Citation
Attenuation correction methods for positron emission tomography

Material Information

Title:
Attenuation correction methods for positron emission tomography
Creator:
Wu, Chen-Hsien
Publication Date:
Language:
English
Physical Description:
xii, 101 leaves : ill. ; 29 cm.

Subjects

Subjects / Keywords:
Algorithms ( jstor )
Density estimation ( jstor )
Diagnostic imaging ( jstor )
Estimation methods ( jstor )
Image reconstruction ( jstor )
Lungs ( jstor )
Maximum likelihood estimations ( jstor )
Pets ( jstor )
Photons ( jstor )
Transcriptional regulatory elements ( jstor )
Dissertations, Academic -- Electrical and Computer Engineering -- UF ( lcsh )
Electrical and Computer Engineering thesis, Ph.D ( lcsh )
Image processing -- Digital techniques ( lcsh )
Imaging systems in medicine -- Mathematical models ( lcsh )
Tomography, Emission ( lcsh )
Genre:
bibliography ( marcgt )
non-fiction ( marcgt )

Notes

Thesis:
Thesis (Ph.D.)--University of Florida, 1998.
Bibliography:
Includes bibliographical references (leaves 97-101).
General Note:
Typescript.
General Note:
Vita.
Statement of Responsibility:
by Chen-Hsien Wu.

Record Information

Source Institution:
University of Florida
Holding Location:
University of Florida
Rights Management:
Copyright [name of dissertation author]. Permission granted to the University of Florida to digitize, archive and distribute this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.
Resource Identifier:
029248464 ( ALEPH )
39669758 ( OCLC )

Downloads

This item has the following downloads:


Full Text














ATTENUATION CORRECTION METHODS FOR
POSITRON EMISSION TOMOGRAPHY












By

CHEN-HSIEN WU


A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY


UNIVERSITY OF FLORIDA


1998


























Dedicate to my parents,

Mr. Min-Ming Wu and Ms. Jau-Jr Wu,

and my sister,

Hwa-Yu.

















ACKNOWLEDGEMENTS


I would like to thank my advisor, mentor and best friend, Professor John M.

M. Anderson for his endless patience and support. His open-mindedness has made

the time I have spent working for him become one of the most precious experiences

in my life.

Thanks go to Professor Bernard Mair for providing me lots of information and

suggestions which have strong influence on my research.

Thanks go to Professor Murali Rao, who has helped me find the solutions to

numerous problems that I have encountered while working on this dissertation.

Thanks go to Professor Jose Principe and Professor William Edmonson for

offering precious opinions which make this dissertation better.

I would like to thank Dr. Chuan Wang, who advised me on heading in the right

direction. Thanks to Pei-Fang, Hsin-Ho, Byron, Kali, Lisa, Chris, Raymond, Ray,

and Anke for their priceless support in helping me to walk through this journey

and overcome all difficulties.

Best thanks go to my family and ancestors.














TABLE OF CONTENTS



ACKNOWLEDGEMENTS .......................... iii

LIST OF FIGURES .............................. vi

LIST OF TABLES ................................ x

ABSTRACT .................................. xi

1 INTRODUCTION ............................. 1

1.1 Overview: Positron Emission Tomography ............... 2
1.2 Emission Data .............. ...................... 5
1.3 Clinical and Research Applications .................... 7
1.4 Advantages and Disadvantages ........................ 8

2 BACKGROUND .............................. 10

2.1 Basic Physics of PET ......................... 10
2.2 Poisson M odel ............................. 14
2.3 Errors in Positron Emission Tomography ................ 15
2.3.1 Background on Attenuation .... .. 15
2.3.2 Accidental Coincidences . .... .. 20
2.3.3 Scatter . .. .. 22
2.3.4 Other Errors . ... .. 24
2.4 Transmission Tomography in PET ... .. 24
2.5 Modified Poisson Model . ... .. 26
2.6 Existing Attenuation Correction Methods .... .. 29
2.6.1 Standard Correction Method .... .. 29
2.6.2 Reprojection Methods ... ................... .. 30
2.6.3 Simultaneous Estimation Methods .... .. 31

3 MAXIMUM LIKELIHOOD METHOD .... .. 33

3.1 Maximum Likelihood Estimator . ... .. 33










3.2 Concavity of The Log Likelihood Function .... .. 35
3.2.1 Example to Illustrate Concavity of the Log Likelihood Func-
tion . .. 37
3.3 Optimization Procedure . ... .. 38
3.3.1 x-step . .. .. 39
3.3.2 si-step . .. .. 40
3.3.3 Monotonically Increasing Value of the Log Likelihood Func-
tion in x-step . ... .. 41
3.3.4 Property of the ML Estimator .... .. 44

4 LEAST SQUARES REPROJECTION METHOD ... ..46

4.1 LS Algorithm for Estimating the Attenuation Density ...... .. 46
4.2 LS Reprojection Method . ... .. 47

5 SIM ULATION . ... .. 49

5.1 Synthetic Data . ... .. 50
5.1.1 ML Method Implementation of Algorithms ... .. 51
5.1.2 EMML Algorithm with Corrected Data ... .. 51
5.1.3 LS Reprojection Method . ... .. 51
5.1.4 Non-smoothed Typical Transmission Scan Data 52
5.1.5 Smoothed Typical Transmission Scan Data ... ..63
5.1.6 Non-smoothed Short Transmission Scan Data ........ 71
5.1.7 Smoothed Short Transmission Scan Data ... .. 77
5.2 Real Data . ... .. 85
5.2.1 Overview of Implementation ... .. 86
5.2.2 Results . .. .. 87

6 CONCLUSIONS . ... .. 92

6.1 Future W ork . ... .. 93

A EMML Algorithm for PET . ... .. 95


REFERENCES . ... .. 97

BIOGRAPHICAL SKETCH . ... .. 101















LIST OF FIGURES




1.1 Quest 250 from GE Medical Systems (courtesy of GE Medical Sys-
tem s) . . 3

1.2 A positron travels a short distance and annihilates with a nearby
electron to produce a pair of photons that fly off at 180-degree
angle. These photons are detected by a ring of detectors 6

2.1 Annihilation Process . ... .. 11

2.2 Scanditronix PC4096-15WB detector module: 16 crystals, 1 light
guide, and 2 dual cathode PMTs .... .. 13

2.3 Attenuation due to absorption . ... .. 16

2.4 Attenuation due to scattering . ... .. 18

2.5 Basic concept of attenuation . ... .. 19

2.6 Accidental coincidence in PET . ... .. 21

2.7 Scatter in PET . ... .. 23

2.8 Blank scan . ... .. 25

2.9 Transmission scan . ... .. 27

5.1 Reconstruction of cardiac emission phantom using non-smoothed
typical transmission scan data (a) phantom (b) 1st global iteration
(c) 2nd global iteration (d) 3rd global iteration ... .. 53

5.2 Reconstruction of cardiac attenuation phantom using non-smoothed
typical transmission scan data (a) attenuation phantom (b) 1st
global iteration (c) 2nd global iteration (d) 3rd global iteration 54









5.3 Reconstruction of cardiac emission phantom using non-smoothed
typical transmission scan data (a) emission phantom (b) EMML
algorithm with corrected data (c) LS reprojection method (d) ML
m ethod . .. .. 56

5.4 Line plot of the 50th row of the reconstruction of cardiac emission
phantom using non-smoothed typical transmission scan data 57
5.5 Reconstruction of cardiac attenuation phantom using non-smoothed
typical transmission scan data (a) attenuation phantom (b) ML
method (c) LS reprojection method .... .. 59

5.6 ML method (a) emission phantom (b) emission image (c) attenua-
tion phantom (d) attenuation image .... .. 60

5.7 Reconstruction of chest emission phantom using non-smoothed typ-
ical transmission scan data (a) emission phantom (b) EMML algo-
rithm with corrected data (c) LS reprojection method (d) ML method 61

5.8 Reconstruction of chest attenuation phantom using non-smoothed
typical transmission scan data (a) attenuation phantom (b) ML
method (c) LS reprojection method .... .. 62
5.9 Reconstruction of cardiac emission phantom using smoothed typical
transmission scan data (a) emission phantom (b) EMML algorithm
with corrected data (c) LS reprojection method (d) ML method 64

5.10 Line plot of the 50h row of reconstruction of cardiac emission phan-
tom using the smoothed transmission scan data ... ..65
5.11 Reconstruction of cardiac attenuation phantom using the smoothed
typical transmission scan data (a) attenuation phantom (b) ML
method (c) LS reprojection method ... .. 67
5.12 ML method (a) emission phantom (b) emission image (c) attenua-
tion phantom (d) attenuation image .... .. 68
5.13 Emission image of the chest phantom using smoothed typical trans-
mission scan data (a) emission phantom (b) EMML algorithm with
corrected data (c) LS reprojection method (d) ML method 69









5.14 Reconstruction of chest attenuation phantom using smoothed typi-
cal transmission scan data (a) attenuation phantom (b) ML method
(c) LS reprojection method . ... .. 70
5.15 Reconstruction of cardiac emission phantom using the non-smoothed
short transmission scan data (a) emission phantom (b) EMML al-
gorithm with corrected data (c) LS reprojection method (d) ML
m ethod . .. ..73
5.16 Reconstruction of cardiac attenuation phantom using non-smoothed
short transmission scan data (a) attenuation phantom (b) ML method
(c) LS reprojection method . ... .. 74
5.17 ML method (a) emission phantom (b) emission image (c) attenua-
tion phantom (d) attenuation image .... .. 75
5.18 Reconstruction of chest emission phantom using non-smoothed short
transmission scan data (a) emission phantom (b) EMML algorithm
with corrected data (c) LS reprojection method (d) ML method 76
5.19 Reconstruction of chest phantom using non-smoothed short trans-
mission scan data (a) attenuation phantom (b) ML method (c) LS
reprojection method . ... .. 78
5.20 Reconstruction of cardiac emission phantom using smoothed short
transmission scan data (a) emission phantom (b) EMML algorithm
with corrected data (c) LS reprojection method (d) ML method 79
5.21 Reconstruction of cardiac attenuation phantom using smoothed
short transmission scan data (a) attenuation phantom (b) ML method
(c) LS reprojection method . ... .. 81
5.22 ML method (a) emission phantom (b) emission image (c) attenua-
tion phantom (d) attenuation image .... .. 82
5.23 Reconstruction of chest emission phantom using smoothed short
transmission scan data (a) emission phantom (b) EMML algorithm
with corrected data (c) LS reprojection method (d) ML method 83
5.24 Reconstruction of chest attenuation phantom using smoothed short
transmission scan data (a) attenuation phantom (b) ML method (c)
LS reprojection method . ... .. 84









5.25 Emission images using high quality smoothed transmission scan
data (a) Filtered back-projection method (b) EMML algorithm
with corrected data (c) LS reprojection method (d) ML method 88
5.26 Attenuation images using high quality smoothed transmission scan
data (a) ML method (b) LS reprojection method ... 90
5.27 Emission image using high quality non-smoothed transmission scan
data (a) EMML algorithm with corrected data (b) ML method. 91
















LIST OF TABLES




5.1 PSNR (dB) of emission images using typical transmission scan data 71

5.2 PSNR (dB) of reconstruction of cardiac emission phantom using
short transmission scan data . ... .. 85
















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


ATTENUATION CORRECTION METHODS FOR
POSITRON EMISSION TOMOGRAPHY


By


Chen-Hsien Wu


May 1998



Chairman: Dr. John M M Anderson
Major Department: Electrical and Computer Engineering


Positron emission tomography (PET) is an imaging modality that provides bio-

chemical information about the human body. Among the sources of error in PET,

attenuation is the one that impacts image quality the most. The standard method

for correcting attenuation consists of the following two step procedure. First, a

blank scan and transmission scan are used to estimate the survival probabilities.

Then, the data for each pair of detectors is divided by the corresponding survival

probability estimate. This corrected data and a reconstruction algorithm, such as










the filtered back-projection algorithm, are then used to generate an estimate of

the emission intensity, which is displayed as an image.

In this dissertation, two iterative methods are proposed for correcting atten-

uation errors in PET. The first method is a maximum likelihood (ML) method

that simultaneously estimates the emission intensity and attenuation density. The

second one is a least squares (LS) reprojection method, where the survival prob-

abilities are estimated by reprojecting a LS estimate of the attenuation density.

The emission intensity is estimated using the expectation maximization maximum

likelihood (EMML) algorithm with a probability matrix that has been modified by

the survival probability estimates. Both methods are guaranteed to produce non-

negative estimates of the emission intensity and the attenuation density, provided

that the initial estimates are nonnegative.

From simulation studies, it appears that the major advantage of the proposed

methods over the standard method with the filtered back-projection algorithm

is that they (1) have less noise, (2) do not have artifacts outside the region of

interest, and (3) they provide high quality images even for short transmission

scans (less than 10 minutes).














CHAPTER 1
INTRODUCTION

For both clinical and research applications, medical imaging modalities are

used to obtain important information on the functionality and anatomy of the

human body. Imaging modalities such as X-ray computed tomography and mag-

netic resonance imaging provide detailed images of the anatomical structure of

the body. However, in many applications, it is more beneficial to have a modality

that determines functional information such as blood flow rates, oxygen utiliza-

tion rates, and glucose utilization rates within the region-of-interest (ROI) (e.g.,

brain, heart) [28]. For example, glucose utilization rates are used in oncology to

diagnose tumors and to determine the efficacy of a treatment.

Functional information on the human body can be obtained using (1) positron

emission tomography (PET), (2) single-photon emission computed tomography

(SPECT) [16], and (3) functional magnetic resonance imaging [51]. Among them,

PET is probably the most flexible imaging modality. For example, PET can

image which parts of the brain are functioning while a subject is performing a

speaking tasking. This is not possible in functional magnetic resonance imaging

because the scanners are too loud [55]. An advantage of PET over SPECT is

that the radionuclides in PET can be incorporated into molecules (or analogs) of

biological interest [56].










In this dissertation, we propose two methods for correcting a certain error in

PET called attenuation. The outline of the dissertation is as follows. An overview

of PET is provided in the remainder of Chapter 1. In Chapter 2, we discuss the

physics behind PET, introduce a mathematical model for the data, and review

existing attenuation correction methods. We introduce the proposed maximum

likelihood and least-squares reprojection methods in Chapter 3 and Chapter 4,

respectively. In Chapter 5, we investigate the performance of these methods using

synthetic and real data. Finally, we discuss our conclusions and future research

direction in Chapter 6.




1.1 Overview: Positron Emission Tomography

PET is exceptional in its capability to quantitatively determine functional

information within a ROI. Since George Hevesy won a Nobel Prize for his pio-

neering research on radionuclides in 1943, numerous researchers have investigated

methods to obtain functional information on the human body. However, the first

PET machine was not built until 1975 by Michael Phelps, Edward Hoffman, and

their colleagues [42]. Since then, scientists and engineers have made significant

advances in the hardware, radionuclides, and reconstruction algorithms used for

PET. In Fig. 1.1 is a PET scanner of the Quest series from the GE Medical

Systems.

We first introduce the major components of a PET scan, which are as follows:


1. A compound is labeled with a radionuclide.




















































Figure 1.1. Quest 250 from GE Medical Systems (courtesy of GE Medical Sys-
tems)










2. The labeled compound or radiopharmaceutical is introduced into the sub-

ject's body.

3. The data are collected.

4. An image is produced that represents the concentration of the radiophar-

maceutical within the ROI.


Radionuclides in PET are (1) incorporated into compounds that are naturally

used by the body, (2) emit positrons, and (3) have short half-lives. A radionuclide's

half-life must be sufficiently long so that there is enough data to get quality

images [38]. The most common radionuclides in PET include "C, 'O50, and

3N, which are used to label glucose, water and amino acid, respectively. The

most widely used radiopharmaceutical is SF-labeled deoxyglucose (FDG), which

is used to study glucose metabolism. Because of the short half-lives of positron-

emitting radionuclides (2 minutes, 10 minutes, 20 minutes, and 110 minutes for
"11C, O150, 13N, and SF, respectively), they need to be produced on-site using a

cyclotron. The cyclotron creates radionuclides by smashing the atomic shell of

certain molecules with protons. Cyclotrons are expensive and costly to maintain

because they require at least two operators.

SPECT uses radionuclides, such as 99mTc, 201TI, and 1231, which have half-

lives that range from 10 hours to several days. Because of their long half-lives,

radionuclides in SPECT can be made in advance and shipped to hospitals so an

on-site cyclotron is unnecessary. However, long half-lives mean that patients are

exposed to radioactivity for long periods.






5


After being administered to a subject, the labeled compound will localize in the

ROI. For example, FDG is absorbed preferentially by the brain because glucose

is a source of energy [11]. Another example is the use of 150 based radiopharma-

ceuticals to index injured tissues by monitoring the oxygen metabolism of tissues

[52].

1.2 Emission Data

As mentioned previously, the radiopharmaceutical localizes in the ROI. When

the radionuclide decays, it emits positrons which travel a short distance (3-4 mm)

before annihilating with nearby electrons. The areas with a high concentration of

the radiopharmaceutical emit more photons than the areas of low concentration.

When a positron annihilates with an electron, two high-energy (511 keV) photons

are produced that fly off at an almost 180-degree angle from each other. This

annihilation process is shown in Fig. 1.2. The photons go through the subject and

are registered by rings of detectors surrounding the subject. When two detectors

detect a pair of photons within a short period of time, these photons are said to

be in coincidence (see Fig. 1.2). For each pair of detectors, there is a count that

corresponds to the number of observed coincidences. The emission data is the

collection of these counts.

From the emission data, we can estimate the concentration of the radiophar-

maceutical using a reconstruction algorithm. A computer is then used to display

images where the intensity is proportional to the estimated concentration of the

radiopharmaceutical. In the next section, some clinical and research applications

of PET will be presented.
















































Figure 1.2. A positron travels a short distance and annihilates with a nearby
electron to produce a pair of photons that fly off at 180-degree angle. These
photons are detected by a ring of detectors









1.3 Clinical and Research Applications

PET is widely used in pathology and oncology because it can quantitatively

measure blood flow, and utilization rates of glucose, C02, and fatty acid ammo-

nia. PET provides functional information about tissues or organs, which cannot

be obtained by modalities such as magnetic resonance imaging and computed to-

mography [52]. It is also a powerful tool for investigating brain functions such

as emotion, thinking, vision, and speech [5, 30]. Since the major energy source

of the brain is glucose, FDG is used to map the brain's energy consumption. In

other words, if a specific region of the brain has increased functionality, then the

regional blood flow increases to provide the necessary energy. As a result, the con-

centration of the radionuclide used in FDG increases in that region. Using PET

to map the concentration of the radionuclide, we can observe the higher function-

ality in that region of the brain. Based on the concepts discussed above, scientists

have used PET images to assess the brains of normal children and children under

abuse. They found that the children who were abused had slower developments

in brain functions such as speech and vision as compared to normal children [7].

PET has clinical applications as well. For example, PET has the ability to

measure the size of infarcts in the heart after a coronary attack, to diagnose dis-

eases such as Alzheimer's [45] and epileptic seizures [36], and to survey the effects

of treatments such as radiation therapy, chemotherapy, and hormone therapy. In

addition, whole-body PET is used to spot metastases, which is very difficult for

other modalities because of the small sizes and the random occurrences of metas-

tases inside cancer patients' bodies. The effect of the chemotherapy on cancer










patients can be evaluated by visualizing the sizes and locations of metastases on

PET images [1].

PET is also used in psychological studies. For instance, it was used in the

early 1990s several times by courts in California to determine whether neural

malfunctions in the brain caused a suspect's violent behavior. Another interesting

experiment was the monitoring of six bank officials' brain functions when they

were exposed to a video of an armed bank robbery. This was done to find out the

connection between stress and visual re-experience [22].




1.4 Advantages and Disadvantages

PET is a better imaging modality than SPECT in the sense of offering better

efficiency, sensitivity, and accuracy as compared to SPECT [11]. Also, unlike

SPECT, the radionuclides used in PET are easily incorporated into compounds,

such as glucose, that are naturally used by the body. Using different radionuclides,

PET is able to image different biochemical activities while functional magnetic

resonance imaging only can image hydrogen activity. Additionally, PET can track

chemical changes that do not affect blood flow enough for functional magnetic

resonance imaging to measure.

Unfortunately, PET has some major disadvantages such as its high cost, high

complexity, and the requirement of highly skilled personnel. The cost of a PET

scanner is similar to a magnetic resonance imaging scanner. However, the cy-

clotron used to generate the radionuclides is also very expensive. Additionally,

at least three skilled professionals one to maintain the scanner, one to run the






9


cyclotron and another one to produce radionuclides are needed to operate a

PET laboratory. For SPECT, it is not necessary to create the radionuclides on

site because the radionuclides used in SPECT can be stored. Even though its

image quality is not as good as that of PET, SPECT is more widely available in

practice because it is significantly cheaper.














CHAPTER 2
BACKGROUND

In this chapter, we review the basic physics of PET, introduce the Poisson

model, and discuss sources of errors in the data. Since the focus of this dissertation

is attenuation correction, we also review existing attenuation correction methods.




2.1 Basic Physics of PET

As discussed in Chapter 1, the subject is given a radiopharmaceutical that

emits positrons. When a positron annihilates with a nearby electron, two photons

are produced that depart at an angle near 180 degrees. The region of interest

(ROI) is surrounded by rings of detectors that detect the photons. We will use

the term tube to refer to a pair of detectors. When a pair of photons is incident

on a tube within a small time interval, a counter is increased (see Fig. 2.1). The

emission data are the counts associated with the various tubes.

The detectors are shielded from radiation coming from outside the region of

interest by heavy lead rings at the front and back of the gantry. In order to

switch between the slice-collimated mode and the three-dimensional mode, most

commercial PET scanners contain spherical thin interplane rings called septa.

When the scanner is in slice-collimated mode, the septa only allows in-plane tubes

to detect photons. When a scanner operates in full three-dimensional mode, the



































Coincidence
Timing -
Circuit


Figure 2.1. Annihilation Process









septa are withdrawn to permit photons to be registered by in-plane tubes and

cross-plane tubes [37].

A detector module consists of an array of crystals coupled to several photomul-

tiplier tubes. In Fig. 2.2, a detector module of the Scanditronix PC4096-15WB

whole-body PET scanner is shown. A detector ring contains several hundred de-

tector modules (e.g., 384 for the Siemens-CTI ECAT EXACT 921 scanner). When

a crystal is hit by a 511 keV photon, it interacts with the photon and generates nu-

merous light photons. Then, the light guide connected to the crystal transfers the

light photons to the nearest photomultiplier tube (PMT). The PMTs collect light

photons and convert them into electrical signals that are stored in a computer.

The outputs of the PMTs are decoded to determine which detector module was

hit by a photon. The most widely used crystal material is Bismuth-Germanate

because it has good properties when it interacts with 511 keV photons such as a

high light photon producing rate, short decay time, and a fast energy rise time

[26].

Once the data is collected, algorithms are used to reconstruct the PET im-

ages. The most popular method is the filtered back-projection algorithm [45]. It

is extremely fast and easy to implement, but the images are noisy, have nega-

tive values, and contain artifacts outside the ROI. Alternative approaches include

statistically based methods such as maximum likelihood [28, 46], maximum a

posterior [20, 21, 30, 31], least squares [15, 40, 24, 46, 49, 50], and weighted least-

squares [3, 12, 17] methods. An advantage of statistically based methods is that

they account for the errors in the data and do not have artifacts outside the ROI.

A disadvantage of them is that they are computationally demanding.


















PMT
















\ Light Guide

Crystals
Figure 2.2. Scanditronix PC4096-15WB detector module: 16 crystals, 1 light
guide, and 2 dual cathode PMTs









2.2 Poisson Model

Since the data collection in PET is actually a photon counting process, it is

reasonable to assume that the emission process is a Poisson process with unknown

mean [38]. Based on this assumption, Shepp and Vardi proposed a model called

the Poisson model [46] in 1982. Their model assumes that the data is error free

and obeys a Poisson process.

In the Poisson model, the ROI is discretized into a grid of J voxels. Let Nj

denote the number of emissions in voxel j, where Nj is assumed to be Poisson with

unknown mean xj = E[Nj]. Let Y, denote the number of coincidences detected in

the ith tube, where Yi is assumed to be Poisson with mean di = E[Yi] and I is the

number of tubes. It is also assumed that (1) {Nj}j' are independent, (2) {Yj}j

are independent, and (3) the probability that an emission in voxel j is detected in

the i1th tube, Pij, is known for all i,j. Under these assumptions, it can be shown

that the mean of the ith tube is given by

J
di = E xjPij (2.1)
j=1
= (Px)i, i = 1,2,...,I, (2.2)


where the (i,j)th element of P is Pij and (Pax), denotes the ith element of the

vector Pax.

Given the data, Y1, Y2, ..., yi, where y, is an observation of Yi, the goal is to

estimate x = [x1 x2 ... xj]T. In this way, an estimate for the number of photons

emitted in each voxel is obtained. Throughout the remainder of this dissertation,

x will be referred to as the emission intensity.









Shepp and Vardi used the expectation maximization algorithm [16] to compute

a ML estimate of the emission intensity. This algorithm, which is referred to as

the expectation maximization maximum likelihood algorithm, is briefly discussed

in the Appendix.

In the next section, we discuss sources of errors in PET.




