vagal tone, development, and Gray's motivational theory:
Toward an integrated model of autonomic nervous system functioning in
psychopathology. Development and Psychopathology, 13, 183-214.

11. Gagnon, R. (1992). Fetal behavioral states: Biological alteration. Seminars in
Perinatology, 16, 234-238.