2.3 Errors in Positron Emission Tomography

In reality, PET data are corrupted by errors. In this section, we go through

most of the important sources of error. Of all the errors, the one that has the

most effect on image quality is attenuation [23]. Since we focus on developing new

algorithms to correct attenuation, we provide an in depth discussion of attenua-

tion.




2.3.1 Background on Attenuation

Attenuation is the main error source in PET emission image reconstruction

[34]. The 511 keV photons produced by the annihilations are attenuated by tissue,

bone, and air. Different mediums have different attenuation effects. Before going

into the details of attenuation correction for PET, we introduce the physics behind

attenuation.

The main sources of attenuation in PET are absorption and Compton scatter-

ing. Photons are rarely absorbed because they carry 511 keV of energy. Hence,


















































Figure 2.3. Attenuation due to absorption










attenuation due to absorption is negligible. Fig. 2.3 is a depiction of photon ab-

sorption. Compton scattering occurs when the path of the photon is deflected by

an outer shell electron [37]. Photons in PET are more likely to undergo Compton

scattering than absorption. Most of the photons that undergo Compton scatter-

ing are not detected. Fig. 2.4 shows a photon undergoing Compton scattering.

The total effect of photon absorption and Compton scattering is referred to as

attenuation.

The survival probability along a line L, a, is the probability that a photon is

not attenuated along that line. It can be shown that the survival probability is

equal to


a = e- fL pd, (2.3)


where it is the attenuation density [29].

Since eq.(2.3) is a fundamental equation in attenuation, we provide an expla-

nation of it. Consider Fig. 2.5, which shows a line source emitting photons toward

an object with attenuation density pt. We assume that N photons incident on a

point within the object travel a short distance, Al, and only N + AN photons

survive. Obviously, AN is a negative number. We may use the following equation

to describe the above process:


AN 1
AN 1 (2.4)




















































Figure 2.4. Attenuation due to scattering





19




Al
-HH-


/line L


/ ...
Source Ni Nout Detector


N N+AN
Tissue

AN 1

N Ai


Figure 2.5. Basic concept of attenuation









As Al and AN go to zero, we get the differential equation

1 dN (2.5)


which can be solved by integrating along the line L. Doing so, we have that

/Nout dN
jNt = -I fPdl (2.6)
,N, N J

where Nin is the number of photons that go into the object and Nout is the number

of photons that come out of the object. From the above equation, we get

InNoit In Ni. fL J dl. (2.7)

It follows from eq. (2.7) that

Nut =e-fLdl (2.8)


Finally, eq. (2.3) follows by noting that the left side of (2.8) is the probability

that a photon along the line L is not attenuated. In the next several sections, we

discuss some other errors in PET.




2.3.2 Accidental Coincidences

If two photons from different annihilations are registered as a coincidence by

a tube, then an error occurs. This kind of error is referred to as an accidental

coincidence [22]. Fig. 2.6, shows an example of an accident coincidence. Two

annihilations happen inside the ROI. However, in both annihilations, a photon

flies out of the detector ring. The remaining photons are detected by a tube

within the coincidence window and incorrectly recorded as a coincidence.




































Out of plane


Figure 2.6. Accidental coincidence in PET










The accidental coincidence rate is proportional to the concentration of the

radionuclide in the ROI. Usually, the accidental coincidences are corrected by

subtracting an estimate of them from the observed photon counts. Accidental

coincidences are estimated either by using an estimate of the single events to

compute them or using a delayed coincidence circuitry to measure them [22].

This subtraction method may produce negative counts in the corrected data,

which means that it is no longer Poisson. Consequently, several methods have

been proposed to resolve this problem [2, 28, 41]. A discussion can be found in

Anderson et al. [4] and Ollinger et al. [37]. In the next section, another error

source, scatter, is reviewed.




2.3.3 Scatter

In a scattered event, one or both of the 511 keV photons created in an an-

nihilation are scattered but they still intersect a detector ring. As a result, the

annihilation event is registered by the wrong tube. An example of an error due

to scatter is shown in Fig. 2.7.

Since photons lose energy after they are scattered, it is very easy to distinguish

them from unscattered photons by measuring the energy they pass to the detector

crystals [8]. This technique can be used to correct the scattered events by setting

a threshold to reject photons with energy sufficiently less than 511 keV. Scatter

between different slices in whole-body PET is minimized by the septa (see Section

2.1).


















































Figure 2.7. Scatter in PET










2.3.4 Other Errors

In reality, positrons travel a short distance (3-4 mm) before annihilating with

an electron and photons shoot off at an angle less than 180. Also, photons can

penetrate the crystal they hit and leak energy to a neighboring tube [26]. Another

type of error, called detector inefficiency, occurs when a detector fails to detect

an incident photon [17]. Fortunately, most of the time, these errors are small and

usually can be ignored or corrected.

In order to correct for attenuation, certain scans called blank scans and trans-

mission scans are used to estimate the survival probability along each tube. In

the next section, we provide a brief discussion of blank scans and transmission

scans.




2.4 Transmission Tomography in PET

In order to correct for attenuation, a blank scan and transmission scan are

performed. For the blank scan, a scan is done without a subject. The external

photon source, as seen in Fig. 2.8, rotates along the interior of the detector ring.

Let bi denote the number of observed coincidences in tube i during the blank scan,

i = 1,2,..., I. The blank scan lasts about an hour to more than 2 hours. It is

performed at the beginning of every day.

After positioning the subject in the scanner, a scan is performed, which is

referred to as a transmission scan. Like the blank scan, the transmission scan uses

a rotating external source of photons (see Fig. 2.9). The photon counts collected

in a transmission scan are the ones that survive the attenuation effect of the ROI.


















































Figure 2.8. Blank scan










Let ri denote the number of observed coincidences during the transmission scan

for tube i, i = 1,2,..., I. A transmission scan needs to be performed each time

a new patient is put in the PET scanner. The duration of a transmission scan

ranges from about 2 to 25 minutes. For the safety and comfort of the patient, the

transmission scans are kept as short as possible. (patient should not be exposed

to the radioactive source too long).

Intuitively, it is reasonable to estimate the survival probability by taking the

ratio of the transmission scan count per unit time and the blank scan count per

unit time. Hence, an estimate of the survival probability in tube i, &j, is


&j -- T (2.9)


where Tb and Tr are the durations of the blank scan and the transmission scan,

respectively. The data can then be corrected for attenuation by dividing the data

by the estimated survival probabilities (i.e., ML). This approach is simple and

intuitively appealing, but it does not work well if the transmission scan is too

short. One way to resolve that problem is to modify the Poisson model instead of

directly correcting the data. In next section, we introduce the modified Poisson

model.




2.5 Modified Poisson Model

In this section, we discuss an extension of the Poisson model for the case

of attenuation errors only [28]. Before discussing this model, we introduce the

concept of thinning a Poisson process.



















































Figure 2.9. Transmission scan










Result 1: Let N be a Poisson random variable with mean x. Suppose a random

variable X is created by thinning N with probability p. Then X is Poisson

with mean px.

Proof:


00
P(X = k) = P(X = kIN = m)P(N = m)
m=k
0(rn)pk(1 p)rn-kP(N =rm)
oo
()kl_ )mpk(_p-k_
=!(P- = m)
m=k


Se-p (l p)kk-kxm
m rk (-
k! m=k ( -k)!


Replacing m k with n, yields


P(X = k)


e-epk- (1 p)nx )+k

k!/- n!
n=O
-e-x(xp)k -,((1 --p)x)"


- e-X(Xp)ke(l-p)x
k!
k-! xp)'
~ k!


which shows that X is a Poisson random variable with mean px.


We can apply Result 1 to model the effect of attenuation in PET (see also

[41]). Let yij be the attenuation density within the j1'h voxel, where it is assumed

that the attenuation density within each voxel is constant. Let L be an I x J

matrix whose (i,j)th element, Lij, is the intersection length of the ith tube and










the Jth voxel. Then, it follows from eq.(2.3) that the probability that a photon

pair is detected by tube i is approximately equal to


-(Lit), (2.10)
O~i = C-


where (LpA), = Zl Lij pj, i = 1,2,... ,JI. Because of attenuation, the data in

tube i, Yi, is created by thinning the error free data in tube i by the survival

probability aj. By Result 1, it follows that Y, is Poisson with unknown mean


E[Y,] = ai(Px)i. (2.11)

With this modified Poisson model, several statistical reconstruction algorithms

have been proposed.




2.6 Existing Attenuation Correction Methods

In this section, we review existing attenuation correction methods including

those based on the modified Poisson model.




2.6.1 Standard Correction Method

The standard way to compensate for the attenuation is to divide the emission

data by the estimate of the survival probabilities. Specifically, the corrected data

for the ith tube, j, is given by

= --,i= ,2,...,J, (2.12)
oti









where &, is given by eq. (2.9).

This method is easy to implement and works fairly well for long transmission

scans (20-25 minutes), for short transmission scans (less than 10 minutes), it

produces noisy emission images. The conventional way this problem is addressed

is to smooth the transmission sinogram. Smoothing reduces the noise, but it also

lessens the quantitative accuracy of the images [34].




2.6.2 Reprojection Methods

Given the blank scan data and the transmission scan data, an estimate of the

attenuation density is obtained [36, 41]. Then, the estimate, it, is "reprojected"

to obtain an estimate of the survival probabilities:


= e- =1L3 ,1 = 1,2, ..., J. (2.13)

The estimated survival probabilities are used to correct the data or modify the

probability matrix, P. In addition to improving the accuracy of the estimated

survival probabilities, the transmission images provide information that can be

used to position patients, estimate scattering errors, and to provide anatomical

landmarks that cannot be seen in emission images [36].

The reprojection method provides a better estimate of the survival probabil-

ities. However, the counting noise from the transmission image introduces addi-

tional error into the emission reconstruction. In regions such as the thorax, where

the attenuation medium is heterogeneous, this problem cannot be resolved by









simply smoothing the estimated survival probabilities because it introduces bias

[13]. Since reprojection methods estimate the attenuation density, they are more

computationally demanding than the standard correction method. In the next

section, we discuss methods that simultaneously estimate the attenuation density

for survival probabilities and the emission intensity.




2.6.3 Simultaneous Estimation Methods

In [41], Politte and Snyder proposed constrained maximum likelihood recon-

struction algorithms by modifying the probability matrix under the assumption of

known survival probabilities. Their methods are not practical since the survival

probabilities must be estimated. The transmission data provides noisy estimate

of the survival probabilities because the count is pretty low in real applications.

Since the emission data are more reliable than the transmission data, in the sense

of having higher data counts, simultaneous estimation methods may provide a

more accurate estimate of the emission intensity.

In [11], a simultaneous estimation method was proposed to estimate the atten-

uation density and the emission intensities for SPECT using the Cyclic Subgradi-

ent Projection (CSP) algorithm. After the Poisson model was proposed, Fessler,

Clinthorne and Rogers proposed a joint maximum likelihood method that esti-

mates the emission intensity and survival probabilities [13, 18]. Lu proposed a

MAP algorithm to estimate the emission intensity and the attenuation density for

SPECT [32]. However, prior information about the attenuation medium must be

known.






32


To simultaneously estimate the attenuation density and the emission intensity,

we propose a new method. In the next chapter, we discuss this proposed method.













CHAPTER 3
MAXIMUM LIKELIHOOD METHOD

In this chapter, we present an attenuation correction method that jointly esti-

mates the attenuation density and emission intensity using a maximum likelihood

(ML) method. Other ML methods have been proposed [13, 18], but they have

the disadvantage of requiring more unknowns than data points. By contrast,

the proposed method has fewer unknowns than data points for typical scanners.

Specifically, it estimates 2J unknowns from I data points, where J and I are the

number of the voxels and the number of the tubes, respectively. A common choice

for the number of voxels is J = 16384 (i.e., J = 128 x 128) and, for the Siemens-

CTI ECAT EXACT 921 PET scanner, the number of the tubes is I = 73536.




3.1 Maximum Likelihood Estimator

As discussed in Section 2.2, the emission data for the ith tube, yi, is assumed

to be an observation of a random variable, Yi, where Yi is Poisson with mean

(Px)ie-(L) (recall, (Px), = xjPij and (Li)i = EJ- 1,jLij). The emis-

sion intensity, ax, and attenuation density, pi, are assumed to be deterministic but

unknown. Given the emission data, yi, y2,... ,yI, blank scan data, bl, b2, ., b,

and transmission scan data, ri,r2,... ,rj, the goal is to estimate the emission

intensity and attenuation density.









It is well known that in many estimation problems ML estimators are consis-

tent and asymptotically unbiased [44]. Thus, we propose to estimate the unknown

parameters using a ML method. In other words, we estimate the emission intensity

and attenuation density by maximizing the log likelihood function, f(ax,/ I):

[a* A*] =rmaa f (x,) ,

where f(x, it) = log P[Yi = y,Y2 = Y2,...,YI = yiIx, A], and x* and i* are ML

estimates of x and p1, respectively. The notation x > 0 means that all of the

elements of x are nonnegative. Under the model assumptions, the log likelihood

function is given by
I
f(x,p.) = logntP[Yi=YjX', ] (3.1)
i=1
Se-[(PX),e-(L) [(pe-(L)]y (3.2)
=logit----L-^-^ ---J (3.2)
= logfly
I
= E[-(pX)ie-(Lit)' + y, log(Px) yi(Lp)i log(yi!)] (3.3)
i=1
Eq. (3.1) follows from the assumption that the data is independent. Note, since

the last term in eq. (3.3) does not contain x or p., it can be ignored in the

maximization procedure. Thus, the above optimization problem is equivalent to

the following one:

(OP) [x* IA*] argwa q(,1),

where
I
O(x, A) = y_[-(Px),e-)(LL) + y, log(Px). yi(LP)i] (3.4)
t=1
The function q(x,p) will be referred to as the log likelihood function for the

remainder of this dissertation.









The problem (OP) is a difficult constrained optimization problem that must

be solved in order to obtain the ML estimate of the attenuation density and the

emission intensity. In the next section, we discuss some properties of the log

likelihood function before presenting our method for solving (OP).




3.2 Concavity of The Log Likelihood Function

Unfortunately, the log likelihood function, S(ax, ft), is not concave over the
A
feasible region, Sx,u = {(Ax, I)Ix > 0, At > 0}. However, for a fixed nonzero

vector zo, q(x, zo) and q(zo,t1z) are concave over the sets Sx = {axax > 0} and

S11 {AIAt > 0}, respectively. These results are proved below.


Result 2: Let zo be a nonzero J x 1 vector. Then, O(x, zo) is concave

over Sx.

Proof:

Our proof is based on the well known fact that a function is concave

if its Hessian is negative semidefinite [6]. Consider l(x, z0), where

Zo > 0 is a fixed J x 1 vector. It follows from eq. (3.4) that (xB, zo)

is given by
I
O(X, zo) = _[-(Px)ie-(Lz)' + y, log(Px), yi(Lz)i] ,
i=1

and its first order partial derivative with respect to xj equals
00(X, Z") Y-P-
_(a -0)- y-3j e-(Lzo)pij] (3.5)
axj j =1(p )









The Hessian of (x, zo,), which we denote by H(x), is a J x J matrix
whose (j, k)th element is given by
H()jk A0 2(X,,') (3.6)

I YiPijPik
= kPi)- ,k=l,2,...,J. (3.7)

Consider the scalar mTH(x)m, where m and x are arbitrary nonzero
vectors:

T J I y2PijmPikmk (3.8)
mTH(x)m = -SEEE ; (3.))
k=1 j=1 i=1 (PA)?
yi(Pm)?
= (px)4 (3.9)

Since for all i, y2 > 0, (Pm)2 > 0, and (Pax), > 0, it follows that
mTH(x)m < 0, which means that the Hessian is negative semidefi-
nite. Thus, O(x, zo) is concave over Sax.

Result 3: Let Zo be a nonzero J x 1 vector. Then, q(zo, Ms) is concave
over the set Sit

Proof:

Similar to the proof for Result 2, we need to prove that the Hessian of
S(zo, Iz) is negative semidefinite for zo, a nonzero fixed J x 1 vector.
From eq. (3.4), it follows that (zo, p) is given by
I
(zo, ,) = Z[-(Pzo),e-(LI)' + y log(Pz), yi(LI),]j ,
i=1
and its first order partial derivative with respect to /,j equals
9 I A I
Z( (^ = ^{Pzo)ie-(LI)'Lij 3yiLij (3.10)
i/ =1 i=1










The Hessian of (Zo, ft), which we denote by H(Is), is a J x J matrix

whose (j, k)t" element is given by


H()jk = ( A)-- (3.11)
I
S- (Pzo)iLjLike-(LA)', j, k = 1,2,...,J.(3.12)
i=1

Consider the scalar mTH(pj)m, where m and 14 are arbitrary nonzero

vectors:

J J I
mTH()m = Ej j -(Pzo)LjmjLikmke-(L)' (3.13)
k=l j=l i=l

-L(Pzo)i(Lm)e-(L) (3.14)
i=1

For all i, (Pzo)i > 0, e-(L>) > 0, and (Lm)? > 0, hence it fol-

lows that mTH(pf)m < 0, which means that the Hessian is negative

semidefinite. Thus, 4(Zo, t) is concave over S1.


3.2.1 Example to Illustrate Concavity of the Log Likelihood Function

To get a sense of the "shape" of the log likelihood function, we consider the

single voxel case. For this case, the log likelihood function becomes


q(x, P) = -pxe-" + y log(px) ylip (3.15)

It is straightforward to show that the Hessian of (x, Ii), which we denote by H,

is given by
__v ple-1p

ple-t. --pxl2e-1o









Consider the scalar mTHm, where m is an arbitrary nonzero vector:

mTHm = --rn2 pxl2e-' m2 + 2ple-mlMrni2. (3.16)
X2

From eq. (3.16), it can be seen that there is no way to guarantee that mTHm is

less than or equal to zero. For p = 1, 1 = 1, x = 40,/p = 0.049, y = 2, mi = 0.2167

and m2 = 0.01,the log likelihood function is not concave (i.e., mTHm > 0).




3.3 Optimization Procedure

We exploit the concavity results discussed in last section and propose the

following optimization procedure to obtain the ML estimates (i.e., solve (OP)):

(Step 1) Given ik-' and xk-1, determine xk by maximizing O(X,pik-l) over Sx.

In other words,
(PI) xk ,k-i1


(Step 2) Given fk-1 and xk, determine ik by maximizing O(Xk, L) over Sif. That

is,
Ak
(P2) a (k

(Step 3) Iterate between Step 1 and Step 2.

Step 1 and Step 2 are referred to as the x-step and the p-step, respectively.

The vectors xk and ,1k are the kth estimate of x and pL, respectively, where an

iteration consists of both steps 1 and 2. Unfortunately, closed form solutions for

x k and ,uk are not available, which means they must be determined using iterative









algorithms. In the next two sections, we develop algorithms for obtaining xk and
k.
A .



3.3.1 x-step

Given Ak-1l > 0, we know that O(X,/1k-l) is concave over Sa from Result 2.
Under the conditions in problem (P1), i is a solution to (P1) if and only if the
Kuhn-Tucker conditions [54] are satisfied:
90(x, -1)
it" I)x=x_ =0 if xj 0 ,j= 1, 2,...,J,
axj
oxj lx.<0O if xj=0, j=l,2,...,J.
ax3
We can rewrite the above two Kuhn-Tucker conditions as
I Yi i
(Cl) ( ) -E(L=-1)i if xj 0, j=1,2,...,J,

(C2) y i i=1 i=1
The first condition (Cl) suggests the following fixed point iteration for solving
(P1):
(Xk)m z yP,
(,k)m+ (k) j = 1,2,..., J (3.17)
(zB )j+ =(x )jE i -(Luk-'),p'P 3~ i"''
ji F=1 e-La-)p

where (x k)7 is the mth estimate of the jth voxel of aXk and (ak)P = (a k-1) for all
j. Another possible iteration is
,--, -(L jAk-1), .
(Xk)m+ = (Xk)m. i Pi j = 1,2,...,J. (3.18)
2^---1 (P(Ck)>),
However, in our simulations, this iteration "blew up" (i.e., the iterates became
unbounded). Consequently, we use the iteration in eq. (3.17) to estimate xk.









The logic behind the fixed point iteration in eq.(3.17) is discussed below. Sup-
pose (aXk) converges to (ak)- 0 then from eq.(3.17) we have

kio kr 1-=1 / Plykixi)
%I at Pij
(^r () =^ T (pak) ) (3.19)


which implies that condition (Cl) is satisfied. In other words, we know that

the fixed point iteration in eq.(3.17) will provide a solution satisfying (Cl) if it
converges to a positive vector. Additional motivation for the technique used to
derive the algorithm in eq. (3.17) is that the EMML algorithm [28, 46] and the
ISRA algorithm [15] can be derived using it. Applying the alternating projection
method [10] to solve (P1) leads to the same iterative equation as eq. (3.17). This

iteration is guaranteed to converge by convergence results in Byrne's paper [10].



3.3.2 uI-step

From Result 3, we know that (xrk, /) is concave over S/z for xk > 0. Thus,
it is a solution to (P2) if and only if the following Kuhn-Tucker conditions hold:
I I
(C3) Z(PaXk),e-(L)'Lj = 3yiLij if /j 0, j = 1,2,...,J,
i=1 i=1
I I
(C4) Z(Pxk)ie-(L)'Lj < yiLij if pj = 0,j = 1,2,...,J.
i=1 i=1
Using the same approach that leads to the iteration in eq. (3.17), we obtain the
following fixed point iteration for solving problem (P2):

I= :l ) ,e-( L(1k ),2(prk)in3ij
1)T+1 = (l k-- t=, j = 1,2,...,J, (3.20)
,Ei=1 yiLij









where (Jtk)T is the m1th estimate of the jt" voxel of itk and (Itk) = (lk-1 )% for all
j. Another possible iterative equation is
E yiLij=
k)rn+i = (,-k)m (Zf- = (P, j= 1,2,..., J (3.21)


In our simulations, this iteration "blew up". Consequently, we estimate Ak using
the iteration in eq. (3.20).
In the next section, we investigate the behavior of the log likelihood function
as a function of the iterates.



3.3.3 Monotonically Increasing Value of the Log Likelihood Function in x-step

From the log likelihood function given in eq.(3.4) and the iterative equation in
eq.(3.17), we can prove that the log likelihood function increases monotonically
with successive iteration in the x-step. Unfortunately, this result cannot be proven
for the ji-step.

Result 5: Let z0 be a fixed nonzero J x 1 vector. Then,

0((xk)m+1, Z.) 0((Xk),, zo) > 0.

Proof:

From eq.(3.4), it is easy to see that

-((Xk)+, z) (k)m, z.)

S (p(k m+ + (o (p(Xk)m+l,)
= El-- ) )"+ (P(X)m),], +-4- E y' o (g k
,=1 i=l (P' V"' ) I


where ai = -(Lz,),.










Hence, using the iterative equation in eq.(3.17), it follows that the

second term of the above derivation can be written as


y log (p(k)m+)i
log (p(Xk)m).


I 1 Yi Pi(
where wj- = ,P "

Let 3ij = (p (a-)) =1 =

is concave, we have


g J
Yi log ij
i1 j=l


I J P.i(Xk)m+l
-- lyog ] "" )J
i=1 j=1 Pz~~
I i P p1-)
= /y'YTlog[ m
= igE [/(p( (k)-),
;=1 J= 1 j=1'J)'
I J p..({kIm
= E Y 1og -[ ( P k k]
i= 1 j= 1- (p (i ), J J '


1. Using the fact that the log function


I J
> EY 1E ij log.
=1 j=l


(3.22)


Multiplying the right side of eq.(3.22) by ( i'" the value of right

side of eq.(3.22) does not change since =' 1. Then, it follows

that


I J
EY /32j log wk
i=1 j=1


I ,ZjEi, logW
- >jy0(^) PZ.7) -

= E^E-.A)j,1j
J I 1 Yi Pi
= E xj(EZaP,) (P(21k)m), log wk
j=l i=1 ZI=1 I Pij
J I~
-=i x(Z )w l=ogw
j=l V=1









J I
E EXk(E -)(
j=l i=1


J I yiPij
= I i= (p(k)Zx)i


I J
S Z 7P=(1 k)?
i=1 .1=l


I I
= Ey y-ET i(PXk))".
i=1 i=1

Since EI Y, log EJ I k EfI y, EJ I A, log we have
Fi=l Yilg fiwj=I Z=I YEj=I Wjkw


I- log (p(k)m+l) I I ),
=ilog (p(Xk)m). j ai(P(


Then, we get

0((Xk)m+, Z.o) -((k)m, Z.)

-- EL1 y F=l[i(P(xk)m)i + a (P(xk)m), a,(P(Xk)m+l),]

Fro eq E1) a(p(Xk)m+l).

From eq.(3.4), we know that


I
i-- (p(--11)


I
=E
i=1


yi (P((Xk)m)
(P(xk)m).


I
= EY1i
i=1


(3.24)


From eq.(3.24), we obtain


I I
ZYi -- Zao(p@(k)m+l)i 0
1=1 i=1


(3.25)


Thus, we get the desired result.

((,k)m+lZo) ((Xk)m, Zo) > 0


In the next section, an interesting property of the ML estimator is presented.


(3.23)









3.3.4 Property of the ML Estimator


Result 4: Su]


?pose there exists a solution (x*, i*) to (OP). Then,
Ei=i
= e(L ) < 1 (3.
E l e- )., -

(Pa*)e-(L*)L < 1, j 1,2,... ,J. (3.
't=1 yiLij


26)

27)


Proof:

Consider the log likelihood function


I
(t,/I) = [-(Px)ie-(LA) + y, log(Px)i y.(Ljz)]
i=1
Suppose that there exists an optimal solution (ax*,/ *) and let t be
a nonnegative scalar, and x and 1A be nonnegative vectors. Then,
(x* + tax, i* + t1A) < (x*, A*). From the above relationship, we get

Ei {Yi log[ (P*'),]+-^ tyi(LA)i

-[(Pa*), + t(PX),]e-[(L/r*)'+t(Lt&)j] + (Px *)e-(L/I)} < 0

Since (Px*)ie-(LA) > 0 for all i, we have
-If orP t(PX)jl
i= loY g[ (PX ), I
L~i\ \t ig[ (Pa;*)


- ty,(LI), [(Pa*), + t(Px),]e-[(Ljr)+t(LA)j]} < 0


(3.28)


Dividing eq.(3.28) by t, we get

{ylog[l+(p*.)]-yi(Lz),-[( 7 +(Px)i]e-[(L#*)i+t(LI)']i < 0.
=(PrX)i 0
(3.29)









Now letting t -- 0 and using the L'Hoptial's rule, we obtain

{,(Px) yY(Lt)i +(Px)(L)e-(L (PX).Ie-(L'*)} < 0.
t=i (P*^x)'
(3.30)
Since eq.(3.30) must hold for any ji > 0 and x > 0, we consider the
case where x = ej (ej is a J x 1 vector of zeros except for the jh
element which is a one) and pt = 0. For this case, eq. (3.30) becomes


pj e-(L*)Pj < 0 j=1,2,...,J. (3.31)
,ti (P *i.- '

Similarly, for x = 0 and 1t = ej, eq. (3.30) yields

I[-yiL3 + (Pa,*);e-(Lw')Li] < 0, j = 1,2,..., J. (3.32)
i=1

Note, eq. (3.31) and (3.32) are only necessary conditions for (x*,/ *)
to be a maximum of (x, 1A) over the feasible region. Since x*, i* > 0,
it follows from eq. (3.31) and eq. (3.32) that




g(L .),<1, j=1,2,..., J, (3.33)
FfI (px*)ie-(LA*) iij
S=<1, j= 1,2,...,J. (3.34)
E.i=l yiLij

In the next chapter, we introduce a new least squares reprojection attenuation
correction method.














CHAPTER 4
LEAST SQUARES REPROJECTION METHOD

In this chapter, we introduce a reprojection method, where the attenuation

density is estimated using a least squares (LS) algorithm. The algorithm is an ex-

tension of one proposed by Anderson, et. al. [4] for X-ray computed tomography.

4.1 LS Algorithm for Estimating the Attenuation Density

As discussed in Section 2.4 and 2.5, let bi denote the number of photons de-

tected in tube i during the blank scan. It is assumed that bi is an observation of

a random variable, Bi, where B, is Poisson with mean Ui > 0. It is also assumed

that the number of photons detected in tube i during the transmission scan, ri, is

an observation of a random variable, Ri, where R, is Poisson with mean Uia,(r)

(recall ai == e_(L)i). Given the observed blank scan data, bl, b2, bi, and the

observed transmission data, ri, r2,..., rI, the goal is to estimate the attenuation

density, i.

It is straightforward to show that b6 and ri are the ML estimates of Ui and

Uiai(!), respectively. Consequently, bi, Ui and r, P Uiaj(r). These approxi-

mations lead to the following one: ai, ()" Taking the logarithm on both sides

of this relationship and multiplying the result by negative one, we get (Lit), z di,

where di = -log(i(1k)). Exploiting this observation, we propose to estimate the

attenuation density by minimizing the 2-norm distance between d and (Lpf):

(P3) p jzp ,










where p(At) = E'=,(di-(L'IA)I)2, tt* is the LS estimate of t, and d = [dj, d2, ..., dr].

It is straightforward to show that p(jt) is convex over Skt.

A vector jt is a solution to (P3) if and only if it satisfies the following Kuhn-

Tucker conditions:
I
(C5) ZLj(d, (Lit)i) = 0 if j $0, j = 2,...,J,
t==l
I
(C6) -L,j(di (L)i) 0 if j = 0, j = 1,2,...,J.
i=1
As before (see Chapter 3), the first condition (C5) motivates the following fixed

point iteration:
k + 1 k i = 1 *
2j c4--3 = -1,2,...,J (4.1)
,,ji E .i (L d kjL jj j .

where yj is the kth estimate of pzj. This iteration has exactly the same form as

the iterative image space reconstruction algorithm for PET [15]. Since it has been

proven that the iterative image space reconstruction algorithm converges [40], the

proposed LS algorithm converges as well. In the next section, we discuss how the

LS estimate of the attenuation density is used to correct for attenuation.

4.2 LS Reprojection Method

Given the blank scan data and the transmission scan data, we estimate the

attenuation density using eq. (4.1). Let A be the resulting estimate. Then, the

survival probabilities are estimated by reprojecting f:

ai = e- ElLjA3 ,i = 1,2,...,j. (4.2)

Next, the estimated survival probabilities are used to create a modified probability

matrix, P, where Pij = diPij. The emission intensity is then estimated by using






48


the EMML algorithm with the modified probability matrix:

EIl_ yPi,,
SP '' ( j) l I 1 2,..., J (43)
$=1 Pij

It is straightforward to show that this iteration is equivalent to

k+1 = k. l (Pak)i
Xj X4 ,' J = 1,2,'...,J. (4.4)
2-r=1 "irij

In the next chapter, simulations are used to compare the performance of the

ML method, LS reprojection method, and EMML algorithm with corrected data.














CHAPTER 5
SIMULATION

Using synthetic data and real data, we assess the proposed methods and com-

pare them to the EMML algorithm with corrected data. Dr. John Votaw, As-

sociate Professor of Radiology at the Emory University Hospital provided the

real data. This data was collected on the Siemens-CTI ECAT EXACT 921 PET

scanner. The probabilities, Pij, and line length, Lij, were determined from the

diameter of the detector ring, number of detectors, the size of ROI (the ROI is

always a square centered in the middle of the detector ring), and the number of

voxels. Specially, the probabilities are calculated using the angle-of-view method

[46]. It should be noted that the probabilities and line lengths are calculated un-

der the assumption that the detectors are contiguous. The number of detectors

for the simulated scanner and the Siemens-CTI ECAT EXACT 921 scanner are

300 and 384, respectively. The computational expense of an iteration for the ML

method, LS reprojection method, and the EMML algorithm with corrected data

are essentially the same. Consequently, the computational complexity of the al-

gorithms can be compared by considering the total number of iterations required

by each one.









5.1 Synthetic Data

For the synthetic data, we use a cardiac emission phantom (see Fig. 5.3 (a)),

chest emission phantom (see Fig. 5.7 (a)), cardiac attenuation phantom (see Fig.

5.5 (a)), and chest attenuation phantom (see Fig. 5.8 (a)). All of the phantoms are

of size 100 x 100, and the voxel size for the cardiac phantoms and chest phantoms

are 0.422 cm and 0.437 cm, respectively. The lungs, spine, and soft tissue region

in the cardiac attenuation phantom have densities 0.025 cm-1, 0.16 cm-1, and

0.096 cm-1, respectively. The low density regions (light color region) and the

high density regions (dark color region) in the chest attenuation phantom have

densities 0.035 cm-1 and 0.095cm-1, respectively. The diameter of the detector

ring for the cardiac phantoms and chest phantoms is 59.68 cm and 61.8 cm,

respectively. The ROI for the cardiac phantoms and chest phantoms is 42.2 cm

and 43.7 cm, respectively. The number of tubes is I = 44850.

We generate the emission scan, transmission scan, and blank scan data in the

following manner. The observed coincidence for the 'th tube, yi, is the observation

of a Poisson random variable with mean (P)ia), (recall a; = e-(LI)i). The

emission phantoms and attenuation phantoms specify x and it, respectively. The

observed count for the blank scan for the 2th tube, bi, and the transmission scan

for the ith tube, ri, are Poisson random variables with means U, and UiQj(3),

respectively. We specify the total count of the transmission scan to be R. and use
Ui R ( ) = 1,2,.. I.
.r"









5.1.1 ML Method Implementation of Algorithms

The term local iteration refers to the iterations used to determine x k and ik

in Step 1 (x-step) and Step 2 (y-step), respectively (see (P1) and (P2) in Section

3.3). The term global iteration refers to a single execution of both the x-step and

the j-step. For the simulations using synthetic data, we run 8 local iterations for

the x-step, 5 local iterations for the p-step, and 2 global iterations. Thus, a total

of 21 iterations and 26 iterations are used to obtain emission intensity (i.e., x2)

and attenuation density (i.e., At2), respectively.

The ML method is initialized using (x0)j = i and ()j = 0.0214,

except that e-(LMO) is replaced with a9 = (). As state in Section 3.3.1 and

Section 3.3.2, (xk) = (xk-1)j and (k) = (k-1)1.

5.1.2 EMML Algorithm with Corrected Data

For the EMML algorithm with corrected data, the iterative equation is given

below:
?+ = X k (p i (5.1)
i= (Pl k)

where y = -. It is initialized as follows: x= for all j. We run 20

iterations of the EMML algorithm with corrected data.

5.1.3 LS Reprojection Method

To initialize the iteration in eq. (4.1), we use a uniform initial estimate: j =

0.0214 for all j. We run 20 iterations to get the LS estimate of t, A. Then, we

run 20 iterations of the algorithm shown in eq. (4.4). Thus, the LS reprojection
method requires 40 iterations to get an emission image.










The total emission count, typical transmission scan count, and short transmis-

sion scan count are 1.5 million, 0.5 million and 0.125 million, respectively. The

blank scan,duration, the typical transmission scan duration, and the short trans-

mission scan duration are 60 minutes, 20 minutes, and 5 minutes, respectively. We

simulate the ML method, LS reprojection method, and the EMML algorithm with

corrected data using typical transmission scan data with and without smoothing

and the short transmission scan data with and without smoothing. In the next

section, we show the results using non-smoothed typical transmission scan data.




5.1.4 Non-smoothed Typical Transmission Scan Data

We present the results using a non-smoothed typical transmission scan data in

this section. The emission image of the cardiac emission phantom using the ML

method after the 1st global iteration, 2nd global iteration, and 3rd global iteration

are shown in Fig. 5.1 (b), Fig. 5.1 (c), and Fig. 5.1 (d), respectively. It can be

seen that the emission images improve with increasing iteration. However, after

3 global iterations, the emission images become noisy and less accurate, which

indicates that a stopping rule needs to be found. This problem is a common one

for statistically based algorithms.

The reconstructions of the cardiac attenuation phantom using the ML method

after the 1st global iteration, 2nd global iteration, and 3rd global iteration are shown

in Fig. 5.2 (b), Fig. 5.2 (c), and Fig. 5.2 (d), respectively. In the attenuation

images, the lungs are not visible, and only the contour and the spine can be seen.

Also, there are artifacts outside the ROI.
























20 40 60 80 100
(a)









20 40 60 80 100
20 40 60 80 1O00


20 40 60 80 100
(b)











20 40 60 80 100


Figure 5.1. Reconstruction of cardiac emission phantom using non-smoothed typ-
ical transmission scan data (a) phantom (b) 1st global iteration (c) 2nd global
iteration (d) 3rd global iteration























100lo aj
20 40 60 80 100
(a)


20 40 60 80 100
(c)


20 40 60 80 100
(b)


100l


Figure 5.2. Reconstruction of cardiac attenuation phantom using non-smoothed
typical transmission scan data (a) attenuation phantom (b) 1st global iteration
(c) 2nd global iteration (d) 3rd global iteration


.I









For the cardiac emission phantom, the results of the EMML algorithm with

corrected data, the LS reprojection method, and the ML method are shown in

Fig. 5.3 (a), Fig. 5.3 (b), Fig. 5.3 (c), and Fig. 5.3(d), respectively. From

this figure, it can be seen that the ML method yields an image with better con-

trast and resolution than the EMML algorithm with corrected data and the LS

reprojection method (note the lungs and the heart at the top center of the cardiac

phantom). It also can be seen that the EMML algorithm with corrected data

does not successfully resolve either the lungs or the heart. In the emission image

of the LS reprojection method, the spine can be seen clearly. However, the LS

reprojection method also provides incorrect information as seen by the fact that

the lungs have a higher activity concentration than the soft tissues around them

(see the phantom in Fig. 5.3 (a)).

In Fig. 5.4, we show the line plots of the 50th row of the emission images in Fig.

5.3. We observe that both the LS reprojection method and the EMML algorithm

with corrected data are more biased than the ML method (note the plot between

52 and 62, which is inside the right lung area). From this line plot, we find that

the ML method not only produces emission images with better contrast but also

gives a more accurate estimate of the emission intensity than other methods.

The EMML algorithm with corrected data does not reconstruct attenuation

images, so we only show the attenuation images of the ML method and the LS

reprojection method in Fig. 5.5 (b) and Fig. 5.5 (c), respectively. The ideal

attenuation phantom is shown in Fig. 5.5 (a). We can see that the attenuation

density reconstruction of the ML method in Fig. 5.5 (b) has a lot of artifacts and























1UU -''
20 40 60 80 100


100' --
20 40 60 80 100


20 40 60 80 100
20 40 60 80 100


,di




20 40 60 80 100


Figure 5.3. Reconstruction of cardiac emission phantom using non-smoothed typ-
ical transmission scan data (a) emission phantom (b) EMML algorithm with cor-
rected data (c) LS reprojection method (d) ML method


I


^; t' v.'. ,
." *. *" '-- .
** -
-- 'j::. ^

















x 104
I I I I


SSolid: Phantom
Dash-Dash: ML Metho
Dash-Dot: LS Reproje
Dot: EM-Standard Cor







S.* .. .


.......... ................
?... ::. I
: I :
2. I
!! i ? : :! *" :


i i i/ V! ^ /- :: '. '- :
* o : l ;..




I: "*- ;; \
\,^^ / N ~: 2;.j** S -'
i V i i i,./ \ /. .
,, N.


I 1\ / / ..I I ,


d
action Method
Tection Method
















I-
2$.:





I.


0 10 20 30 40 50 60 70 80 90 100

Figure 5.4. Line plot of the 50th row of the reconstruction of cardiac emission
phantom using non-smoothed typical transmission scan data


1.4


0.8


0.6


0.4


0.2


n









does not resolve the lungs, spine, and soft tissue. The LS reprojection method

resolves lungs and soft tissues but also has artifacts.

We also show the emission images of the ML method initialized with a LS

estimate of the attenuation density. Since the yt-step is supposed to determine a

better estimate of /t, we want to check if a better initial it will correct the artifact

problem of the attenuation images and provide better emission images for the ML

method. In Fig. 5.6 (b), we can observe that spine is clearer, but the lung area is

incorrect. However, the attenuation image in Fig. 5.6 (d) is better than Fig. 5.5

(b) (resolve lungs, soft tissue but no spine).

The results of the EMML algorithm with corrected data, the LS reprojection

method and the ML method using the chest phantom are shown in Fig. 5.7 (b),

5.7 (c), and 5.7 (d), respectively. From this figure, we can observe that the ML

method and the LS reprojection method produce comparable emission images.

Both images have better contrast and resolution than the EMML algorithm with

corrected data. Note, the emission image produced by the EM algorithm with

corrected data is very noisy.

We present the attenuation images of the ML method and the LS reprojection

method for the chest phantom in Fig. 5.8 (b) and Fig. 5.8 (c), respectively. The

ideal attenuation phantom is shown in Fig. 5.8 (a). For the chest phantom, the

ML method provides better contrast and more details (big and small spots in the

center) than the LS reprojection method. The LS reprojection method fails to

resolve any information.






















Ip

20 40 60 80 100
(a)











20 40 60 80 100
(c)


20 40 60 80 100
(b)


Figure 5.5. Reconstruction of cardiac attenuation phantom using non-smoothed
typical transmission scan data (a) attenuation phantom (b) ML method (c) LS
reprojection method



























1UU ---- ---- ---- ---- ----
20 40 60 80 100
(a)



20

40-

604

80

100 .
20 40 60 80 100


1o0 o jL-
20 40 60 80 100
(b)



20

40

60 -

80

100 -...
20 40 60 80 100


Figure 5.6. ML method (a) emission phantom (b) emission image (c) attenuation
phantom (d) attenuation image


.





























20 40 60 80 100


20 40 60 80 100


100, ---
20 40 60 80 100
(b)


20 40 60 80 100


Figure 5.7. Reconstruction of chest emission phantom using non-smoothed typical
transmission scan data (a) emission phantom (b) EMML algorithm with corrected
data (c) LS reprojection method (d) ML method


J A



























1UU -
20 40 60 80 100


20 40 60 80 100
(b)


1UU --
20 40 60 80 100
(c)









Figure 5.8. Reconstruction of chest attenuation phantom using non-smoothed
typical transmission scan data (a) attenuation phantom (b) ML method (c) LS
reprojection method









In practice, the transmission scan data is smoothed even though it is ad hoc to

do so [34]. In next section, we show the results of the simulations using smoothed

transmission scan data.




5.1.5 Smoothed Typical Transmission Scan Data

Since the survival probability estimated by the ratio of transmission scan data

and blank scan data is usually noisy and biased, a popular way to address this

problem is to smooth the transmission scan data [13, 34, 41]. We use a 1 x 3 box

car filter, [1 1 '] to smooth the transmission scan data.

We show the emission images of the EMML algorithm with corrected data,

the LS reprojection method and the ML method for the cardiac phantom using

smoothed typical transmission scan data in Fig. 5.9 (b), 5.9 (c), and Fig. 5.9 (d),

respectively. Note that the LS reprojection method resolves more details of the

lungs, heart and the spine than the ML method. Both methods produce images

much better than the EMML algorithm with corrected data.

This study demonstrates that smoothing the transmission scan data improves

the quality of the emission images. In order to compare the accuracy of the

emission image, in Fig. 5.10, we show the line plots of the 50th row of the images

in Fig. 5.9. We can see that the line plot of the LS reprojection method and

the EMML algorithm with corrected data both provide worse estimates of the

emission intensity than the ML method.

We also compare the attenuation images. We show the result of the ML method

and the LS reprojection method in Fig. 5.11 (b) and Fig. 5.11 (c), respectively.





























20 40 60 80 100


20 40 60 80 100


., ." ?. | I' ""
S' <-" *-" N' ^ r.





20 40 60 80 100
(b)


Iv|W -
20 40 60 80 100


Figure 5.9. Reconstruction of cardiac emission phantom using smoothed typical
transmission scan data (a) emission phantom (b) EMML algorithm with corrected
data (c) LS reprojection method (d) ML method







65








x 104
4.5iiiii

Solid: Phantom
4 Dash-Dash: ML Method :
Dash-Dot: LS Reprojection Method
Dot: EM-Standard Correction Method
3.5-


3


2.5


2-


1.5
2 "- .: .- .

0.5 ., :

"I A' '" :" "
1... I .
..,.J ,: ." I x.. :... .,.- i

0.5 .. t ./ -,. \: /:* ". s / : *. : :
.- .' "*.. '. .. ". '1
0. .-

0
0 10 20 30 40 50 60 70 80 90 100










Figure 5.10. Line plot of the 50th row of reconstruction of cardiac emission phan-
tom using the smoothed transmission scan data









The ML method still produces an attenuation image with heavy artifacts while

the LS method provides an image with better contrast and quality. Compared to

the attenuation images using non-smoothed transmission scan data in Fig. 5.5,

both attenuation images show more features.

The emission and attenuation image of the ML method using an initial esti-

mate for IA was obtained from the LS reprojection method are shown in Fig. 5.12

(b), Fig. 5.12 (d), respectively. The ML method produces emission image with

better contrast in the heart area in than Fig. 5.9 (b). The attenuation image

provides more information than both attenuation images in Fig. 5.11 (note the

black spine at the center below).

The emission images of the EMML algorithm with corrected data, the LS

reprojection method and the ML method of the chest phantom using smoothed

typical transmission scan data are shown in Fig. 5.13 (b), 5.13 (c), and 5.13 (d),

respectively. We can see that the LS reprojection method and the ML method

produce comparable images. Both methods provide better contrast and the image

quality than the EMML algorithm with corrected data.

The attenuation images of the ML method and the LS reprojection method

using the typical smoothed transmission data for chest attenuation phantom are

shown in Fig. 5.14 (b) and Fig. 5.14 (c), respectively. The ML method resolves

lungs, a heart and soft tissue while the LS method fails to provide any information.

The attenuation image of the ML method is smoother than Fig. 5.8 (b) but the

contrast and quality do not improve.























20 40 60 80 100
(a)


20 40 60 80 100
(b)


20 40 60 80 100
(c)


Figure 5.11. Reconstruction of cardiac attenuation phantom using the smoothed
typical transmission scan data (a) attenuation phantom (b) ML method (c) LS
reprojection method


I


*4 <"<**

















60


20 40 60 80 100


20 40 60 80 100
20 40 60 80 100


1001' -
20 40 60


-I (I]


20 40 60


Figure 5.12. ML method (a) emission phantom (b) emission image (c) attenuation
phantom (d) attenuation image


4*an


I


.I4


80 100


80 100


























100. .
20 40 60 80 100


20 40 60 80 100


20 40 60 80 100


20 40 60 80 100


Figure 5.13. Emission image of the chest phantom using smoothed typical trans-
mission scan data (a) emission phantom (b) EMML algorithm with corrected data
(c) LS reprojection method (d) ML method


















20

40-

60

80-

100 .
20 40 60 80 100
(a)


20 40 60 80 100
(b)


1Ug"-L-
20 40 60 80 100
(c)


Figure 5.14. Reconstruction of chest attenuation phantom using smoothed typical
transmission scan data (a) attenuation phantom (b) ML method (c) LS reprojec-
tion method









In order to compare the methods numerically, we compute their peak-signal-

to-noise-ratio (PSNR) [43] for the different images. PSNR is defined by
X2k
PSNR(dB) = 10logo10 xpe, (5.2)
rose
1 E XJ j2X i
where Xpeak is the peak value of the phantom, mse = 2j=l(j Jj)2, x= is

the jth voxel in phantom, and ij is an estimate of xj. We compare the PSNR

of the methods using the typical transmission scan data in Table 5.1. We can


dB cardiac phantom chest phantom
Algorithm Standard ML LS-Rep. Standard ML LS-Rep.
Non-smooth -7.5 1112.5111 -5.92 -6.3 12.73 -4.8
Smooth -15.5 8.62 -5.62 -6.7 12.78 -4.56


Table 5.1. PSNR (dB) of emission images using typical transmission scan data


observe that the emission images of the ML method have the highest PSNR and

the EMML algorithm with corrected data has the lowest PSNR.

Another application of the proposed methods is the case of short duration

transmission scan data. In the next two sections, we show the results using a 5

minutes short transmission scan data with and without smoothing.




5.1.6 Non-smoothed Short Transmission Scan Data

In this section, we show the results using a 5 minutes transmission scan. The

emission images of the EMML algorithm with corrected data, the LS reprojection

method and the ML method of cardiac emission phantom using non-smoothed

typical transmission scan data are shown in Fig. 5.15 (b), 5.15 (c), and 5.15 (d),









respectively. From this figure, we know that the ML method resolves the heart

and lungs better than the the EMML algorithm with corrected data and the LS

reprojection method. The emission image of the LS reprojection method produces

incorrect information in the lungs region.

We also compare the attenuation images. We show the result of the ML method

and the LS reprojection method in Fig. 5.16 (b) and Fig. 5.16 (c), respectively.

We can see that the attenuation image of ML method in Fig. 5.16 (b) has a lot of

artifacts and only resolves spine (black spot at the center below). The attenuation

image of the LS reprojection method in Fig. 5.16 (c) shows the basic contours

better than the ML method but also suffers from the artifacts. Both methods

provide worse attenuation images when using short transmission scan data.

The emission and attenuation image of the ML method initialized using the

LS estimate of ft are shown in Fig. 5.17 (b), Fig. 5.17 (d), respectively. The ML

method resolves the heart area better than Fig. 5.15 (d) but produces the wrong

information of the lungs. The attenuation image provides more information and

less artifacts than Fig. 5.16 (c).

The results of the EMML algorithm with corrected data, the LS reprojection

method and the ML method of chest emission phantom using non-smoothed short

transmission scan data are shown in Fig. 5.18 (b), 5.18 (c), and 5.18 (d), respec-

tively. From this figure, we can observe that the ML method that produces

image with better contrast, resolution than the EMML algorithm with corrected

data and the LS reprojection method. The EMML algorithm with corrected data

and LS reprojection method both produce image with wrong information. The




























1UU .. .
20 40 60 80 100


20 40 60 80 100
20 40 60 80 1O00


1 UUl- -
20 40 60 80 100


20 40 60 80 100


Figure 5.15. Reconstruction of cardiac emission phantom using the non-smoothed
short transmission scan data (a) emission phantom (b) EMML algorithm with
corrected data (c) LS reprojection method (d) ML method


* ,' ,,a. *

























20 40 60 80 100
(a)













20 40 60 80 100
(c)


20 40 60 80 100
(b)


Figure 5.16. Reconstruction of cardiac attenuation phantom using non-smoothed
short transmission scan data (a) attenuation phantom (b) ML method (c) LS
reprojection method



















0

0


0.



0
20 40 60 80 100
(a)


40

604 1^l^^y

80

100 --.
20 40 60 80 100
(c)


20 40 60 80 100
(b)


20 40 60 80 100
(d)


Figure 5.17. ML method (a) emission phantom (b) emission image (c) attenuation
phantom (d) attenuation image


I-





























20 40 60 80 100


20

40

60

80

100 ',.. .
20 40 60 80 100


..^ '*:






20 40 60 80 100

(b)


20 40 60 80 100


Figure 5.18. Reconstruction of chest emission phantom using non-smoothed short
transmission scan data (a) emission phantom (b) EMML algorithm with corrected
data (c) LS reprojection method (d) ML method









emission image of the LS reprojection method is smoother than the one of the

EMML algorithm with corrected data.

We present the attenuation images of the ML method and the LS reprojection

method in Fig. 5.19 (b) and Fig. 5.19 (c), respectively. The LS method does not

provide too much information while the ML method provides a noisy attenuation

image. However, the ML method resolves the high density region at the center.

In the next section, we show the results of the simulations using smoothed

short transmission scan data.




5.1.7 Smoothed Short Transmission Scan Data

In this section, we show the results with short transmission scan smoothed

by the same 1 x 3 box car filter, [1 1 1]. The emission images of the EMML

algorithm with corrected data, the LS reprojection method nd the ML method of

cardiac phantom using smoothed short transmission scan data are shown in Fig.

5.20 (b), 5.20 (c), and 5.20 (d), respectively. Again, we observe that We can see

that the LS reprojection method provides the comparable emission image with

the ML method in the sense of the contrast and image quality. Note that the

LS reprojection method resolves more details of the lungs, heart and spine but

also suffers from quantitative inaccuracy. Especially, the LS reprojection method

provides clearer heart area than other methods. The ML method and the LS

reprojection method both methods produce images much better than the EMML

algorithm with corrected data.


























20 40 60 80 100
(a)












20 40 60 80 100
(c)


20 40 60 80 100
(b)


Figure 5.19. Reconstruction of chest phantom using non-smoothed short trans-
mission scan data (a) attenuation phantom (b) ML method (c) LS reprojection
method



























4 ^An


20 40 60 80 100


20

40-

60 d

80

100 .
20 40 60 80 100


Wk


20 40 60 80 100
(b)



0



0


0.

0
20 40 60 80 100
(d)


Figure 5.20. Reconstruction of cardiac emission phantom using smoothed short
transmission scan data (a) emission phantom (b) EMML algorithm with corrected
data (c) LS reprojection method (d) ML method


II










We show the attenuation images of the ML method and the LS reprojection

method in Fig. 5.21 (b) and Fig. 5.21 (c), respectively. The attenuation image

of the LS method provides the shape of the attenuation map while the image of

the ML method suffers from heavy artifacts.

The emission and attenuation image of the ML method initialized with fs

are shown in Fig. 5.22 (b), Fig. 5.22 (d), respectively. The ML method almost

produces the best emission images for the all short transmission scan experiments.

In particularly, it provides a very clear reconstruction of the heart area (see Fig.

5.9 (b)).

The emission images of the EMML algorithm with corrected data, the LS

reprojection method and the ML method of the chest emission phantom using

smoothed short transmission scan data are shown in Fig. 5.23 (b), 5.23 (c), and

5.23 (d), respectively. From this figure, we can observe that the ML method

produces an image with more details than the other images (Note, the small high

density spot area at the left below).

We present the attenuation images of the ML method and the LS reprojection

method in Fig. 5.24 (b) and Fig. 5.24 (c), respectively. We can see that the

ML method produces a more accurate attenuation image than the LS reprojection

method fails to resolve anything (note the two spots at the center). Compared to

the attenuation images using non-smoothed transmission scan data, attenuation

images here have more details and features.

The comparisons of the PSNR for the emission images using the short trans-

mission scan data are shown in Table 5.2. Again, the ML method has the largest

PSNR. In the next section, we show the results using the real data.



























20 40 60 80 100
(a)












20 40 60 80 100
(c)


20 40 60 80 100
(b)


Figure 5.21. Reconstruction of cardiac attenuation phantom using smoothed short
transmission scan data (a) attenuation phantom (b) ML method (c) LS reprojec-
tion method





















20 40 60 80 100
(a)


20 40 60 80 100
(c)


20 40 60 80 100
(b)


20 40 60 80 100
(d)


Figure 5.22. ML method (a) emission phantom (b) emission image (c) attenuation
phantom (d) attenuation image


U-


I


I


























20 40 60 80 100
(a)












20 40 60 80 100


40 -


20 40 60 80 100


20 40 60 80 100


Figure 5.23. Reconstruction of chest emission phantom using smoothed short
transmission scan data (a) emission phantom (b) EMML algorithm with corrected
data (c) LS reprojection method (d) ML method



























20 40 60 80 100
(a)


20 40 60 80 100
(b)


20 40 60 80 100
(c)









Figure 5.24. Reconstruction of chest attenuation phantom using smoothed short
transmission scan data (a) attenuation phantom (b) ML method (c) LS reprojec-
tion method


i tV ,











dB cardiac phantom chest phantom
Algorithm Standard 11 ML 11 LS-Rep. Standard ML LS-Rep.
Non-smooth -13.9 22.13 -9 -5.8 10.33 9.28
Smooth -13.62 19.85 -11 -15.2 9.02 -8.22


Table 5.2. PSNR (dB) of reconstruction of cardiac emission phantom using short
transmission scan data

5.2 Real Data

We obtained real data from Dr. John Votaw, Associate Professor of Radiology

at the Emory University Hospital. To create the data, a thorax phantom with

a hot heart, two lungs, a spine, and soft tissue was scanned using the Siemens-

CTI ECAT EXACT 921 PET scanner. This scanner is comprised of 24 rings of

diameter 82.5 cm, where each ring contains 384 detectors of size 6.35 x 6.35 mm2.

In addition, the diameter of the scanner's patient port is 56.2 cm and its axial

field of view is 16.2 cm. The field of view is a square where each side equals to

45 cm. There are J = 128 x 128 voxels in the field of view. In two-dimensional

mode, virtual rings are created by the cross-ring tubes. Since the ECAT EXACT

scanner consists of 24 rings and 23 virtual rings, it provides 47 planes of data.

Additional details on the scanner can be found in the paper of Wienhard et. al

[53].

The duration of the emission scan, transmission scan, and blank scan are 25

minutes, 25 minutes, and 60 minutes, respectively. Associated with each plane

are the sinograms for the emission data, transmission data, and the accidental

coincidence data from the emission and transmission scan. The sinograms are of









size 160 x 192 corresponding to 192 angles with 160 bins per angle. The number of
tube is I = 73536. In the next section, we discuss details of the implementation.



5.2.1 Overview of Implementation

All the images shown are for the 10th plane. This plane was chosen because
it contains most of the major features of the thorax phantom. The 10th plane
contains 0.288 million emission counts (i.e., E=1i yi = 2.88 x 105) and 13.82 million
blank scan counts. The number of counts for the low quality transmission scan
is 0.615 million (i.e., EL1 r, = 6.15 x 105 ), while the high quality transmission
scan contains 3.67 million counts (i.e., Eil ri = 3.67 x 106). It should be noted
that the transmission scans were automatically smoothed using the scanner's data
processing algorithm. The emission and transmission data have been corrected
for detector inefficiency and accidental coincidences using standard procedures,
however, no scatter correction has been done.
For the ML method, we run 2 global iterations, 7 and 5 iterations for the
x-step and jz-step, respectively. In other words, the ML method requires total 19
iterations and 24 iterations to get an emission image and an attenuation image,
respectively. The initial estimate for x-step is (x)j = fi or all where
a6 = l The initial estimate for ji-step is (p)j = 0.07 for all j. The initial
bi rr "
estimate for the survival probability is a = where rb = 60 minutes and
Tr = 25 minutes.

Twenty iterations of the EMML algorithm with corrected data method is initial-
ized using x = for all j, where = 4. For the LS reprojection method,









to initialize eq. (4.1), we use a uniform initial estimate: f0 = 0.07 for all j. We

run 20 iterations to get he LS estimate of ft, A. Then, we run 20 iterations of

the algorithm shown in eq. (4.4). Thus, LS reprojection method requires total 40

iterations to get an emission image.

The transmission scan is a high quality scan because it was performed long

after the emission scan was done (no emission counts will be mis-registered as

transmission counts).

5.2.2 Results

The emission images of the filtered back-projection method, the EMML algo-

rithm with corrected data, the LS reprojection method, and the ML method are

shown in Fig. 5.25 (a), Fig. 5.25 (b), Fig. 5.25 (c), and Fig. 5.25 (d), respec-

tively. From this figure, we observe that the EMML algorithm with corrected

data produces an image with clearer lungs boundaries of the other methods. Also,

the ML method produces comparable but smoother image than the EMML algo-

rithm with corrected data. The filtered back-projection method provides a noisy
emission image with lots of artifacts everywhere.

Except the emission images of the filtered back-projection method, all emission

images have a "ring" artifact outside the phantom (see Fig. 5.25 (c) and Fig. 5.25

(d)), which is not visible in the image using the EMML algorithm with corrected
data because the limited gray level scales of the image (the heart area in Fig. 5.25

(b) has high intensity values compared to the background, so the "ring" artifacts
on the background cannot be seen).


























20 40 60 80 100 120
(a)













20 40 60 80 100 120
(C)


20 40 60 80 100 120
(b)













20 40 60 80 100 120
(d)


Figure 5.25. Emission images using high quality smoothed transmission scan data
(a) Filtered back-projection method (b) EMML algorithm with corrected data (c)
LS reprojection method (d) ML method




Full Text
1780
1991
UNIVERSITY OF FLORIDA



PAGE 1

$77(18$7,21 &255(&7,21 0(7+2'6 )25 326,7521 (0,66,21 7202*5$3+< %\ &+(1+6,(1 :8 $ ',66(57$7,21 35(6(17(' 72 7+( *5$'8$7( 6&+22/ 2) 7+( 81,9(56,7< 2) )/25,'$ ,1 3$57,$/ )8/),//0(17 2) 7+( 5(48,5(0(176 )25 7+( '(*5(( 2) '2&725 2) 3+,/2623+< 81,9(56,7< 2) )/25,'$

PAGE 2

'HGLFDWH WR P\ SDUHQWV 0U 0LQ0LQJ :X DQG 0V -DX-U :X DQG P\ VLVWHU +ZD
PAGE 3

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

PAGE 4

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

PAGE 5

&RQFDYLW\ RI 7KH /RJ /LNHOLKRRG )XQFWLRQ ([DPSOH WR ,OOXVWUDWH &RQFDYLW\ RI WKH /RJ /LNHOLKRRG )XQFn WLRQ 2SWLPL]DWLRQ 3URFHGXUH [VWHS ILVWHS 0RQRWRQLFDOO\ ,QFUHDVLQJ 9DOXH RI WKH /RJ /LNHOLKRRG )XQFn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

PAGE 6

/,67 2) ),*85(6 4XHVW IURP *( 0HGLFDO 6\VWHPV FRXUWHV\ RI *( 0HGLFDO 6\Vn WHPVf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f SKDQWRP Ef VW JOREDO LWHUDWLRQ Ff QG JOREDO LWHUDWLRQ Gf UG JOREDO LWHUDWLRQ 5HFRQVWUXFWLRQ RI FDUGLDF DWWHQXDWLRQ SKDQWRP XVLQJ QRQVPRRWKHG W\SLFDO WUDQVPLVVLRQ VFDQ GDWD Df DWWHQXDWLRQ SKDQWRP Ef VW JOREDO LWHUDWLRQ Ff QG JOREDO LWHUDWLRQ Gf UG JOREDO LWHUDWLRQ YL

PAGE 7

5HFRQVWUXFWLRQ RI FDUGLDF HPLVVLRQ SKDQWRP XVLQJ QRQVPRRWKHG W\SLFDO WUDQVPLVVLRQ VFDQ GDWD Df HPLVVLRQ SKDQWRP Ef (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD Ff /6 UHSURMHFWLRQ PHWKRG Gf 0/ PHWKRG /LQH SORW RI WKH LO URZ RI WKH UHFRQVWUXFWLRQ RI FDUGLDF HPLVVLRQ SKDQWRP XVLQJ QRQVPRRWKHG W\SLFDO WUDQVPLVVLRQ VFDQ GDWD 5HFRQVWUXFWLRQ RI FDUGLDF DWWHQXDWLRQ SKDQWRP XVLQJ QRQVPRRWKHG W\SLFDO WUDQVPLVVLRQ VFDQ GDWD Df DWWHQXDWLRQ SKDQWRP Ef 0/ PHWKRG Ff /6 UHSURMHFWLRQ PHWKRG 0/ PHWKRG Df HPLVVLRQ SKDQWRP Ef HPLVVLRQ LPDJH Ff DWWHQXDn WLRQ SKDQWRP Gf DWWHQXDWLRQ LPDJH 5HFRQVWUXFWLRQ RI FKHVW HPLVVLRQ SKDQWRP XVLQJ QRQVPRRWKHG W\Sn LFDO WUDQVPLVVLRQ VFDQ GDWD Df HPLVVLRQ SKDQWRP Ef (00/ DOJRn ULWKP ZLWK FRUUHFWHG GDWD Ff /6 UHSURMHFWLRQ PHWKRG Gf 0/ PHWKRG 5HFRQVWUXFWLRQ RI FKHVW DWWHQXDWLRQ SKDQWRP XVLQJ QRQVPRRWKHG W\SLFDO WUDQVPLVVLRQ VFDQ GDWD Df DWWHQXDWLRQ SKDQWRP Ef 0/ PHWKRG Ff /6 UHSURMHFWLRQ PHWKRG 5HFRQVWUXFWLRQ RI FDUGLDF HPLVVLRQ SKDQWRP XVLQJ VPRRWKHG W\SLFDO WUDQVPLVVLRQ VFDQ GDWD Df HPLVVLRQ SKDQWRP Ef (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD Ff /6 UHSURMHFWLRQ PHWKRG Gf 0/ PHWKRG /LQH SORW RI WKH WO URZ RI UHFRQVWUXFWLRQ RI FDUGLDF HPLVVLRQ SKDQn WRP XVLQJ WKH VPRRWKHG WUDQVPLVVLRQ VFDQ GDWD 5HFRQVWUXFWLRQ RI FDUGLDF DWWHQXDWLRQ SKDQWRP XVLQJ WKH VPRRWKHG W\SLFDO WUDQVPLVVLRQ VFDQ GDWD Df DWWHQXDWLRQ SKDQWRP Ef 0/ PHWKRG Ff /6 UHSURMHFWLRQ PHWKRG 0/ PHWKRG Df HPLVVLRQ SKDQWRP Ef HPLVVLRQ LPDJH Ff DWWHQXDn WLRQ SKDQWRP Gf DWWHQXDWLRQ LPDJH (PLVVLRQ LPDJH RI WKH FKHVW SKDQWRP XVLQJ VPRRWKHG W\SLFDO WUDQVn PLVVLRQ VFDQ GDWD Df HPLVVLRQ SKDQWRP Ef (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD Ff /6 UHSURMHFWLRQ PHWKRG Gf 0/ PHWKRG 9OO

PAGE 8

5HFRQVWUXFWLRQ RI FKHVW DWWHQXDWLRQ SKDQWRP XVLQJ VPRRWKHG W\SLn FDO WUDQVPLVVLRQ VFDQ GDWD Df DWWHQXDWLRQ SKDQWRP Ef 0/ PHWKRG Ff /6 UHSURMHFWLRQ PHWKRG 5HFRQVWUXFWLRQ RI FDUGLDF HPLVVLRQ SKDQWRP XVLQJ WKH QRQVPRRWKHG VKRUW WUDQVPLVVLRQ VFDQ GDWD Df HPLVVLRQ SKDQWRP Ef (00/ DOn JRULWKP ZLWK FRUUHFWHG GDWD Ff /6 UHSURMHFWLRQ PHWKRG Gf 0/ PHWKRG 5HFRQVWUXFWLRQ RI FDUGLDF DWWHQXDWLRQ SKDQWRP XVLQJ QRQVPRRWKHG VKRUW WUDQVPLVVLRQ VFDQ GDWD Df DWWHQXDWLRQ SKDQWRP Ef 0/ PHWKRG Ff /6 UHSURMHFWLRQ PHWKRG 0/ PHWKRG Df HPLVVLRQ SKDQWRP Ef HPLVVLRQ LPDJH Ff DWWHQXDn WLRQ SKDQWRP Gf DWWHQXDWLRQ LPDJH 5HFRQVWUXFWLRQ RI FKHVW HPLVVLRQ SKDQWRP XVLQJ QRQVPRRWKHG VKRUW WUDQVPLVVLRQ VFDQ GDWD Df HPLVVLRQ SKDQWRP Ef (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD Ff /6 UHSURMHFWLRQ PHWKRG Gf 0/ PHWKRG 5HFRQVWUXFWLRQ RI FKHVW SKDQWRP XVLQJ QRQVPRRWKHG VKRUW WUDQVn PLVVLRQ VFDQ GDWD Df DWWHQXDWLRQ SKDQWRP Ef 0/ PHWKRG Ff /6 UHSURMHFWLRQ PHWKRG 5HFRQVWUXFWLRQ RI FDUGLDF HPLVVLRQ SKDQWRP XVLQJ VPRRWKHG VKRUW WUDQVPLVVLRQ VFDQ GDWD Df HPLVVLRQ SKDQWRP Ef (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD Ff /6 UHSURMHFWLRQ PHWKRG Gf 0/ PHWKRG 5HFRQVWUXFWLRQ RI FDUGLDF DWWHQXDWLRQ SKDQWRP XVLQJ VPRRWKHG VKRUW WUDQVPLVVLRQ VFDQ GDWD Df DWWHQXDWLRQ SKDQWRP Ef 0/ PHWKRG Ff /6 UHSURMHFWLRQ PHWKRG 0/ PHWKRG Df HPLVVLRQ SKDQWRP Ef HPLVVLRQ LPDJH Ff DWWHQXDn WLRQ SKDQWRP Gf DWWHQXDWLRQ LPDJH 5HFRQVWUXFWLRQ RI FKHVW HPLVVLRQ SKDQWRP XVLQJ VPRRWKHG VKRUW WUDQVPLVVLRQ VFDQ GDWD Df HPLVVLRQ SKDQWRP Ef (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD Ff /6 UHSURMHFWLRQ PHWKRG Gf 0/ PHWKRG 5HFRQVWUXFWLRQ RI FKHVW DWWHQXDWLRQ SKDQWRP XVLQJ VPRRWKHG VKRUW WUDQVPLVVLRQ VFDQ GDWD Df DWWHQXDWLRQ SKDQWRP Ef 0/ PHWKRG Ff /6 UHSURMHFWLRQ PHWKRG YLLL

PAGE 9

(PLVVLRQ LPDJHV XVLQJ KLJK TXDOLW\ VPRRWKHG WUDQVPLVVLRQ VFDQ GDWD Df )LOWHUHG EDFNSURMHFWLRQ PHWKRG Ef (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD Ff /6 UHSURMHFWLRQ PHWKRG Gf 0/ PHWKRG $WWHQXDWLRQ LPDJHV XVLQJ KLJK TXDOLW\ VPRRWKHG WUDQVPLVVLRQ VFDQ GDWD Df 0/ PHWKRG Ef /6 UHSURMHFWLRQ PHWKRG (PLVVLRQ LPDJH XVLQJ KLJK TXDOLW\ QRQVPRRWKHG WUDQVPLVVLRQ VFDQ GDWD Df (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD Ef 0/ PHWKRG ,;

PAGE 10

/,67 2) 7$%/(6 3615 G%f RI HPLVVLRQ LPDJHV XVLQJ W\SLFDO WUDQVPLVVLRQ VFDQ GDWD 3615 G%f RI UHFRQVWUXFWLRQ RI FDUGLDF HPLVVLRQ SKDQWRP XVLQJ VKRUW WUDQVPLVVLRQ VFDQ GDWD [

PAGE 11

$EVWUDFW RI 7KHVLV 3UHVHQWHG WR WKH *UDGXDWH 6FKRRO RI WKH 8QLYHUVLW\ RI )ORULGD LQ 3DUWLDO )XOILOOPHQW RI WKH 5HTXLUHPHQWV IRU WKH 'HJUHH RI 'RFWRU RI 3KLORVRSK\ $77(18$7,21 &255(&7,21 0(7+2'6 )25 326,7521 (0,66,21 7202*5$3+< %\ &KHQ+VLHQ :X 0D\ &KDLUPDQ 'U -RKQ 0 0 $QGHUVRQ 0DMRU 'HSDUWPHQW (OHFWULFDO DQG &RPSXWHU (QJLQHHULQJ 3RVLWURQ HPLVVLRQ WRPRJUDSK\ 3(7f LV DQ LPDJLQJ PRGDOLW\ WKDW SURYLGHV ELRn FKHPLFDO LQIRUPDWLRQ DERXW WKH KXPDQ ERG\ $PRQJ WKH VRXUFHV RI HUURU LQ 3(7 DWWHQXDWLRQ LV WKH RQH WKDW LPSDFWV LPDJH TXDOLW\ WKH PRVW 7KH VWDQGDUG PHWKRG IRU FRUUHFWLQJ DWWHQXDWLRQ FRQVLVWV RI WKH IROORZLQJ WZR VWHS SURFHGXUH )LUVW D EODQN VFDQ DQG WUDQVPLVVLRQ VFDQ DUH XVHG WR HVWLPDWH WKH VXUYLYDO SUREDELOLWLHV 7KHQ WKH GDWD IRU HDFK SDLU RI GHWHFWRUV LV GLYLGHG E\ WKH FRUUHVSRQGLQJ VXUYLYDO SUREDELOLW\ HVWLPDWH 7KLV FRUUHFWHG GDWD DQG D UHFRQVWUXFWLRQ DOJRULWKP VXFK DV [L

PAGE 12

WKH ILOWHUHG EDFNSURMHFWLRQ DOJRULWKP DUH WKHQ XVHG WR JHQHUDWH DQ HVWLPDWH RI WKH HPLVVLRQ LQWHQVLW\ ZKLFK LV GLVSOD\HG DV DQ LPDJH ,Q WKLV GLVVHUWDWLRQ WZR LWHUDWLYH PHWKRGV DUH SURSRVHG IRU FRUUHFWLQJ DWWHQn XDWLRQ HUURUV LQ 3(7 7KH ILUVW PHWKRG LV D PD[LPXP OLNHOLKRRG 0/f PHWKRG WKDW VLPXOWDQHRXVO\ HVWLPDWHV WKH HPLVVLRQ LQWHQVLW\ DQG DWWHQXDWLRQ GHQVLW\ 7KH VHFRQG RQH LV D OHDVW VTXDUHV /6f UHSURMHFWLRQ PHWKRG ZKHUH WKH VXUYLYDO SUREn DELOLWLHV DUH HVWLPDWHG E\ UHSURMHFWLQJ D /6 HVWLPDWH RI WKH DWWHQXDWLRQ GHQVLW\ 7KH HPLVVLRQ LQWHQVLW\ LV HVWLPDWHG XVLQJ WKH H[SHFWDWLRQ PD[LPL]DWLRQ PD[LPXP OLNHOLKRRG (00/f DOJRULWKP ZLWK D SUREDELOLW\ PDWUL[ WKDW KDV EHHQ PRGLILHG E\ WKH VXUYLYDO SUREDELOLW\ HVWLPDWHV %RWK PHWKRGV DUH JXDUDQWHHG WR SURGXFH QRQn QHJDWLYH HVWLPDWHV RI WKH HPLVVLRQ LQWHQVLW\ DQG WKH DWWHQXDWLRQ GHQVLW\ SURYLGHG WKDW WKH LQLWLDO HVWLPDWHV DUH QRQQHJDWLYH )URP VLPXODWLRQ VWXGLHV LW DSSHDUV WKDW WKH PDMRU DGYDQWDJH RI WKH SURSRVHG PHWKRGV RYHU WKH VWDQGDUG PHWKRG ZLWK WKH ILOWHUHG EDFNSURMHFWLRQ DOJRULWKP LV WKDW WKH\ f KDYH OHVV QRLVH f GR QRW KDYH DUWLIDFWV RXWVLGH WKH UHJLRQ RI LQWHUHVW DQG f WKH\ SURYLGH KLJK TXDOLW\ LPDJHV HYHQ IRU VKRUW WUDQVPLVVLRQ VFDQV OHVV WKDQ PLQXWHVf ;OO

PAGE 13

&+$37(5 ,1752'8&7,21 )RU ERWK FOLQLFDO DQG UHVHDUFK DSSOLFDWLRQV PHGLFDO LPDJLQJ PRGDOLWLHV DUH XVHG WR REWDLQ LPSRUWDQW LQIRUPDWLRQ RQ WKH IXQFWLRQDOLW\ DQG DQDWRP\ RI WKH KXPDQ ERG\ ,PDJLQJ PRGDOLWLHV VXFK DV ;UD\ FRPSXWHG WRPRJUDSK\ DQG PDJn QHWLF UHVRQDQFH LPDJLQJ SURYLGH GHWDLOHG LPDJHV RI WKH DQDWRPLFDO VWUXFWXUH RI WKH ERG\ +RZHYHU LQ PDQ\ DSSOLFDWLRQV LW LV PRUH EHQHILFLDO WR KDYH D PRGDOLW\ WKDW GHWHUPLQHV IXQFWLRQDO LQIRUPDWLRQ VXFK DV EORRG IORZ UDWHV R[\JHQ XWLOL]Dn WLRQ UDWHV DQG JOXFRVH XWLOL]DWLRQ UDWHV ZLWKLQ WKH UHJLRQRILQWHUHVW 52,f HJ EUDLQ KHDUWf >@ )RU H[DPSOH JOXFRVH XWLOL]DWLRQ UDWHV DUH XVHG LQ RQFRORJ\ WR GLDJQRVH WXPRUV DQG WR GHWHUPLQH WKH HIILFDF\ RI D WUHDWPHQW )XQFWLRQDO LQIRUPDWLRQ RQ WKH KXPDQ ERG\ FDQ EH REWDLQHG XVLQJ f SRVLWURQ HPLVVLRQ WRPRJUDSK\ 3(7f f VLQJOHSKRWRQ HPLVVLRQ FRPSXWHG WRPRJUDSK\ 63(&7f >@ DQG f IXQFWLRQDO PDJQHWLF UHVRQDQFH LPDJLQJ >@ $PRQJ WKHP 3(7 LV SUREDEO\ WKH PRVW IOH[LEOH LPDJLQJ PRGDOLW\ )RU H[DPSOH 3(7 FDQ LPDJH ZKLFK SDUWV RI WKH EUDLQ DUH IXQFWLRQLQJ ZKLOH D VXEMHFW LV SHUIRUPLQJ D VSHDNLQJ WDVNLQJ 7KLV LV QRW SRVVLEOH LQ IXQFWLRQDO PDJQHWLF UHVRQDQFH LPDJLQJ EHFDXVH WKH VFDQQHUV DUH WRR ORXG >@ $Q DGYDQWDJH RI 3(7 RYHU 63(&7 LV WKDW WKH UDGLRQXFOLGHV LQ 3(7 FDQ EH LQFRUSRUDWHG LQWR PROHFXOHV RU DQDORJVf RI ELRORJLFDO LQWHUHVW >@

PAGE 14

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n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

PAGE 15

)LJXUH 4XHVW IURP *( 0HGLFDO 6\VWHPV FRXUWHV\ RI *( 0HGLFDO 6\Vn WHPVf

PAGE 16

7KH ODEHOHG FRPSRXQG RU UDGLRSKDUPDFHXWLFDO LV LQWURGXFHG LQWR WKH VXEn MHFWfV ERG\ 7KH GDWD DUH FROOHFWHG $Q LPDJH LV SURGXFHG WKDW UHSUHVHQWV WKH FRQFHQWUDWLRQ RI WKH UDGLRSKDUn PDFHXWLFDO ZLWKLQ WKH 52, 5DGLRQXFOLGHV LQ 3(7 DUH f LQFRUSRUDWHG LQWR FRPSRXQGV WKDW DUH QDWXUDOO\ XVHG E\ WKH ERG\ f HPLW SRVLWURQV DQG f KDYH VKRUW KDOIOLYHV $ UDGLRQXFOLGHfV KDOIOLIH PXVW EH VXIILFLHQWO\ ORQJ VR WKDW WKHUH LV HQRXJK GDWD WR JHW TXDOLW\ LPDJHV >@ 7KH PRVW FRPPRQ UDGLRQXFOLGHV LQ 3(7 LQFOXGH Q& DQG 1 ZKLFK DUH XVHG WR ODEHO JOXFRVH ZDWHU DQG DPLQR DFLG UHVSHFWLYHO\ 7KH PRVW ZLGHO\ XVHG UDGLRSKDUPDFHXWLFDO LV )ODEHOHG GHR[\JOXFRVH )'*f ZKLFK LV XVHG WR VWXG\ JOXFRVH PHWDEROLVP %HFDXVH RI WKH VKRUW KDOIOLYHV RI SRVLWURQ HPLWWLQJ UDGLRQXFOLGHV PLQXWHV PLQXWHV PLQXWHV DQG PLQXWHV IRU Q& OV2 1 DQG ) UHVSHFWLYHO\f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

PAGE 17

$IWHU EHLQJ DGPLQLVWHUHG WR D VXEMHFW WKH ODEHOHG FRPSRXQG ZLOO ORFDOL]H LQ WKH 52, )RU H[DPSOH )'* LV DEVRUEHG SUHIHUHQWLDOO\ E\ WKH EUDLQ EHFDXVH JOXFRVH LV D VRXUFH RI HQHUJ\ >@ $QRWKHU H[DPSOH LV WKH XVH RI OV2 EDVHG UDGLRSKDUPDn FHXWLFDOV WR LQGH[ LQMXUHG WLVVXHV E\ PRQLWRULQJ WKH R[\JHQ PHWDEROLVP RI WLVVXHV >@ (PLVVLRQ 'DWD $V PHQWLRQHG SUHYLRXVO\ WKH UDGLRSKDUPDFHXWLFDO ORFDOL]HV LQ WKH 52, :KHQ WKH UDGLRQXFOLGH GHFD\V LW HPLWV SRVLWURQV ZKLFK WUDYHO D VKRUW GLVWDQFH PPf EHIRUH DQQLKLODWLQJ ZLWK QHDUE\ HOHFWURQV 7KH DUHDV ZLWK D KLJK FRQFHQWUDWLRQ RI WKH UDGLRSKDUPDFHXWLFDO HPLW PRUH SKRWRQV WKDQ WKH DUHDV RI ORZ FRQFHQWUDWLRQ :KHQ D SRVLWURQ DQQLKLODWHV ZLWK DQ HOHFWURQ WZR KLJKHQHUJ\ NH9f SKRWRQV DUH SURGXFHG WKDW IO\ RII DW DQ DOPRVW GHJUHH DQJOH IURP HDFK RWKHU 7KLV DQQLKLODWLRQ SURFHVV LV VKRZQ LQ )LJ 7KH SKRWRQV JR WKURXJK WKH VXEMHFW DQG DUH UHJLVWHUHG E\ ULQJV RI GHWHFWRUV VXUURXQGLQJ WKH VXEMHFW :KHQ WZR GHWHFWRUV GHWHFW D SDLU RI SKRWRQV ZLWKLQ D VKRUW SHULRG RI WLPH WKHVH SKRWRQV DUH VDLG WR EH LQ FRLQFLGHQFH VHH )LJ f )RU HDFK SDLU RI GHWHFWRUV WKHUH LV D FRXQW WKDW FRUUHVSRQGV WR WKH QXPEHU RI REVHUYHG FRLQFLGHQFHV 7KH HPLVVLRQ GDWD LV WKH FROOHFWLRQ RI WKHVH FRXQWV )URP WKH HPLVVLRQ GDWD ZH FDQ HVWLPDWH WKH FRQFHQWUDWLRQ RI WKH UDGLRSKDUn PDFHXWLFDO XVLQJ D UHFRQVWUXFWLRQ DOJRULWKP $ FRPSXWHU LV WKHQ XVHG WR GLVSOD\ LPDJHV ZKHUH WKH LQWHQVLW\ LV SURSRUWLRQDO WR WKH HVWLPDWHG FRQFHQWUDWLRQ RI WKH UDGLRSKDUPDFHXWLFDO ,Q WKH QH[W VHFWLRQ VRPH FOLQLFDO DQG UHVHDUFK DSSOLFDWLRQV RI 3(7 ZLOO EH SUHVHQWHG

PAGE 18

5HJLRQ RI ,QWHUHVW )LJXUH $ SRVLWURQ WUDYHOV D VKRUW GLVWDQFH DQG DQQLKLODWHV ZLWK D QHDUE\ HOHFWURQ WR SURGXFH D SDLU RI SKRWRQV WKDW IO\ RII DW GHJUHH DQJOH 7KHVH SKRWRQV DUH GHWHFWHG E\ D ULQJ RI GHWHFWRUV

PAGE 19

&OLQLFDO DQG 5HVHDUFK $SSOLFDWLRQV 3(7 LV ZLGHO\ XVHG LQ SDWKRORJ\ DQG RQFRORJ\ EHFDXVH LW FDQ TXDQWLWDWLYHO\ PHDVXUH EORRG IORZ DQG XWLOL]DWLRQ UDWHV RI JOXFRVH & DQG IDWW\ DFLG DPPRn QLD 3(7 SURYLGHV IXQFWLRQDO LQIRUPDWLRQ DERXW WLVVXHV RU RUJDQV ZKLFK FDQQRW EH REWDLQHG E\ PRGDOLWLHV VXFK DV PDJQHWLF UHVRQDQFH LPDJLQJ DQG FRPSXWHG WRn PRJUDSK\ >@ ,W LV DOVR D SRZHUIXO WRRO IRU LQYHVWLJDWLQJ EUDLQ IXQFWLRQV VXFK DV HPRWLRQ WKLQNLQJ YLVLRQ DQG VSHHFK > @ 6LQFH WKH PDMRU HQHUJ\ VRXUFH RI WKH EUDLQ LV JOXFRVH )'* LV XVHG WR PDS WKH EUDLQfV HQHUJ\ FRQVXPSWLRQ ,Q RWKHU ZRUGV LI D VSHFLILF UHJLRQ RI WKH EUDLQ KDV LQFUHDVHG IXQFWLRQDOLW\ WKHQ WKH UHJLRQDO EORRG IORZ LQFUHDVHV WR SURYLGH WKH QHFHVVDU\ HQHUJ\ $V D UHVXOW WKH FRQn FHQWUDWLRQ RI WKH UDGLRQXFOLGH XVHG LQ )'* LQFUHDVHV LQ WKDW UHJLRQ 8VLQJ 3(7 WR PDS WKH FRQFHQWUDWLRQ RI WKH UDGLRQXFOLGH ZH FDQ REVHUYH WKH KLJKHU IXQFWLRQn DOLW\ LQ WKDW UHJLRQ RI WKH EUDLQ %DVHG RQ WKH FRQFHSWV GLVFXVVHG DERYH VFLHQWLVWV KDYH XVHG 3(7 LPDJHV WR DVVHVV WKH EUDLQV RI QRUPDO FKLOGUHQ DQG FKLOGUHQ XQGHU DEXVH 7KH\ IRXQG WKDW WKH FKLOGUHQ ZKR ZHUH DEXVHG KDG VORZHU GHYHORSPHQWV LQ EUDLQ IXQFWLRQV VXFK DV VSHHFK DQG YLVLRQ DV FRPSDUHG WR QRUPDO FKLOGUHQ >@ 3(7 KDV FOLQLFDO DSSOLFDWLRQV DV ZHOO )RU H[DPSOH 3(7 KDV WKH DELOLW\ WR PHDVXUH WKH VL]H RI LQIDUFWV LQ WKH KHDUW DIWHU D FRURQDU\ DWWDFN WR GLDJQRVH GLVn HDVHV VXFK DV $O]KHLPHUfV >@ DQG HSLOHSWLF VHL]XUHV >@ DQG WR VXUYH\ WKH HIIHFWV RI WUHDWPHQWV VXFK DV UDGLDWLRQ WKHUDS\ FKHPRWKHUDS\ DQG KRUPRQH WKHUDS\ ,Q DGGLWLRQ ZKROHERG\ 3(7 LV XVHG WR VSRW PHWDVWDVHV ZKLFK LV YHU\ GLIILFXOW IRU RWKHU PRGDOLWLHV EHFDXVH RI WKH VPDOO VL]HV DQG WKH UDQGRP RFFXUUHQFHV RI PHWDVn WDVHV LQVLGH FDQFHU SDWLHQWVf ERGLHV 7KH HIIHFW RI WKH FKHPRWKHUDS\ RQ FDQFHU

PAGE 20

SDWLHQWV FDQ EH HYDOXDWHG E\ YLVXDOL]LQJ WKH VL]HV DQG ORFDWLRQV RI PHWDVWDVHV RQ 3(7 LPDJHV >@ 3(7 LV DOVR XVHG LQ SV\FKRORJLFDO VWXGLHV )RU LQVWDQFH LW ZDV XVHG LQ WKH HDUO\ V VHYHUDO WLPHV E\ FRXUWV LQ &DOLIRUQLD WR GHWHUPLQH ZKHWKHU QHXUDO PDOIXQFWLRQV LQ WKH EUDLQ FDXVHG D VXVSHFWfV YLROHQW EHKDYLRU $QRWKHU LQWHUHVWLQJ H[SHULPHQW ZDV WKH PRQLWRULQJ RI VL[ EDQN RIILFLDOVf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n FORWURQ XVHG WR JHQHUDWH WKH UDGLRQXFOLGHV LV DOVR YHU\ H[SHQVLYH $GGLWLRQDOO\ DW OHDVW WKUHH VNLOOHG SURIHVVLRQDOV RQH WR PDLQWDLQ WKH VFDQQHU RQH WR UXQ WKH

PAGE 21

F\FORWURQ DQG DQRWKHU RQH WR SURGXFH UDGLRQXFOLGHV DUH QHHGHG WR RSHUDWH D 3(7 ODERUDWRU\ )RU 63(&7 LW LV QRW QHFHVVDU\ WR FUHDWH WKH UDGLRQXFOLGHV RQ VLWH EHFDXVH WKH UDGLRQXFOLGHV XVHG LQ 63(&7 FDQ EH VWRUHG (YHQ WKRXJK LWV LPDJH TXDOLW\ LV QRW DV JRRG DV WKDW RI 3(7 63(&7 LV PRUH ZLGHO\ DYDLODEOH LQ SUDFWLFH EHFDXVH LW LV VLJQLILFDQWO\ FKHDSHU

PAGE 22

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f LV VXUURXQGHG E\ ULQJV RI GHWHFWRUV WKDW GHWHFW WKH SKRWRQV :H ZLOO XVH WKH WHUP WXEH WR UHIHU WR D SDLU RI GHWHFWRUV :KHQ D SDLU RI SKRWRQV LV LQFLGHQW RQ D WXEH ZLWKLQ D VPDOO WLPH LQWHUYDO D FRXQWHU LV LQFUHDVHG VHH )LJ f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

PAGE 23

5LQJ RI 'HWHFWRUV &RLQFLGHQFH 7LPLQJ &LUFXLW )LJXUH $QQLKLODWLRQ 3URFHVV

PAGE 24

VHSWD DUH ZLWKGUDZQ WR SHUPLW SKRWRQV WR EH UHJLVWHUHG E\ LQSODQH WXEHV DQG FURVVSODQH WXEHV >@ $ GHWHFWRU PRGXOH FRQVLVWV RI DQ DUUD\ RI FU\VWDOV FRXSOHG WR VHYHUDO SKRWRPXOn WLSOLHU WXEHV ,Q )LJ D GHWHFWRU PRGXOH RI WKH 6FDQGLWURQL[ 3&:% ZKROHERG\ 3(7 VFDQQHU LV VKRZQ $ GHWHFWRU ULQJ FRQWDLQV VHYHUDO KXQGUHG GHn WHFWRU PRGXOHV HJ IRU WKH 6LHPHQV&7, (&$7 (;$&7 VFDQQHUf :KHQ D FU\VWDO LV KLW E\ D NH9 SKRWRQ LW LQWHUDFWV ZLWK WKH SKRWRQ DQG JHQHUDWHV QXn PHURXV OLJKW SKRWRQV 7KHQ WKH OLJKW JXLGH FRQQHFWHG WR WKH FU\VWDO WUDQVIHUV WKH OLJKW SKRWRQV WR WKH QHDUHVW SKRWRPXOWLSOLHU WXEH 307f 7KH 307V FROOHFW OLJKW SKRWRQV DQG FRQYHUW WKHP LQWR HOHFWULFDO VLJQDOV WKDW DUH VWRUHG LQ D FRPSXWHU 7KH RXWSXWV RI WKH 307V DUH GHFRGHG WR GHWHUPLQH ZKLFK GHWHFWRU PRGXOH ZDV KLW E\ D SKRWRQ 7KH PRVW ZLGHO\ XVHG FU\VWDO PDWHULDO LV %LVPXWK*HUPDQDWH EHFDXVH LW KDV JRRG SURSHUWLHV ZKHQ LW LQWHUDFWV ZLWK NH9 SKRWRQV VXFK DV D KLJK OLJKW SKRWRQ SURGXFLQJ UDWH VKRUW GHFD\ WLPH DQG D IDVW HQHUJ\ ULVH WLPH >@ 2QFH WKH GDWD LV FROOHFWHG DOJRULWKPV DUH XVHG WR UHFRQVWUXFW WKH 3(7 LPn DJHV 7KH PRVW SRSXODU PHWKRG LV WKH ILOWHUHG EDFNSURMHFWLRQ DOJRULWKP >@ ,W LV H[WUHPHO\ IDVW DQG HDV\ WR LPSOHPHQW EXW WKH LPDJHV DUH QRLV\ KDYH QHJDn WLYH YDOXHV DQG FRQWDLQ DUWLIDFWV RXWVLGH WKH 52, $OWHUQDWLYH DSSURDFKHV LQFOXGH VWDWLVWLFDOO\ EDVHG PHWKRGV VXFK DV PD[LPXP OLNHOLKRRG > @ PD[LPXP D SRVWHULRU > @ OHDVW VTXDUHV > @ DQG ZHLJKWHG OHDVW VTXDUHV > @ PHWKRGV $Q DGYDQWDJH RI VWDWLVWLFDOO\ EDVHG PHWKRGV LV WKDW WKH\ DFFRXQW IRU WKH HUURUV LQ WKH GDWD DQG GR QRW KDYH DUWLIDFWV RXWVLGH WKH 52, $ GLVDGYDQWDJH RI WKHP LV WKDW WKH\ DUH FRPSXWDWLRQDOO\ GHPDQGLQJ

PAGE 25

307 &U\VWDOV )LJXUH 6FDQGLWURQL[ 3&:% GHWHFWRU PRGXOH FU\VWDOV OLJKW JXLGH DQG GXDO FDWKRGH 307V

PAGE 26

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">7@ DQG LV WKH QXPEHU RI WXEHV ,W LV DOVR DVVXPHG WKDW f ^1M`a DUH LQGHSHQGHQW f ^[L [ f f f DM@7 ,Q WKLV ZD\ DQ HVWLPDWH IRU WKH QXPEHU RI SKRWRQV HPLWWHG LQ HDFK YR[HO LV REWDLQHG 7KURXJKRXW WKH UHPDLQGHU RI WKLV GLVVHUWDWLRQ [ ZLOO EH UHIHUUHG WR DV WKH HPLVVLRQ LQWHQVLW\

PAGE 27

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n WLRQ %DFNJURXQG RQ $WWHQXDWLRQ $WWHQXDWLRQ LV WKH PDLQ HUURU VRXUFH LQ 3(7 HPLVVLRQ LPDJH UHFRQVWUXFWLRQ >@ 7KH NH9 SKRWRQV SURGXFHG E\ WKH DQQLKLODWLRQV DUH DWWHQXDWHG E\ WLVVXH ERQH DQG DLU 'LIIHUHQW PHGLXPV KDYH GLIIHUHQW DWWHQXDWLRQ HIIHFWV %HIRUH JRLQJ LQWR WKH GHWDLOV RI DWWHQXDWLRQ FRUUHFWLRQ IRU 3(7 ZH LQWURGXFH WKH SK\VLFV EHKLQG DWWHQXDWLRQ 7KH PDLQ VRXUFHV RI DWWHQXDWLRQ LQ 3(7 DUH DEVRUSWLRQ DQG &RPSWRQ VFDWWHUn LQJ 3KRWRQV DUH UDUHO\ DEVRUEHG EHFDXVH WKH\ FDUU\ NH9 RI HQHUJ\ +HQFH

PAGE 28

)LJXUH $WWHQXDWLRQ GXH WR DEVRUSWLRQ

PAGE 29

DWWHQXDWLRQ GXH WR DEVRUSWLRQ LV QHJOLJLEOH )LJ LV D GHSLFWLRQ RI SKRWRQ DEn VRUSWLRQ &RPSWRQ VFDWWHULQJ RFFXUV ZKHQ WKH SDWK RI WKH SKRWRQ LV GHIOHFWHG E\ DQ RXWHU VKHOO HOHFWURQ >@ 3KRWRQV LQ 3(7 DUH PRUH OLNHO\ WR XQGHUJR &RPSWRQ VFDWWHULQJ WKDQ DEVRUSWLRQ 0RVW RI WKH SKRWRQV WKDW XQGHUJR &RPSWRQ VFDWWHUn LQJ DUH QRW GHWHFWHG )LJ VKRZV D SKRWRQ XQGHUJRLQJ &RPSWRQ VFDWWHULQJ 7KH WRWDO HIIHFW RI SKRWRQ DEVRUSWLRQ DQG &RPSWRQ VFDWWHULQJ LV UHIHUUHG WR DV DWWHQXDWLRQ 7KH VXUYLYDO SUREDELOLW\ DORQJ D OLQH / D LV WKH SUREDELOLW\ WKDW D SKRWRQ LV QRW DWWHQXDWHG DORQJ WKDW OLQH ,W FDQ EH VKRZQ WKDW WKH VXUYLYDO SUREDELOLW\ LV HTXDO WR D HaIAGO f ZKHUH S LV WKH DWWHQXDWLRQ GHQVLW\ >@ 6LQFH HTf LV D IXQGDPHQWDO HTXDWLRQ LQ DWWHQXDWLRQ ZH SURYLGH DQ H[SODn QDWLRQ RI LW &RQVLGHU )LJ ZKLFK VKRZV D OLQH VRXUFH HPLWWLQJ SKRWRQV WRZDUG DQ REMHFW ZLWK DWWHQXDWLRQ GHQVLW\ S :H DVVXPH WKDW 1 SKRWRQV LQFLGHQW RQ D SRLQW ZLWKLQ WKH REMHFW WUDYHO D VKRUW GLVWDQFH $ DQG RQO\ 1 $1 SKRWRQV VXUYLYH 2EYLRXVO\ $1 LV D QHJDWLYH QXPEHU :H PD\ XVH WKH IROORZLQJ HTXDWLRQ WR GHVFULEH WKH DERYH SURFHVV $1 1 $ O 9 f f

PAGE 30

)LJXUH $WWHQXDWLRQ GXH WR VFDWWHULQJ

PAGE 31

6RXUFH $ 1 1$1 7LVVXH $1 1 $ )LJXUH %DVLF FRQFHSW RI DWWHQXDWLRQ

PAGE 32

$V $O DQG $1 JR WR ]HUR ZH JHW WKH GLIIHUHQWLDO HTXDWLRQ G1 1 GO a r n f ZKLFK FDQ EH VROYHG E\ LQWHJUDWLQJ DORQJ WKH OLQH / 'RLQJ VR ZH KDYH WKDW f$IRXW G1 1LQ 1 f ZKHUH L9Q LV WKH QXPEHU RI SKRWRQV WKDW JR LQWR WKH REMHFW DQG 1RXW LV WKH QXPEHU RI SKRWRQV WKDW FRPH RXW RI WKH REMHFW )URP WKH DERYH HTXDWLRQ ZH JHW ,Q 1RXW ,Q 1^Q f§ f ,W IROORZV IURP HT f WKDW Hf§ 1Q f )LQDOO\ HT f IROORZV E\ QRWLQJ WKDW WKH OHIW VLGH RI f LV WKH SUREDELOLW\ WKDW D SKRWRQ DORQJ WKH OLQH / LV QRW DWWHQXDWHG ,Q WKH QH[W VHYHUDO VHFWLRQV ZH GLVFXVV VRPH RWKHU HUURUV LQ 3(7 $FFLGHQWDO &RLQFLGHQFHV ,I WZR SKRWRQV IURP GLIIHUHQW DQQLKLODWLRQV DUH UHJLVWHUHG DV D FRLQFLGHQFH E\ D WXEH WKHQ DQ HUURU RFFXUV 7KLV NLQG RI HUURU LV UHIHUUHG WR DV DQ DFFLGHQWDO FRLQFLGHQFH >@ )LJ VKRZV DQ H[DPSOH RI DQ DFFLGHQW FRLQFLGHQFH 7ZR DQQLKLODWLRQV KDSSHQ LQVLGH WKH 52, +RZHYHU LQ ERWK DQQLKLODWLRQV D SKRWRQ IOLHV RXW RI WKH GHWHFWRU ULQJ 7KH UHPDLQLQJ SKRWRQV DUH GHWHFWHG E\ D WXEH ZLWKLQ WKH FRLQFLGHQFH ZLQGRZ DQG LQFRUUHFWO\ UHFRUGHG DV D FRLQFLGHQFH

PAGE 33

)LJXUH $FFLGHQWDO FRLQFLGHQFH LQ 3(7

PAGE 34

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n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f

PAGE 35

)LJXUH 6FDWWHU LQ 3(7

PAGE 36

2WKHU (UURUV ,Q UHDOLW\ SRVLWURQV WUDYHO D VKRUW GLVWDQFH PPf EHIRUH DQQLKLODWLQJ ZLWK DQ HOHFWURQ DQG SKRWRQV VKRRW RII DW DQ DQJOH OHVV WKDQ r $OVR SKRWRQV FDQ SHQHWUDWH WKH FU\VWDO WKH\ KLW DQG OHDN HQHUJ\ WR D QHLJKERULQJ WXEH >@ $QRWKHU W\SH RI HUURU FDOOHG GHWHFWRU LQHIILFLHQF\ RFFXUV ZKHQ D GHWHFWRU IDLOV WR GHWHFW DQ LQFLGHQW SKRWRQ >@ )RUWXQDWHO\ PRVW RI WKH WLPH WKHVH HUURUV DUH VPDOO DQG XVXDOO\ FDQ EH LJQRUHG RU FRUUHFWHG ,Q RUGHU WR FRUUHFW IRU DWWHQXDWLRQ FHUWDLQ VFDQV FDOOHG EODQN VFDQV DQG WUDQVn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f 7KH SKRWRQ FRXQWV FROOHFWHG LQ D WUDQVPLVVLRQ VFDQ DUH WKH RQHV WKDW VXUYLYH WKH DWWHQXDWLRQ HIIHFW RI WKH 52,

PAGE 37

)LJXUH %ODQN VFDQ

PAGE 38

/HW U GHQRWH WKH QXPEHU RI REVHUYHG FRLQFLGHQFHV GXULQJ WKH WUDQVPLVVLRQ VFDQ IRU WXEH  $ WUDQVPLVVLRQ VFDQ QHHGV WR EH SHUIRUPHG HDFK WLPH D QHZ SDWLHQW LV SXW LQ WKH 3(7 VFDQQHU 7KH GXUDWLRQ RI D WUDQVPLVVLRQ VFDQ UDQJHV IURP DERXW WR PLQXWHV )RU WKH VDIHW\ DQG FRPIRUW RI WKH SDWLHQW WKH WUDQVPLVVLRQ VFDQV DUH NHSW DV VKRUW DV SRVVLEOH SDWLHQW VKRXOG QRW EH H[SRVHG WR WKH UDGLRDFWLYH VRXUFH WRR ORQJf ,QWXLWLYHO\ LW LV UHDVRQDEOH WR HVWLPDWH WKH VXUYLYDO SUREDELOLW\ E\ WDNLQJ WKH UDWLR RI WKH WUDQVPLVVLRQ VFDQ FRXQW SHU XQLW WLPH DQG WKH EODQN VFDQ FRXQW SHU XQLW WLPH +HQFH DQ HVWLPDWH RI WKH VXUYLYDO SUREDELOLW\ LQ WXEH L £ LV D $ ‘ r Y 7 2WL N f f ZKHUH X DQG UU DUH WKH GXUDWLRQV RI WKH EODQN VFDQ DQG WKH WUDQVPLVVLRQ VFDQ UHVSHFWLYHO\ 7KH GDWD FDQ WKHQ EH FRUUHFWHG IRU DWWHQXDWLRQ E\ GLYLGLQJ WKH GDWD E\ WKH HVWLPDWHG VXUYLYDO SUREDELOLWLHV LH Af 7KLV DSSURDFK LV VLPSOH DQG LQWXLWLYHO\ DSSHDOLQJ EXW LW GRHV QRW ZRUN ZHOO LI WKH WUDQVPLVVLRQ VFDQ LV WRR VKRUW 2QH ZD\ WR UHVROYH WKDW SUREOHP LV WR PRGLI\ WKH 3RLVVRQ PRGHO LQVWHDG RI GLUHFWO\ FRUUHFWLQJ WKH GDWD ,Q QH[W VHFWLRQ ZH LQWURGXFH WKH PRGLILHG 3RLVVRQ PRGHO 0RGLILHG 3RLVVRQ 0RGHO ,Q WKLV VHFWLRQ ZH GLVFXVV DQ H[WHQVLRQ RI WKH 3RLVVRQ PRGHO IRU WKH FDVH RI DWWHQXDWLRQ HUURUV RQO\ >@ %HIRUH GLVFXVVLQJ WKLV PRGHO ZH LQWURGXFH WKH FRQFHSW RI WKLQQLQJ D 3RLVVRQ SURFHVV

PAGE 39

)LJXUH 7UDQVPLVVLRQ VFDQ

PAGE 40

5HVXOW /HW 1 EH D 3RLVVRQ UDQGRP YDULDEOH ZLWK PHDQ [ 6XSSRVH D UDQGRP YDULDEOH ; LV FUHDWHG E\ WKLQQLQJ 1 ZLWK SUREDELOLW\ S 7KHQ ; LV 3RLVVRQ ZLWK PHDQ S[ 3URRI 3^; Nf  3; N?1 Pf31 Pf P N RR (*983\PaN3^1 UQf Pf§N PO ( WN N@P Nf S L Sfn LaNn PO Ha[SN a SfPN[P N KN P Nf 5HSODFLQJ P f§ N ZLWK Q \LHOGV 3; Nf Ha[SN J SfQ[QN N? Q Q? Hr[SfN aOSf[U ,G  Q? Ha[[SfN BSf[ N? Ha[S[SfN ,G f ZKLFK VKRZV WKDW ; LV D 3RLVVRQ UDQGRP YDULDEOH ZLWK PHDQ S[ :H FDQ DSSO\ 5HVXOW WR PRGHO WKH HIIHFW RI DWWHQXDWLRQ LQ 3(7 VHH DOVR >@f /HW SM EH WKH DWWHQXDWLRQ GHQVLW\ ZLWKLQ WKH MWK YR[HO ZKHUH LW LV DVVXPHG WKDW WKH DWWHQXDWLRQ GHQVLW\ ZLWKLQ HDFK YR[HO LV FRQVWDQW /HW / EH DQ [ PDWUL[ ZKRVH LMfWK HOHPHQW LV WKH LQWHUVHFWLRQ OHQJWK RI WKH LWK WXEH DQG

PAGE 41

WKH MWK YR[HO 7KHQ LW IROORZV IURP HTf WKDW WKH SUREDELOLW\ WKDW D SKRWRQ SDLU LV GHWHFWHG E\ WXEH L LV DSSUR[LPDWHO\ HTXDO WR Hr!m f ZKHUH =8 / MILM L %HFDXVH RI DWWHQXDWLRQ WKH GDWD LQ WXEH L
PAGE 42

ZKHUH D LV JLYHQ E\ HT f 7KLV PHWKRG LV HDV\ WR LPSOHPHQW DQG ZRUNV IDLUO\ ZHOO IRU ORQJ WUDQVPLVVLRQ VFDQV PLQXWHVf IRU VKRUW WUDQVPLVVLRQ VFDQV OHVV WKDQ PLQXWHVf LW SURGXFHV QRLV\ HPLVVLRQ LPDJHV 7KH FRQYHQWLRQDO ZD\ WKLV SUREOHP LV DGGUHVVHG LV WR VPRRWK WKH WUDQVPLVVLRQ VLQRJUDP 6PRRWKLQJ UHGXFHV WKH QRLVH EXW LW DOVR OHVVHQV WKH TXDQWLWDWLYH DFFXUDF\ RI WKH LPDJHV >@ 5HSURMHFWLRQ 0HWKRGV *LYHQ WKH EODQN VFDQ GDWD DQG WKH WUDQVPLVVLRQ VFDQ GDWD DQ HVWLPDWH RI WKH DWWHQXDWLRQ GHQVLW\ LV REWDLQHG > @ 7KHQ WKH HVWLPDWH W LV fUHSURMHFWHGf WR REWDLQ DQ HVWLPDWH RI WKH VXUYLYDO SUREDELOLWLHV G  f 7KH HVWLPDWHG VXUYLYDO SUREDELOLWLHV DUH XVHG WR FRUUHFW WKH GDWD RU PRGLI\ WKH SUREDELOLW\ PDWUL[ 3 ,Q DGGLWLRQ WR LPSURYLQJ WKH DFFXUDF\ RI WKH HVWLPDWHG VXUYLYDO SUREDELOLWLHV WKH WUDQVPLVVLRQ LPDJHV SURYLGH LQIRUPDWLRQ WKDW FDQ EH XVHG WR SRVLWLRQ SDWLHQWV HVWLPDWH VFDWWHULQJ HUURUV DQG WR SURYLGH DQDWRPLFDO ODQGPDUNV WKDW FDQQRW EH VHHQ LQ HPLVVLRQ LPDJHV >@ 7KH UHSURMHFWLRQ PHWKRG SURYLGHV D EHWWHU HVWLPDWH RI WKH VXUYLYDO SUREDELOn LWLHV +RZHYHU WKH FRXQWLQJ QRLVH IURP WKH WUDQVPLVVLRQ LPDJH LQWURGXFHV DGGLn WLRQDO HUURU LQWR WKH HPLVVLRQ UHFRQVWUXFWLRQ ,Q UHJLRQV VXFK DV WKH WKRUD[ ZKHUH WKH DWWHQXDWLRQ PHGLXP LV KHWHURJHQHRXV WKLV SUREOHP FDQQRW EH UHVROYHG E\

PAGE 43

VLPSO\ VPRRWKLQJ WKH HVWLPDWHG VXUYLYDO SUREDELOLWLHV EHFDXVH LW LQWURGXFHV ELDV >@ 6LQFH UHSURMHFWLRQ PHWKRGV HVWLPDWH WKH DWWHQXDWLRQ GHQVLW\ WKH\ DUH PRUH FRPSXWDWLRQDOO\ GHPDQGLQJ WKDQ WKH VWDQGDUG FRUUHFWLRQ PHWKRG ,Q WKH QH[W VHFWLRQ ZH GLVFXVV PHWKRGV WKDW VLPXOWDQHRXVO\ HVWLPDWH WKH DWWHQXDWLRQ GHQVLW\ IRU VXUYLYDO SUREDELOLWLHV DQG WKH HPLVVLRQ LQWHQVLW\ 6LPXOWDQHRXV (VWLPDWLRQ 0HWKRGV ,Q >@ 3ROLWWH DQG 6Q\GHU SURSRVHG FRQVWUDLQHG PD[LPXP OLNHOLKRRG UHFRQn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n XDWLRQ GHQVLW\ DQG WKH HPLVVLRQ LQWHQVLWLHV IRU 63(&7 XVLQJ WKH &\FOLF 6XEJUDGLn HQW 3URMHFWLRQ &63f DOJRULWKP $IWHU WKH 3RLVVRQ PRGHO ZDV SURSRVHG )HVVOHU &OLQWKRUQH DQG 5RJHUV SURSRVHG D MRLQW PD[LPXP OLNHOLKRRG PHWKRG WKDW HVWLn PDWHV WKH HPLVVLRQ LQWHQVLW\ DQG VXUYLYDO SUREDELOLWLHV > @ /X SURSRVHG D 0$3 DOJRULWKP WR HVWLPDWH WKH HPLVVLRQ LQWHQVLW\ DQG WKH DWWHQXDWLRQ GHQVLW\ IRU 63(&7 >@ +RZHYHU SULRU LQIRUPDWLRQ DERXW WKH DWWHQXDWLRQ PHGLXP PXVW EH NQRZQ

PAGE 44

7R VLPXOWDQHRXVO\ HVWLPDWH WKH DWWHQXDWLRQ GHQVLW\ DQG WKH HPLVVLRQ LQWHQVLW\ ZH SURSRVH D QHZ PHWKRG ,Q WKH QH[W FKDSWHU ZH GLVFXVV WKLV SURSRVHG PHWKRG

PAGE 45

&+$37(5 0$;,080 /,.(/,+22' 0(7+2' ,Q WKLV FKDSWHU ZH SUHVHQW DQ DWWHQXDWLRQ FRUUHFWLRQ PHWKRG WKDW MRLQWO\ HVWLn PDWHV WKH DWWHQXDWLRQ GHQVLW\ DQG HPLVVLRQ LQWHQVLW\ XVLQJ D PD[LPXP OLNHOLKRRG 0/f PHWKRG 2WKHU 0/ PHWKRGV KDYH EHHQ SURSRVHG > @ EXW WKH\ KDYH WKH GLVDGYDQWDJH RI UHTXLULQJ PRUH XQNQRZQV WKDQ GDWD SRLQWV %\ FRQWUDVW WKH SURSRVHG PHWKRG KDV IHZHU XQNQRZQV WKDQ GDWD SRLQWV IRU W\SLFDO VFDQQHUV 6SHFLILFDOO\ LW HVWLPDWHV XQNQRZQV IURP GDWD SRLQWV ZKHUH DQG DUH WKH QXPEHU RI WKH YR[HOV DQG WKH QXPEHU RI WKH WXEHV UHVSHFWLYHO\ $ FRPPRQ FKRLFH IRU WKH QXPEHU RI YR[HOV LV LH [ f DQG IRU WKH 6LHPHQV &7, (&$7 (;$&7 3(7 VFDQQHU WKH QXPEHU RI WKH WXEHV LV 0D[LPXP /LNHOLKRRG (VWLPDWRU $V GLVFXVVHG LQ 6HFWLRQ WKH HPLVVLRQ GDWD IRU WKH "WK WXEH W LV DVVXPHG WR EH DQ REVHUYDWLRQ RI D UDQGRP YDULDEOH ZKHUH
PAGE 46

,W LV ZHOO NQRZQ WKDW LQ PDQ\ HVWLPDWLRQ SUREOHPV 0/ HVWLPDWRUV DUH FRQVLVn WHQW DQG DV\PSWRWLFDOO\ XQELDVHG >@ 7KXV ZH SURSRVH WR HVWLPDWH WKH XQNQRZQ SDUDPHWHUV XVLQJ D 0/ PHWKRG ,Q RWKHU ZRUGV ZH HVWLPDWH WKH HPLVVLRQ LQWHQVLW\ DQG DWWHQXDWLRQ GHQVLW\ E\ PD[LPL]LQJ WKH ORJ OLNHOLKRRG IXQFWLRQ p]f >[r ]r@ QDAF []f ZKHUH [ ]f ORJ 3>3[f[H/:r@XS ? /QK@\! ORJ Q e / f W L \r >f§ \L ORJ3[fc f§ \c/ILfc f§ ORJJ-f@ f (T f IROORZV IURP WKH DVVXPSWLRQ WKDW WKH GDWD LV LQGHSHQGHQW 1RWH VLQFH WKH ODVW WHUP LQ HT f GRHV QRW FRQWDLQ [ RU ] LW FDQ EH LJQRUHG LQ WKH PD[LPL]DWLRQ SURFHGXUH 7KXV WKH DERYH RSWLPL]DWLRQ SUREOHP LV HTXLYDOHQW WR WKH IROORZLQJ RQH 23f >[r ILr@ DUA[QDA If^[Qf ZKHUH 3f <+a3[fLHa^/IOfn 9L ORJSrf a \L/ILfL@ f 7KH IXQFWLRQ A!p]f ZLOO EH UHIHUUHG WR DV WKH ORJ OLNHOLKRRG IXQFWLRQ IRU WKH UHPDLQGHU RI WKLV GLVVHUWDWLRQ

PAGE 47

7KH SUREOHP 23f LV D GLIILFXOW FRQVWUDLQHG RSWLPL]DWLRQ SUREOHP WKDW PXVW EH VROYHG LQ RUGHU WR REWDLQ WKH 0/ HVWLPDWH RI WKH DWWHQXDWLRQ GHQVLW\ DQG WKH HPLVVLRQ LQWHQVLW\ ,Q WKH QH[W VHFWLRQ ZH GLVFXVV VRPH SURSHUWLHV RI WKH ORJ OLNHOLKRRG IXQFWLRQ EHIRUH SUHVHQWLQJ RXU PHWKRG IRU VROYLQJ 23f &RQFDYLW\ RI 7KH /RJ /LNHOLKRRG )XQFWLRQ 8QIRUWXQDWHO\ WKH ORJ OLNHOLKRRG IXQFWLRQ Wf LV QRW FRQFDYH RYHU WKH IHDVLEOH UHJLRQ 6[IL ^[ILf?[ IL ` +RZHYHU IRU D IL[HG QRQ]HUR YHFWRU ] If[]f DQG DUH FRQFDYH RYHU WKH VHWV 6[ ^rr ` DQG 6IL ^[_Q ` UHVSHFWLYHO\ 7KHVH UHVXOWV DUH SURYHG EHORZ 5HVXOW /HW ]4 EH D QRQ]HUR [ YHFWRU 7KHQ I![ ]f LV FRQFDYH RYHU 6[ 3URRI 2XU SURRI LV EDVHG RQ WKH ZHOO NQRZQ IDFW WKDW D IXQFWLRQ LV FRQFDYH LI LWV +HVVLDQ LV QHJDWLYH VHPLGHILQLWH >@ &RQVLGHU M![]f ZKHUH ]4 LV D IL[HG [ YHFWRU ,W IROORZV IURP HT f WKDW M![]f LV JLYHQ E\ W!^[ =Rf 3[fLHa^/=RfL 9L ORJ3[fL \L/]fL`  L DQG LWV ILUVW RUGHU SDUWLDO GHULYDWLYH ZLWK UHVSHFW WR ;M HTXDOV GMf[]f 3[fL f

PAGE 48

7KH +HVVLDQ RI ![]f ZKLFK ZH GHQRWH E\ +[f LV D [ PDWUL[ ZKRVH M tfWK HOHPHQW LV JLYHQ E\ $ GO![]f +^[fLN G[G[ \ MOLMN ‘ A B f f 6 Af &RQVLGHU WKH VFDODU P7+[fP ZKHUH P DQG D DUH DUELWUDU\ QRQ]HUR YHFWRUV M M O P +[fP ;,(( \L3LMPM3LNPN 3rf f f ‹ 3rf" f 6LQFH IRU DOO ] ] 3Pf DQG Hrf LW IROORZV WKDW P +[fP ZKLFK PHDQV WKDW WKH +HVVLDQ LV QHJDWLYH VHPLGHIL QLWH 7KXV I![]f LV FRQFDYH RYHU 6[ 5HVXOW /HW ] EH D QRQ]HUR [ YHFWRU 7KHQ Mf] ILf LV FRQFDYH RYHU WKH VHW Y 3URRI 6LPLODU WR WKH SURRI IRU 5HVXOW ZH QHHG WR SURYH WKDW WKH +HVVLDQ RI W!]ILf LV QHJDWLYH VHPLGHILQLWH IRU ]4 D QRQ]HUR IL[HG [ YHFWRU )URP HT f LW IROORZV WKDW A!]Lf LV JLYHQ E\ W!]' Qf e>3rRfH/;! 9L ORJ3]4fL \L/ILfL@ W L DQG LWV ILUVW RUGHU SDUWLDO GHULYDWLYH ZLWK UHVSHFW WR \M HTXDOV +]RQf ne3]RfLHO/A/LM ‘ W L GILM f

PAGE 49

7KH +HVVLDQ RI I!] ILf ZKLFK ZH GHQRWH E\ LV D [ PDWUL[ ZKRVH M NfWK HOHPHQW LV JLYHQ E\ GM!] ILf GQMGXN f
PAGE 50

&RQVLGHU WKH VFDODU P7+P ZKHUH P LV DQ DUELWUDU\ QRQ]HUR YHFWRU P7+P f§ f§PO f§ S[OH OWOP??SOH OIOP?P f [O )URP HT f LW FDQ EH VHHQ WKDW WKHUH LV QR ZD\ WR JXDUDQWHH WKDW PU+P LV OHVV WKDQ RU HTXDO WR ]HUR )RU S [ S \ PL DQG P WKH ORJ OLNHOLKRRG IXQFWLRQ LV QRW FRQFDYH LH P +P f 2SWLPL]DWLRQ 3URFHGXUH :H H[SORLW WKH FRQFDYLW\ UHVXOWV GLVFXVVHG LQ ODVW VHFWLRQ DQG SURSRVH WKH IROORZLQJ RSWLPL]DWLRQ SURFHGXUH WR REWDLQ WKH 0/ HVWLPDWHV LH VROYH 23ff 6WHS f *LYHQ SLNB DQG [Na GHWHUPLQH [N E\ PD[LPL]LQJ Af[SLNaf RYHU 6[ ,Q RWKHU ZRUGV 3,f [N DUJMQDA I!^[ Bf 6WHS f *LYHQ ]r DQG [N GHWHUPLQH cLN E\ PD[LPL]LQJ M![N SLf RYHU 6M 7KDW LV 3f +N UTDA W![NILf 6WHS f ,WHUDWH EHWZHHQ 6WHS DQG 6WHS 6WHS DQG 6WHS DUH UHIHUUHG WR DV WKH [VWHS DQG WKH SVWHS UHVSHFWLYHO\ 7KH YHFWRUV [N DQG SLN DUH WKH NWK HVWLPDWH RI [ DQG SL UHVSHFWLYHO\ ZKHUH DQ LWHUDWLRQ FRQVLVWV RI ERWK VWHSV DQG 8QIRUWXQDWHO\ FORVHG IRUP VROXWLRQV IRU [N DQG SLN DUH QRW DYDLODEOH ZKLFK PHDQV WKH\ PXVW EH GHWHUPLQHG XVLQJ LWHUDWLYH

PAGE 51

DOJRULWKPV ,Q WKH QH[W WZR VHFWLRQV ZH GHYHORS DOJRULWKPV IRU REWDLQLQJ [N DQG MVWHS *LYHQ SNaO ZH NQRZ WKDW Mf[SNaOf LV FRQFDYH RYHU 6[ IURP 5HVXOW 8QGHU WKH FRQGLWLRQV LQ SUREOHP 3,f [ LV D VROXWLRQ WR 3Lf LI DQG RQO\ LI WKH .XKQ7XFNHU FRQGLWLRQV >@ DUH VDWLVILHG ArLIFBf G[M nr ; LI ;M A M GI![WLN rf G[M LI ;M :H FDQ UHZULWH WKH DERYH WZR .XKQ7XFNHU FRQGLWLRQV DV &Of LI M OL ?U[K  L &f I LI M 3[fL L 7KH ILUVW FRQGLWLRQ &Of VXJJHVWV WKH IROORZLQJ IL[HG SRLQW LWHUDWLRQ IRU VROYLQJ 3,f rrfff§fn n -n  f§ QQ f L f (/ ZKHUH [IFfr1 LV WKH PWK HVWLPDWH RI WKH MWK YR[HO RI [IF DQG [fr [IFBfM IRU DOO M $QRWKHU SRVVLEOH LWHUDWLRQ LV 6n 3/SN Lf S N?P B N?P AL ]O H Ubf [Nfr1 [rffAA \7X M Of Ar 3mrfrf +RZHYHU LQ RXU VLPXODWLRQV WKLV LWHUDWLRQ fEOHZ XSf LH WKH LWHUDWHV EHFDPH XQERXQGHGf &RQVHTXHQWO\ ZH XVH WKH LWHUDWLRQ LQ HT f WR HVWLPDWH [N

PAGE 52

7KH ORJLF EHKLQG WKH IL[HG SRLQW LWHUDWLRQ LQ HTf LV GLVFXVVHG EHORZ 6XSn SRVH [Nfr1 FRQYHUJHV WR [NfMr A WKHQ IURP HTf ZH KDYH f ZKLFK LPSOLHV WKDW FRQGLWLRQ &Of LV VDWLVILHG ,Q RWKHU ZRUGV ZH NQRZ WKDW WKH IL[HG SRLQW LWHUDWLRQ LQ HTf ZLOO SURYLGH D VROXWLRQ VDWLVI\LQJ &Of LI LW FRQYHUJHV WR D SRVLWLYH YHFWRU $GGLWLRQDO PRWLYDWLRQ IRU WKH WHFKQLTXH XVHG WR GHULYH WKH DOJRULWKP LQ HT f LV WKDW WKH (00/ DOJRULWKP > @ DQG WKH ,65$ DOJRULWKP >@ FDQ EH GHULYHG XVLQJ LW $SSO\LQJ WKH DOWHUQDWLQJ SURMHFWLRQ PHWKRG >@ WR VROYH 3Lf OHDGV WR WKH VDPH LWHUDWLYH HTXDWLRQ DV HT f 7KLV LWHUDWLRQ LV JXDUDQWHHG WR FRQYHUJH E\ FRQYHUJHQFH UHVXOWV LQ %\UQHfV SDSHU >@ VWHS )URP 5HVXOW ZH NQRZ WKDW ![NQf LV FRQFDYH RYHU 6Lr IRU [N 7KXV ML LV D VROXWLRQ WR 3f LI DQG RQO\ LI WKH IROORZLQJ .XKQ7XFNHU FRQGLWLRQV KROG &f  \c/ LI Z M m 8VLQJ WKH VDPH DSSURDFK WKDW OHDGV WR WKH LWHUDWLRQ LQ HT f ZH REWDLQ WKH IROORZLQJ IL[HG SRLQW LWHUDWLRQ IRU VROYLQJ SUREOHP 3f LrfU PrfM f

PAGE 53

ZKHUH Afr1 LV WKH PWK HVWLPDWH RI WKH MWK YR[HO RI ILN DQG IFfr QN fM IRU DOO M $QRWKHU SRVVLEOH LWHUDWLYH HTXDWLRQ LV $P 0Lf) U ( L 9L/LM M f =/ H/:NfPff3[NfL/LM ,Q RXU VLPXODWLRQV WKLV LWHUDWLRQ fEOHZ XSf &RQVHTXHQWO\ ZH HVWLPDWH c[N XVLQJ WKH LWHUDWLRQ LQ HT f ,Q WKH QH[W VHFWLRQ ZH LQYHVWLJDWH WKH EHKDYLRU RI WKH ORJ OLNHOLKRRG IXQFWLRQ DV D IXQFWLRQ RI WKH LWHUDWHV 0RQRWRQLFDOOY ,QFUHDVLQJ 9DOXH RI WKH /RJ /LNHOLKRRG )XQFWLRQ LQ MVWHS )URP WKH ORJ OLNHOLKRRG IXQFWLRQ JLYHQ LQ HTf DQG WKH LWHUDWLYH HTXDWLRQ LQ HTf ZH FDQ SURYH WKDW WKH ORJ OLNHOLKRRG IXQFWLRQ LQFUHDVHV PRQRWRQLFDOO\ ZLWK VXFFHVVLYH LWHUDWLRQ LQ WKH D!VWHS 8QIRUWXQDWHO\ WKLV UHVXOW FDQQRW EH SURYHQ IRU WKH IMVWHS 5HVXOW /HW ]4 EH D IL[HG QRQ]HUR [ YHFWRU 7KHQ M![NfP ]f f[NfP]f 3URRI )URP HTf LW LV HDV\ WR VHH WKDW ![NU?]f![NU]f fSAQMR( r LrJ ?S-Q f ZKHUH D HaA]rf?

PAGE 54

+HQFH XVLQJ WKH LWHUDWLYH HTXDWLRQ LQ HTf LW IROORZV WKDW WKH VHFRQG WHUP RI WKH DERYH GHULYDWLRQ FDQ EH ZULWWHQ DV n 3[NUOfW rJ 3[Nfr1fc 3 LnUN?P ‘ e!‘n‘e LLEN 3 e\LRJ'A $ Wf§L n8?S^rNfPfL ZKHUH XN (/L /HW ID L ( L 3LM 8VLQJ WKH IDFW WKDW WKH ORJ IXQFWLRQ LV FRQFDYH ZH KDYH = ORV = 3Q8M ( \ 3L ORV W L M L } L M L f 0XOWLSO\LQJ WKH ULJKW VLGH RI HTf E\ WKH YDOXH RI ULJKW VLGH RI HTf GRHV QRW FKDQJH VLQFH D33 7KHQ LW IROORZV WKDW ORJZM } L M L W O   8LrLM r \M3MM LRJAf M O ( L DL 3L@ N ORJM O Wn O

PAGE 55

! [D$fAOf M L  L (rI( \3LM !e rfmrrff 6L n 6 Srffff 6 6 ;!!L3r7fL 6LQFH (nL 6L OrJ e L $MAfM (6L\L(6LIWMORJX" ZH KDYH 7KHQ ZH JHW (/ 6L (/LOfL3r7fL DV3rrffL Dc3rrfmfL@ (/L 6L (/O 2L3rffPfL )URP HTf ZH NQRZ WKDW f IR O3,Urf B \ \c3[LfPfc B\  n nf IH 3}rfffL 6 )URP HTf ZH REWDLQ VL‹rLSLQfUfL R m L  L 7KXV ZH JHW WKH GHVLUHG UHVXOW f f rrfffrRfr7rf! N?P ,Q WKH QH[W VHFWLRQ DQ LQWHUHVWLQJ SURSHUW\ RI WKH 0/ HVWLPDWRU LV SUHVHQWHG

PAGE 56

3URSHUW\ RI WKH 0/ (VWLPDWRU 5HVXOW 6XSSRVH WKHUH H[LVWV D VROXWLRQ rrrf WR 23f 7KHQ \W3W 3Yf (O \M3MM r O r1 (/ =/ O3[fL&/U!f/L (/ YcN f  f mM 3URRI &RQVLGHU WKH ORJ OLNHOLKRRG IXQFWLRQ e>3rfL&BnenrLfr 9L ORJ3[fc \;Lf@ L L 6XSSRVH WKDW WKHUH H[LVWV DQ RSWLPDO VROXWLRQ [rILrf DQG OHW W EH D QRQQHJDWLYH VFDODU DQG r DQG LL EH QRQQHJDWLYH YHFWRUV 7KHQ ![r W[ ILr WSLf I![rILrf )URP WKH DERYH UHODWLRQVKLS ZH JHW (/IFLRJLAA0VAf f§ >3[rf 3rf@H>,rfW,f 3[rfLH^/:n` 6LQFH 3[rf^H J IRU DMM A ZH KDYH (/L ID LrJ>Sr L W\L/Qf^ >3DMrf W^3[fL@HA/AfLLA/A` f 'LYLGLQJ HTf E\ W ZH JHW f

PAGE 57

1RZ OHWWLQJ W f§ DQG XVLQJ WKH /f+RSWLDOfV UXOH ZH REWDLQ  ?A[ fW f 6LQFH HTf PXVW KROG IRU DQ\ Q DQG [ ZH FRQVLGHU WKH FDVH ZKHUH [ HM HM LV D [ YHFWRU RI ]HURV H[FHSW IRU WKH WK HOHPHQW ZKLFK LV D RQHf DQG c[ )RU WKLV FDVH HT f EHFRPHV 7: f  ?A; fL 6LPLODUO\ IRU [ DQG IL HM HT f \LHOGV >cIFOWI 3[AHAA/LMO M f L L 1RWH HT f DQG f DUH RQO\ QHFHVVDU\ FRQGLWLRQV IRU pr[rf WR EH D PD[LPXP RI Mf[ ILf RYHU WKH IHDVLEOH UHJLRQ 6LQFH [r Qr LW IROORZV IURP HT f DQG HT f WKDW \n A 3Df =/ HA:3MM =/L 9L/LM   f f ,Q WKH QH[W FKDSWHU ZH LQWURGXFH D QHZ OHDVW VTXDUHV UHSURMHFWLRQ DWWHQXDWLRQ FRUUHFWLRQ PHWKRG

PAGE 58

&+$37(5 /($67 648$5(6 5(352-(&7,21 0(7+2' ,Q WKLV FKDSWHU ZH LQWURGXFH D UHSURMHFWLRQ PHWKRG ZKHUH WKH DWWHQXDWLRQ GHQVLW\ LV HVWLPDWHG XVLQJ D OHDVW VTXDUHV /6f DOJRULWKP 7KH DOJRULWKP LV DQ H[n WHQVLRQ RI RQH SURSRVHG E\ $QGHUVRQ HW DO >@ IRU ;UD\ FRPSXWHG WRPRJUDSK\ /6 $OJRULWKP IRU (VWLPDWLQJ WKH $WWHQXDWLRQ 'HQVLW\ $V GLVFXVVHG LQ 6HFWLRQ DQG OHW  GHQRWH WKH QXPEHU RI SKRWRQV GHn WHFWHG LQ WXEH L GXULQJ WKH EODQN VFDQ ,W LV DVVXPHG WKDW t‘ LV DQ REVHUYDWLRQ RI D UDQGRP YDULDEOH /" ZKHUH %^ LV 3RLVVRQ ZLWK PHDQ 8L ,W LV DOVR DVVXPHG WKDW WKH QXPEHU RI SKRWRQV GHWHFWHG LQ WXEH L GXULQJ WKH WUDQVPLVVLRQ VFDQ U LV DQ REVHUYDWLRQ RI D UDQGRP YDULDEOH -" ZKHUH 5^ LV 3RLVVRQ ZLWK PHDQ eDWAf UHFDOO Dc *LYHQ WKH REVHUYHG EODQN VFDQ GDWD T f f f DQG WKH REVHUYHG WUDQVPLVVLRQ GDWD UT U U WKH JRDO LV WR HVWLPDWH WKH DWWHQXDWLRQ GHQVLW\ I[ ,W LV VWUDLJKWIRUZDUG WR VKRZ WKDW  DQG U DUH WKH 0/ HVWLPDWHV RI 8L DQG 8FTAf UHVSHFWLYHO\ &RQVHTXHQWO\  m 8^ DQG U m 7KHVH DSSUR[Ln PDWLRQV OHDG WR WKH IROORZLQJ RQH D a 7DNLQJ WKH ORJDULWKP RQ ERWK VLGHV RI WKLV UHODWLRQVKLS DQG PXOWLSO\LQJ WKH UHVXOW E\ QHJDWLYH RQH ZH JHW /ILfL m G ZKHUH G f§ ORJAAff ([SORLWLQJ WKLV REVHUYDWLRQ ZH SURSRVH WR HVWLPDWH WKH DWWHQXDWLRQ GHQVLW\ E\ PLQLPL]LQJ WKH QRUP GLVWDQFH EHWZHHQ G DQG /ILf 3f X cOUA,E S0 f

PAGE 59

ZKHUH SI[f =L L^GLf§^/QfLf 6 WKH/6 HVWLPDWH RI D DQG G >F"[ F f f f G@ ,W LV VWUDLJKWIRUZDUG WR VKRZ WKDW MDf LV FRQYH[ RYHU 6A $ YHFWRU LW LV D VROXWLRQ WR 3f LI DQG RQO\ LI LW VDWLVILHV WKH IROORZLQJ .XKQ 7XFNHU FRQGLWLRQV &f -/n$GL a /$fLf LI 3M L L &f e/ILGcrf‘f! LI A  L L $V EHIRUH VHH &KDSWHU f WKH ILUVW FRQGLWLRQ &f PRWLYDWHV WKH IROORZLQJ IL[HG SRLQW LWHUDWLRQ L ] L GL/LM (/ M f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f /HW IL EH WKH UHVXOWLQJ HVWLPDWH 7KHQ WKH VXUYLYDO SUREDELOLWLHV DUH HVWLPDWHG E\ UHSURMHFWLQJ IL £L Ha =8 L O f 1H[W WKH HVWLPDWHG VXUYLYDO SUREDELOLWLHV DUH XVHG WR FUHDWH D PRGLILHG SUREDELOLW\ PDWUL[ 3 ZKHUH 3M D3M 7KH HPLVVLRQ LQWHQVLW\ LV WKHQ HVWLPDWHG E\ XVLQJ

PAGE 60

WKH (00/ DOJRULWKP ZLWK WKH PRGLILHG SUREDELOLW\ PDWUL[ \} 9L 3L /AL? BIFO ;M ; 3[Nf S Q L UrM ,W LV VWUDLJKWIRUZDUG WR VKRZ WKDW WKLV LWHUDWLRQ LV HTXLYDOHQW WR 9L3LM 3[b ; r [?‘‘ (O 9L 3LM r a M f§ f f =/ L3LL ,Q WKH QH[W FKDSWHU VLPXODWLRQV DUH XVHG WR FRPSDUH WKH SHUIRUPDQFH RI WKH 0/ PHWKRG /6 UHSURMHFWLRQ PHWKRG DQG (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD

PAGE 61

&+$37(5 6,08/$7,21 8VLQJ V\QWKHWLF GDWD DQG UHDO GDWD ZH DVVHVV WKH SURSRVHG PHWKRGV DQG FRPn SDUH WKHP WR WKH (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD 'U -RKQ 9RWDZ $Vn VRFLDWH 3URIHVVRU RI 5DGLRORJ\ DW WKH (PRU\ 8QLYHUVLW\ +RVSLWDO SURYLGHG WKH UHDO GDWD 7KLV GDWD ZDV FROOHFWHG RQ WKH 6LHPHQV&7, (&$7 (;$&7 3(7 VFDQQHU 7KH SUREDELOLWLHV 3M DQG OLQH OHQJWK 7M ZHUH GHWHUPLQHG IURP WKH GLDPHWHU RI WKH GHWHFWRU ULQJ QXPEHU RI GHWHFWRUV WKH VL]H RI 52, WKH 52, LV DOZD\V D VTXDUH FHQWHUHG LQ WKH PLGGOH RI WKH GHWHFWRU ULQJf DQG WKH QXPEHU RI YR[HOV 6SHFLDOO\ WKH SUREDELOLWLHV DUH FDOFXODWHG XVLQJ WKH DQJOHRIYLHZ PHWKRG >@ ,W VKRXOG EH QRWHG WKDW WKH SUREDELOLWLHV DQG OLQH OHQJWKV DUH FDOFXODWHG XQn GHU WKH DVVXPSWLRQ WKDW WKH GHWHFWRUV DUH FRQWLJXRXV 7KH QXPEHU RI GHWHFWRUV IRU WKH VLPXODWHG VFDQQHU DQG WKH 6LHPHQV&7, (&$7 (;$&7 VFDQQHU DUH DQG UHVSHFWLYHO\ 7KH FRPSXWDWLRQDO H[SHQVH RI DQ LWHUDWLRQ IRU WKH 0/ PHWKRG /6 UHSURMHFWLRQ PHWKRG DQG WKH (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD DUH HVVHQWLDOO\ WKH VDPH &RQVHTXHQWO\ WKH FRPSXWDWLRQDO FRPSOH[LW\ RI WKH DOn JRULWKPV FDQ EH FRPSDUHG E\ FRQVLGHULQJ WKH WRWDO QXPEHU RI LWHUDWLRQV UHTXLUHG E\ HDFK RQH

PAGE 62

6\QWKHWLF 'DWD )RU WKH V\QWKHWLF GDWD ZH XVH D FDUGLDF HPLVVLRQ SKDQWRP VHH )LJ Dff FKHVW HPLVVLRQ SKDQWRP VHH )LJ Dff FDUGLDF DWWHQXDWLRQ SKDQWRP VHH )LJ Dff DQG FKHVW DWWHQXDWLRQ SKDQWRP VHH )LJ Dff $OO RI WKH SKDQWRPV DUH RI VL]H [ DQG WKH YR[HO VL]H IRU WKH FDUGLDF SKDQWRPV DQG FKHVW SKDQWRPV DUH FP DQG FP UHVSHFWLYHO\ 7KH OXQJV VSLQH DQG VRIW WLVVXH UHJLRQ LQ WKH FDUGLDF DWWHQXDWLRQ SKDQWRP KDYH GHQVLWLHV FP FP DQG FP UHVSHFWLYHO\ 7KH ORZ GHQVLW\ UHJLRQV OLJKW FRORU UHJLRQf DQG WKH KLJK GHQVLW\ UHJLRQV GDUN FRORU UHJLRQf LQ WKH FKHVW DWWHQXDWLRQ SKDQWRP KDYH GHQVLWLHV FP DQG FPB UHVSHFWLYHO\ 7KH GLDPHWHU RI WKH GHWHFWRU ULQJ IRU WKH FDUGLDF SKDQWRPV DQG FKHVW SKDQWRPV LV FP DQG FP UHVSHFWLYHO\ 7KH 52, IRU WKH FDUGLDF SKDQWRPV DQG FKHVW SKDQWRPV LV FP DQG FP UHVSHFWLYHO\ 7KH QXPEHU RI WXEHV LV f§ :H JHQHUDWH WKH HPLVVLRQ VFDQ WUDQVPLVVLRQ VFDQ DQG EODQN VFDQ GDWD LQ WKH IROORZLQJ PDQQHU 7KH REVHUYHG FRLQFLGHQFH IRU WKH 7K WXEH LV WKH REVHUYDWLRQ RI D 3RLVVRQ UDQGRP YDULDEOH ZLWK PHDQ 3[fDc UHFDOO D 7KH HPLVVLRQ SKDQWRPV DQG DWWHQXDWLRQ SKDQWRPV VSHFLI\ [ DQG UHVSHFWLYHO\ 7KH REVHUYHG FRXQW IRU WKH EODQN VFDQ IRU WKH "WK WXEH  DQG WKH WUDQVPLVVLRQ VFDQ IRU WKH WK WXEH U DUH 3RLVVRQ UDQGRP YDULDEOHV ZLWK PHDQV 8^ DQG WD,Wf UHVSHFWLYHO\ :H VSHFLI\ WKH WRWDO FRXQW RI WKH WUDQVPLVVLRQ VFDQ WR EH 5 DQG XVH

PAGE 63

0/ 0HWKRG ,PSOHPHQWDWLRQ RI $OJRULWKPV 7KH WHUP ORFDO LWHUDWLRQ UHIHUV WR WKH LWHUDWLRQV XVHG WR GHWHUPLQH [N DQG >LK LQ 6WHS [VWHSf DQG 6WHS VWHSf UHVSHFWLYHO\ VHH 3,f DQG 3f LQ 6HFWLRQ f 7KH WHUP JOREDO LWHUDWLRQ UHIHUV WR D VLQJOH H[HFXWLRQ RI ERWK WKH [VWHS DQG WKH VWHS )RU WKH VLPXODWLRQV XVLQJ V\QWKHWLF GDWD ZH UXQ ORFDO LWHUDWLRQV IRU WKH [VWHS ORFDO LWHUDWLRQV IRU WKH VWHS DQG JOREDO LWHUDWLRQV 7KXV D WRWDO RI LWHUDWLRQV DQG LWHUDWLRQV DUH XVHG WR REWDLQ HPLVVLRQ LQWHQVLW\ LH [f DQG DWWHQXDWLRQ GHQVLW\ LH ILf UHVSHFWLYHO\ 7KH 0/ PHWKRG LV LQLWLDOL]HG XVLQJ DfM -L L DQG $rf" H[FHSW WKDW LV UHSODFHG ZLWK Dp MMrAf $V VWDWH LQ 6HFWLRQ DQG 6HFWLRQ [Nfr [NafM DQG WLNf ]IHfM (00/ $OJRULWKP ZLWK &RUUHFWHG 'DWD )RU WKH (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD WKH LWHUDWLYH HTXDWLRQ LV JLYHQ EHORZ $ Q N B N 9LALM NO N ; ,L/ f 3[NfL f ;r LWHUDWLRQV RI WKH (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD f ZKHUH \ br ,W LV LQLWLDOL]HG DV IROORZV [r \f IRU DOO M :H UXQ 9 2WL /6 5HSURMHFWLRQ 0HWKRG 7R LQLWLDOL]H WKH LWHUDWLRQ LQ HT f ZH XVH D XQLIRUP LQLWLDO HVWLPDWH r IRU DOO M :H UXQ LWHUDWLRQV WR JHW WKH /6 HVWLPDWH RI ] IL 7KHQ ZH UXQ LWHUDWLRQV RI WKH DOJRULWKP VKRZQ LQ HT f 7KXV WKH /6 UHSURMHFWLRQ PHWKRG UHTXLUHV M2 LWHUDWLRQV WR JHW DQ HPLVVLRQ LPDJH

PAGE 64

7KH WRWDO HPLVVLRQ FRXQW W\SLFDO WUDQVPLVVLRQ VFDQ FRXQW DQG VKRUW WUDQVPLVn VLRQ VFDQ FRXQW DUH PLOOLRQ PLOOLRQ DQG PLOOLRQ UHVSHFWLYHO\ 7KH EODQN VFDQGXUDWLRQ WKH W\SLFDO WUDQVPLVVLRQ VFDQ GXUDWLRQ DQG WKH VKRUW WUDQVn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f )LJ Ff DQG )LJ Gf UHVSHFWLYHO\ ,W FDQ EH VHHQ WKDW WKH HPLVVLRQ LPDJHV LPSURYH ZLWK LQFUHDVLQJ LWHUDWLRQ +RZHYHU DIWHU JOREDO LWHUDWLRQV WKH HPLVVLRQ LPDJHV EHFRPH QRLV\ DQG OHVV DFFXUDWH ZKLFK LQGLFDWHV WKDW D VWRSSLQJ UXOH QHHGV WR EH IRXQG 7KLV SUREOHP LV D FRPPRQ RQH IRU VWDWLVWLFDOO\ EDVHG DOJRULWKPV 7KH UHFRQVWUXFWLRQV RI WKH FDUGLDF DWWHQXDWLRQ SKDQWRP XVLQJ WKH 0/ PHWKRG DIWHU WKH VW JOREDO LWHUDWLRQ QG JOREDO LWHUDWLRQ DQG UG JOREDO LWHUDWLRQ DUH VKRZQ LQ )LJ Ef )LJ Ff DQG )LJ Gf UHVSHFWLYHO\ ,Q WKH DWWHQXDWLRQ LPDJHV WKH OXQJV DUH QRW YLVLEOH DQG RQO\ WKH FRQWRXU DQG WKH VSLQH FDQ EH VHHQ $OVR WKHUH DUH DUWLIDFWV RXWVLGH WKH 52,

PAGE 65

)LJXUH 5HFRQVWUXFWLRQ RI FDUGLDF HPLVVLRQ SKDQWRP XVLQJ QRQVPRRWKHG W\Sn LFDO WUDQVPLVVLRQ VFDQ GDWD Df SKDQWRP Ef VW JOREDO LWHUDWLRQ Ff QG JOREDO LWHUDWLRQ Gf UG JOREDO LWHUDWLRQ

PAGE 66

)LJXUH 5HFRQVWUXFWLRQ RI FDUGLDF DWWHQXDWLRQ SKDQWRP XVLQJ QRQVPRRWKHG W\SLFDO WUDQVPLVVLRQ VFDQ GDWD Df DWWHQXDWLRQ SKDQWRP Ef VW JOREDO LWHUDWLRQ Ff QG JOREDO LWHUDWLRQ Gf UG JOREDO LWHUDWLRQ

PAGE 67

)RU WKH FDUGLDF HPLVVLRQ SKDQWRP WKH UHVXOWV RI WKH (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD WKH /6 UHSURMHFWLRQ PHWKRG DQG WKH 0/ PHWKRG DUH VKRZQ LQ )LJ Df )LJ Ef )LJ Ff DQG )LJ Gf UHVSHFWLYHO\ )URP WKLV ILJXUH LW FDQ EH VHHQ WKDW WKH 0/ PHWKRG \LHOGV DQ LPDJH ZLWK EHWWHU FRQn WUDVW DQG UHVROXWLRQ WKDQ WKH (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD DQG WKH /6 UHSURMHFWLRQ PHWKRG QRWH WKH OXQJV DQG WKH KHDUW DW WKH WRS FHQWHU RI WKH FDUGLDF SKDQWRPf ,W DOVR FDQ EH VHHQ WKDW WKH (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD GRHV QRW VXFFHVVIXOO\ UHVROYH HLWKHU WKH OXQJV RU WKH KHDUW ,Q WKH HPLVVLRQ LPDJH RI WKH /6 UHSURMHFWLRQ PHWKRG WKH VSLQH FDQ EH VHHQ FOHDUO\ +RZHYHU WKH /6 UHSURMHFWLRQ PHWKRG DOVR SURYLGHV LQFRUUHFW LQIRUPDWLRQ DV VHHQ E\ WKH IDFW WKDW WKH OXQJV KDYH D KLJKHU DFWLYLW\ FRQFHQWUDWLRQ WKDQ WKH VRIW WLVVXHV DURXQG WKHP VHH WKH SKDQWRP LQ )LJ Dff ,Q )LJ ZH VKRZ WKH OLQH SORWV RI WKH WK URZ RI WKH HPLVVLRQ LPDJHV LQ )LJ :H REVHUYH WKDW ERWK WKH /6 UHSURMHFWLRQ PHWKRG DQG WKH (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD DUH PRUH ELDVHG WKDQ WKH 0/ PHWKRG QRWH WKH SORW EHWZHHQ DQG ZKLFK LV LQVLGH WKH ULJKW OXQJ DUHDf )URP WKLV OLQH SORW ZH ILQG WKDW WKH 0/ PHWKRG QRW RQO\ SURGXFHV HPLVVLRQ LPDJHV ZLWK EHWWHU FRQWUDVW EXW DOVR JLYHV D PRUH DFFXUDWH HVWLPDWH RI WKH HPLVVLRQ LQWHQVLW\ WKDQ RWKHU PHWKRGV 7KH (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD GRHV QRW UHFRQVWUXFW DWWHQXDWLRQ LPDJHV VR ZH RQO\ VKRZ WKH DWWHQXDWLRQ LPDJHV RI WKH 0/ PHWKRG DQG WKH /6 UHSURMHFWLRQ PHWKRG LQ )LJ Ef DQG )LJ Ff UHVSHFWLYHO\ 7KH LGHDO DWWHQXDWLRQ SKDQWRP LV VKRZQ LQ )LJ Df :H FDQ VHH WKDW WKH DWWHQXDWLRQ GHQVLW\ UHFRQVWUXFWLRQ RI WKH 0/ PHWKRG LQ )LJ Ef KDV D ORW RI DUWLIDFWV DQG

PAGE 68

)LJXUH 5HFRQVWUXFWLRQ RI FDUGLDF HPLVVLRQ SKDQWRP XVLQJ QRQVPRRWKHG W\Sn LFDO WUDQVPLVVLRQ VFDQ GDWD Df HPLVVLRQ SKDQWRP Ef (00/ DOJRULWKP ZLWK FRUn UHFWHG GDWD Ff /6 UHSURMHFWLRQ PHWKRG Gf 0/ PHWKRG

PAGE 69

)LJXUH /LQH SORW RI WKH WO URZ RI WKH UHFRQVWUXFWLRQ RI FDUGLDF HPLVVLRQ SKDQWRP XVLQJ QRQVPRRWKHG W\SLFDO WUDQVPLVVLRQ VFDQ GDWD [ M OL LnOO n ? L L L }  L L U 6ROLG 3KDQWRP 'DVK'DVK 0/ 0HWKRG 'DVK'RW /6 5HSURMHFWLRQ 0HWKRG 'RW (06WDQGDUG &RUUHFWLRQ 0HWKRG f W f f ? L W ‘ L L ‘ A 6 f n‘ L? f nL 9U I f nW r ff ? f f L B > n r L ƒ W O ?? f fr $ W ? W L ? ‘ L  W n LI ?L L L L L L L ?-L L ?L

PAGE 70

GRHV QRW UHVROYH WKH OXQJV VSLQH DQG VRIW WLVVXH 7KH /6 UHSURMHFWLRQ PHWKRG UHVROYHV OXQJV DQG VRIW WLVVXHV EXW DOVR KDV DUWLIDFWV :H DOVR VKRZ WKH HPLVVLRQ LPDJHV RI WKH 0/ PHWKRG LQLWLDOL]HG ZLWK D /6 HVWLPDWH RI WKH DWWHQXDWLRQ GHQVLW\ 6LQFH WKH VWHS LV VXSSRVHG WR GHWHUPLQH D EHWWHU HVWLPDWH RI ZH ZDQW WR FKHFN LI D EHWWHU LQLWLDO Q ZLOO FRUUHFW WKH DUWLIDFW SUREOHP RI WKH DWWHQXDWLRQ LPDJHV DQG SURYLGH EHWWHU HPLVVLRQ LPDJHV IRU WKH 0/ PHWKRG ,Q )LJ Ef ZH FDQ REVHUYH WKDW VSLQH LV FOHDUHU EXW WKH OXQJ DUHD LV LQFRUUHFW +RZHYHU WKH DWWHQXDWLRQ LPDJH LQ )LJ Gf LV EHWWHU WKDQ )LJ Ef UHVROYH OXQJV VRIW WLVVXH EXW QR VSLQHf 7KH UHVXOWV RI WKH (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD WKH /6 UHSURMHFWLRQ PHWKRG DQG WKH 0/ PHWKRG XVLQJ WKH FKHVW SKDQWRP DUH VKRZQ LQ )LJ Ef Ff DQG Gf UHVSHFWLYHO\ )URP WKLV ILJXUH ZH FDQ REVHUYH WKDW WKH 0/ PHWKRG DQG WKH /6 UHSURMHFWLRQ PHWKRG SURGXFH FRPSDUDEOH HPLVVLRQ LPDJHV %RWK LPDJHV KDYH EHWWHU FRQWUDVW DQG UHVROXWLRQ WKDQ WKH (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD 1RWH WKH HPLVVLRQ LPDJH SURGXFHG E\ WKH (0 DOJRULWKP ZLWK FRUUHFWHG GDWD LV YHU\ QRLV\ :H SUHVHQW WKH DWWHQXDWLRQ LPDJHV RI WKH 0/ PHWKRG DQG WKH /6 UHSURMHFWLRQ PHWKRG IRU WKH FKHVW SKDQWRP LQ )LJ Ef DQG )LJ Ff UHVSHFWLYHO\ 7KH LGHDO DWWHQXDWLRQ SKDQWRP LV VKRZQ LQ )LJ Df )RU WKH FKHVW SKDQWRP WKH 0/ PHWKRG SURYLGHV EHWWHU FRQWUDVW DQG PRUH GHWDLOV ELJ DQG VPDOO VSRWV LQ WKH FHQWHUf WKDQ WKH /6 UHSURMHFWLRQ PHWKRG 7KH /6 UHSURMHFWLRQ PHWKRG IDLOV WR UHVROYH DQ\ LQIRUPDWLRQ

PAGE 71

Ff )LJXUH 5HFRQVWUXFWLRQ RI FDUGLDF DWWHQXDWLRQ SKDQWRP XVLQJ QRQVPRRWKHG W\SLFDO WUDQVPLVVLRQ VFDQ GDWD Df DWWHQXDWLRQ SKDQWRP Ef 0/ PHWKRG Ff /6 UHSURMHFWLRQ PHWKRG

PAGE 72

)LJXUH 0/ PHWKRG Df HPLVVLRQ SKDQWRP Ef HPLVVLRQ LPDJH Ff DWWHQXDWLRQ SKDQWRP Gf DWWHQXDWLRQ LPDJH

PAGE 73

)LJXUH 5HFRQVWUXFWLRQ RI FKHVW HPLVVLRQ SKDQWRP XVLQJ QRQVPRRWKHG W\SLFDO WUDQVPLVVLRQ VFDQ GDWD Df HPLVVLRQ SKDQWRP Ef (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD Ff /6 UHSURMHFWLRQ PHWKRG Gf 0/ PHWKRG

PAGE 74

r r r &f )LJXUH 5HFRQVWUXFWLRQ RI FKHVW DWWHQXDWLRQ SKDQWRP XVLQJ QRQVPRRWKHG W\SLFDO WUDQVPLVVLRQ VFDQ GDWD Df DWWHQXDWLRQ SKDQWRP Ef 0/ PHWKRG Ff /6 UHSURMHFWLRQ PHWKRG

PAGE 75

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f Ff DQG )LJ Gf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f DQG )LJ Ff UHVSHFWLYHO\

PAGE 76

)LJXUH 5HFRQVWUXFWLRQ RI FDUGLDF HPLVVLRQ SKDQWRP XVLQJ VPRRWKHG W\SLFDO WUDQVPLVVLRQ VFDQ GDWD Df HPLVVLRQ SKDQWRP Ef (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD Ff /6 UHSURMHFWLRQ PHWKRG Gf 0/ PHWKRG

PAGE 77

[ 6ROLG 3KDQWRP 'DVK'DVK 0/ 0HWKRG 'DVK'RW /6 5HSURMHFWLRQ 0HWKRG 'RW (06WDQGDUG &RUUHFWLRQ 0HWKRG 7 / W $r W ‘n \ ‘L f f Y ? A ‘‘‘ A ? f f ‘ !‘! U L f O f 1 s===AA 9V P f§ L L M W : L B M ?Ya f O?UUW )LJXUH /LQH SORW RI WKH WO URZ RI UHFRQVWUXFWLRQ RI FDUGLDF HPLVVLRQ SKDQ WRP XVLQJ WKH VPRRWKHG WUDQVPLVVLRQ VFDQ GDWD

PAGE 78

7KH 0/ PHWKRG VWLOO SURGXFHV DQ DWWHQXDWLRQ LPDJH ZLWK KHDY\ DUWLIDFWV ZKLOH WKH /6 PHWKRG SURYLGHV DQ LPDJH ZLWK EHWWHU FRQWUDVW DQG TXDOLW\ &RPSDUHG WR WKH DWWHQXDWLRQ LPDJHV XVLQJ QRQVPRRWKHG WUDQVPLVVLRQ VFDQ GDWD LQ )LJ ERWK DWWHQXDWLRQ LPDJHV VKRZ PRUH IHDWXUHV 7KH HPLVVLRQ DQG DWWHQXDWLRQ LPDJH RI WKH 0/ PHWKRG XVLQJ DQ LQLWLDO HVWLn PDWH IRU Q ZDV REWDLQHG IURP WKH /6 UHSURMHFWLRQ PHWKRG DUH VKRZQ LQ )LJ Ef )LJ Gf UHVSHFWLYHO\ 7KH 0/ PHWKRG SURGXFHV HPLVVLRQ LPDJH ZLWK EHWWHU FRQWUDVW LQ WKH KHDUW DUHD LQ WKDQ )LJ Ef 7KH DWWHQXDWLRQ LPDJH SURYLGHV PRUH LQIRUPDWLRQ WKDQ ERWK DWWHQXDWLRQ LPDJHV LQ )LJ QRWH WKH EODFN VSLQH DW WKH FHQWHU EHORZf 7KH HPLVVLRQ LPDJHV RI WKH (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD WKH /6 UHSURMHFWLRQ PHWKRG DQG WKH 0/ PHWKRG RI WKH FKHVW SKDQWRP XVLQJ VPRRWKHG W\SLFDO WUDQVPLVVLRQ VFDQ GDWD DUH VKRZQ LQ )LJ Ef Ff DQG Gf UHVSHFWLYHO\ :H FDQ VHH WKDW WKH /6 UHSURMHFWLRQ PHWKRG DQG WKH 0/ PHWKRG SURGXFH FRPSDUDEOH LPDJHV %RWK PHWKRGV SURYLGH EHWWHU FRQWUDVW DQG WKH LPDJH TXDOLW\ WKDQ WKH (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD 7KH DWWHQXDWLRQ LPDJHV RI WKH 0/ PHWKRG DQG WKH /6 UHSURMHFWLRQ PHWKRG XVLQJ WKH W\SLFDO VPRRWKHG WUDQVPLVVLRQ GDWD IRU FKHVW DWWHQXDWLRQ SKDQWRP DUH VKRZQ LQ )LJ Ef DQG )LJ Ff UHVSHFWLYHO\ 7KH 0/ PHWKRG UHVROYHV OXQJV D KHDUW DQG VRIW WLVVXH ZKLOH WKH /6 PHWKRG IDLOV WR SURYLGH DQ\ LQIRUPDWLRQ 7KH DWWHQXDWLRQ LPDJH RI WKH 0/ PHWKRG LV VPRRWKHU WKDQ )LJ Ef EXW WKH FRQWUDVW DQG TXDOLW\ GR QRW LPSURYH

PAGE 79

)LJXUH 5HFRQVWUXFWLRQ RI FDUGLDF DWWHQXDWLRQ SKDQWRP XVLQJ WKH VPRRWKHG W\SLFDO WUDQVPLVVLRQ VFDQ GDWD Df DWWHQXDWLRQ SKDQWRP Ef 0/ PHWKRG Ff /6 UHSURMHFWLRQ PHWKRG

PAGE 80

)LJXUH 0/ PHWKRG Df HPLVVLRQ SKDQWRP Ef HPLVVLRQ LPDJH Ff DWWHQXDWLRQ SKDQWRP Gf DWWHQXDWLRQ LPDJH

PAGE 81

)LJXUH (PLVVLRQ LPDJH RI WKH FKHVW SKDQWRP XVLQJ VPRRWKHG W\SLFDO WUDQVn PLVVLRQ VFDQ GDWD Df HPLVVLRQ SKDQWRP Ef (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD Ff /6 UHSURMHFWLRQ PHWKRG Gf 0/ PHWKRG

PAGE 82

)LJXUH 5HFRQVWUXFWLRQ RI FKHVW DWWHQXDWLRQ SKDQWRP XVLQJ VPRRWKHG W\SLFDO WUDQVPLVVLRQ VFDQ GDWD Df DWWHQXDWLRQ SKDQWRP Ef 0/ PHWKRG Ff /6 UHSURMHFn WLRQ PHWKRG

PAGE 83

,Q RUGHU WR FRPSDUH WKH PHWKRGV QXPHULFDOO\ ZH FRPSXWH WKHLU SHDNVLJQDO WRQRLVHUDWLR 3615f >@ IRU WKH GLIIHUHQW LPDJHV 3615 LV GHILQHG E\ [ 3615G%f ORJ r f PVH ZKHUH [SHDN LV WKH SHDN YDOXH RI WKH SKDQWRP PVH MeM L[ f§ ef ;M LV WKH MWK YR[HO LQ SKDQWRP DQG [M LV DQ HVWLPDWH RI ;M :H FRPSDUH WKH 3615 RI WKH PHWKRGV XVLQJ WKH W\SLFDO WUDQVPLVVLRQ VFDQ GDWD LQ 7DEOH :H FDQ G% FDUGLDF SKDQWRP FKHVW SKDQWRP $OJRULWKP 6WDQGDUG 0/ /65HS 6WDQGDUG 0/ /65HS 1RQVPRRWK 6PRRWK 7DEOH 3615 G%f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f Ff DQG Gf

PAGE 84

UHVSHFWLYHO\ )URP WKLV ILJXUH ZH NQRZ WKDW WKH 0/ PHWKRG UHVROYHV WKH KHDUW DQG OXQJV EHWWHU WKDQ WKH WKH (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD DQG WKH /6 UHSURMHFWLRQ PHWKRG 7KH HPLVVLRQ LPDJH RI WKH /6 UHSURMHFWLRQ PHWKRG SURGXFHV LQFRUUHFW LQIRUPDWLRQ LQ WKH OXQJV UHJLRQ :H DOVR FRPSDUH WKH DWWHQXDWLRQ LPDJHV :H VKRZ WKH UHVXOW RI WKH 0/ PHWKRG DQG WKH /6 UHSURMHFWLRQ PHWKRG LQ )LJ Ef DQG )LJ Ff UHVSHFWLYHO\ :H FDQ VHH WKDW WKH DWWHQXDWLRQ LPDJH RI 0/ PHWKRG LQ )LJ Ef KDV D ORW RI DUWLIDFWV DQG RQO\ UHVROYHV VSLQH EODFN VSRW DW WKH FHQWHU EHORZf 7KH DWWHQXDWLRQ LPDJH RI WKH /6 UHSURMHFWLRQ PHWKRG LQ )LJ Ff VKRZV WKH EDVLF FRQWRXUV EHWWHU WKDQ WKH 0/ PHWKRG EXW DOVR VXIIHUV IURP WKH DUWLIDFWV %RWK PHWKRGV SURYLGH ZRUVH DWWHQXDWLRQ LPDJHV ZKHQ XVLQJ VKRUW WUDQVPLVVLRQ VFDQ GDWD 7KH HPLVVLRQ DQG DWWHQXDWLRQ LPDJH RI WKH 0/ PHWKRG LQLWLDOL]HG XVLQJ WKH /6 HVWLPDWH RI Q DUH VKRZQ LQ )LJ Ef )LJ Gf UHVSHFWLYHO\ 7KH 0/ PHWKRG UHVROYHV WKH KHDUW DUHD EHWWHU WKDQ )LJ Gf EXW SURGXFHV WKH ZURQJ LQIRUPDWLRQ RI WKH OXQJV 7KH DWWHQXDWLRQ LPDJH SURYLGHV PRUH LQIRUPDWLRQ DQG OHVV DUWLIDFWV WKDQ )LJ Ff 7KH UHVXOWV RI WKH (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD WKH /6 UHSURMHFWLRQ PHWKRG DQG WKH 0/ PHWKRG RI FKHVW HPLVVLRQ SKDQWRP XVLQJ QRQVPRRWKHG VKRUW WUDQVPLVVLRQ VFDQ GDWD DUH VKRZQ LQ )LJ Ef Ff DQG Gf UHVSHFn WLYHO\ )URP WKLV ILJXUH ZH FDQ REVHUYH WKDW WKH 0/ PHWKRG WKDW SURGXFHV LPDJH ZLWK EHWWHU FRQWUDVW UHVROXWLRQ WKDQ WKH (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD DQG WKH /6 UHSURMHFWLRQ PHWKRG 7KH (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD DQG /6 UHSURMHFWLRQ PHWKRG ERWK SURGXFH LPDJH ZLWK ZURQJ LQIRUPDWLRQ 7KH

PAGE 85

)LJXUH 5HFRQVWUXFWLRQ RI FDUGLDF HPLVVLRQ SKDQWRP XVLQJ WKH QRQVPRRWKHG VKRUW WUDQVPLVVLRQ VFDQ GDWD Df HPLVVLRQ SKDQWRP Ef (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD Ff /6 UHSURMHFWLRQ PHWKRG Gf 0/ PHWKRG

PAGE 86

Ff )LJXUH 5HFRQVWUXFWLRQ RI FDUGLDF DWWHQXDWLRQ SKDQWRP XVLQJ QRQVPRRWKHG VKRUW WUDQVPLVVLRQ VFDQ GDWD Df DWWHQXDWLRQ SKDQWRP Ef 0/ PHWKRG Ff /6 UHSURMHFWLRQ PHWKRG

PAGE 87

)LJXUH 0/ PHWKRG Df HPLVVLRQ SKDQWRP Ef HPLVVLRQ LPDJH Ff DWWHQXDWLRQ SKDQWRP Gf DWWHQXDWLRQ LPDJH

PAGE 88

)LJXUH 5HFRQVWUXFWLRQ RI FKHVW HPLVVLRQ SKDQWRP XVLQJ QRQVPRRWKHG VKRUW WUDQVPLVVLRQ VFDQ GDWD Df HPLVVLRQ SKDQWRP Ef (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD Ff /6 UHSURMHFWLRQ PHWKRG Gf 0/ PHWKRG

PAGE 89

HPLVVLRQ LPDJH RI WKH /6 UHSURMHFWLRQ PHWKRG LV VPRRWKHU WKDQ WKH RQH RI WKH (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD :H SUHVHQW WKH DWWHQXDWLRQ LPDJHV RI WKH 0/ PHWKRG DQG WKH /6 UHSURMHFWLRQ PHWKRG LQ )LJ Ef DQG )LJ Ff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f Ff DQG Gf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

PAGE 90

Ff )LJXUH 5HFRQVWUXFWLRQ RI FKHVW SKDQWRP XVLQJ QRQVPRRWKHG VKRUW WUDQVn PLVVLRQ VFDQ GDWD Df DWWHQXDWLRQ SKDQWRP Ef 0/ PHWKRG Ff /6 UHSURMHFWLRQ PHWKRG

PAGE 91

)LJXUH 5HFRQVWUXFWLRQ RI FDUGLDF HPLVVLRQ SKDQWRP XVLQJ VPRRWKHG VKRUW WUDQVPLVVLRQ VFDQ GDWD Df HPLVVLRQ SKDQWRP Ef (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD Ff /6 UHSURMHFWLRQ PHWKRG Gf 0/ PHWKRG

PAGE 92

:H VKRZ WKH DWWHQXDWLRQ LPDJHV RI WKH 0/ PHWKRG DQG WKH /6 UHSURMHFWLRQ PHWKRG LQ )LJ Ef DQG )LJ Ff UHVSHFWLYHO\ 7KH DWWHQXDWLRQ LPDJH RI WKH /6 PHWKRG SURYLGHV WKH VKDSH RI WKH DWWHQXDWLRQ PDS ZKLOH WKH LPDJH RI WKH 0/ PHWKRG VXIIHUV IURP KHDY\ DUWLIDFWV 7KH HPLVVLRQ DQG DWWHQXDWLRQ LPDJH RI WKH 0/ PHWKRG LQLWLDOL]HG ZLWK IL DUH VKRZQ LQ )LJ Ef )LJ Gf UHVSHFWLYHO\ 7KH 0/ PHWKRG DOPRVW SURGXFHV WKH EHVW HPLVVLRQ LPDJHV IRU WKH DOO VKRUW WUDQVPLVVLRQ VFDQ H[SHULPHQWV ,Q SDUWLFXODUO\ LW SURYLGHV D YHU\ FOHDU UHFRQVWUXFWLRQ RI WKH KHDUW DUHD VHH )LJ Eff 7KH HPLVVLRQ LPDJHV RI WKH (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD WKH /6 UHSURMHFWLRQ PHWKRG DQG WKH 0/ PHWKRG RI WKH FKHVW HPLVVLRQ SKDQWRP XVLQJ VPRRWKHG VKRUW WUDQVPLVVLRQ VFDQ GDWD DUH VKRZQ LQ )LJ Ef Ff DQG Gf UHVSHFWLYHO\ )URP WKLV ILJXUH ZH FDQ REVHUYH WKDW WKH 0/ PHWKRG SURGXFHV DQ LPDJH ZLWK PRUH GHWDLOV WKDQ WKH RWKHU LPDJHV 1RWH WKH VPDOO KLJK GHQVLW\ VSRW DUHD DW WKH OHIW EHORZf :H SUHVHQW WKH DWWHQXDWLRQ LPDJHV RI WKH 0/ PHWKRG DQG WKH /6 UHSURMHFWLRQ PHWKRG LQ )LJ Ef DQG )LJ Ff UHVSHFWLYHO\ :H FDQ VHH WKDW WKH 0/ PHWKRG SURGXFHV D PRUH DFFXUDWH DWWHQXDWLRQ LPDJH WKDQ WKH /6 UHSURMHFWLRQ PHWKRG IDLOV WR UHVROYH DQ\WKLQJ QRWH WKH WZR VSRWV DW WKH FHQWHUf &RPSDUHG WR WKH DWWHQXDWLRQ LPDJHV XVLQJ QRQVPRRWKHG WUDQVPLVVLRQ VFDQ GDWD DWWHQXDWLRQ LPDJHV KHUH KDYH PRUH GHWDLOV DQG IHDWXUHV 7KH FRPSDULVRQV RI WKH 3615 IRU WKH HPLVVLRQ LPDJHV XVLQJ WKH VKRUW WUDQVn PLVVLRQ VFDQ GDWD DUH VKRZQ LQ 7DEOH $JDLQ WKH 0/ PHWKRG KDV WKH ODUJHVW 3615 ,Q WKH QH[W VHFWLRQ ZH VKRZ WKH UHVXOWV XVLQJ WKH UHDO GDWD

PAGE 93

‘ ‘ ‘ ‘ /f§ r r ‘ ‘ A Ff )LJXUH 5HFRQVWUXFWLRQ RI FDUGLDF DWWHQXDWLRQ SKDQWRP XVLQJ VPRRWKHG VKRUW WUDQVPLVVLRQ VFDQ GDWD Df DWWHQXDWLRQ SKDQWRP Ef 0/ PHWKRG Ff /6 UHSURMHFn WLRQ PHWKRG ,

PAGE 94

)LJXUH 0/ PHWKRG Df HPLVVLRQ SKDQWRP Ef HPLVVLRQ LPDJH Ff DWWHQXDWLRQ SKDQWRP Gf DWWHQXDWLRQ LPDJH

PAGE 95

)LJXUH 5HFRQVWUXFWLRQ RI FKHVW HPLVVLRQ SKDQWRP XVLQJ VPRRWKHG VKRUW WUDQVPLVVLRQ VFDQ GDWD Df HPLVVLRQ SKDQWRP Ef (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD Ff /6 UHSURMHFWLRQ PHWKRG Gf 0/ PHWKRG

PAGE 96

r n f Ff )LJXUH 5HFRQVWUXFWLRQ RI FKHVW DWWHQXDWLRQ SKDQWRP XVLQJ VPRRWKHG VKRUW WUDQVPLVVLRQ VFDQ GDWD Df DWWHQXDWLRQ SKDQWRP Ef 0/ PHWKRG Ff /6 UHSURMHFn WLRQ PHWKRG

PAGE 97

G% FDUGLDF SKDQWRP FKHVW SKDQWRP $OJRULWKP 6WDQGDUG 0/ /65HS 6WDQGDUG 0/ /65HS 1RQVPRRWK 6PRRWK 7DEOH 3615 G%f RI UHFRQVWUXFWLRQ RI FDUGLDF HPLVVLRQ SKDQWRP XVLQJ VKRUW WUDQVPLVVLRQ VFDQ GDWD 5HDO 'DWD :H REWDLQHG UHDO GDWD IURP 'U -RKQ 9RWDZ $VVRFLDWH 3URIHVVRU RI 5DGLRORJ\ DW WKH (PRU\ 8QLYHUVLW\ +RVSLWDO 7R FUHDWH WKH GDWD D WKRUD[ SKDQWRP ZLWK D KRW KHDUW WZR OXQJV D VSLQH DQG VRIW WLVVXH ZDV VFDQQHG XVLQJ WKH 6LHPHQV &7, (&$7 (;$&7 3(7 VFDQQHU 7KLV VFDQQHU LV FRPSULVHG RI ULQJV RI GLDPHWHU FP ZKHUH HDFK ULQJ FRQWDLQV GHWHFWRUV RI VL]H [ PP ,Q DGGLWLRQ WKH GLDPHWHU RI WKH VFDQQHUf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

PAGE 98

VL]H [ FRUUHVSRQGLQJ WR DQJOHV ZLWK ELQV SHU DQJOH 7KH QXPEHU RI WXEH LV f§ ,Q WKH QH[W VHFWLRQ ZH GLVFXVV GHWDLOV RI WKH LPSOHPHQWDWLRQ 2YHUYLHZ RI ,PSOHPHQWDWLRQ $OO WKH LPDJHV VKRZQ DUH IRU WKH WK SODQH 7KLV SODQH ZDV FKRVHQ EHFDXVH LW FRQWDLQV PRVW RI WKH PDMRU IHDWXUHV RI WKH WKRUD[ SKDQWRP 7KH WK SODQH FRQWDLQV PLOOLRQ HPLVVLRQ FRXQWV LH 9L [ f DQG PLOOLRQ EODQN VFDQ FRXQWV 7KH QXPEHU RI FRXQWV IRU WKH ORZ TXDOLW\ WUDQVPLVVLRQ VFDQ LV PLOOLRQ LH - L UL [ f ZKLOH WKH KLJK TXDOLW\ WUDQVPLVVLRQ VFDQ FRQWDLQV PLOOLRQ FRXQWV LH U [ f ,W VKRXOG EH QRWHG WKDW WKH WUDQVPLVVLRQ VFDQV ZHUH DXWRPDWLFDOO\ VPRRWKHG XVLQJ WKH VFDQQHUfV GDWD SURFHVVLQJ DOJRULWKP 7KH HPLVVLRQ DQG WUDQVPLVVLRQ GDWD KDYH EHHQ FRUUHFWHG IRU GHWHFWRU LQHIILFLHQF\ DQG DFFLGHQWDO FRLQFLGHQFHV XVLQJ VWDQGDUG SURFHGXUHV KRZHYHU QR VFDWWHU FRUUHFWLRQ KDV EHHQ GRQH )RU WKH 0/ PHWKRG ZH UXQ JOREDO LWHUDWLRQV DQG LWHUDWLRQV IRU WKH [VWHS DQG SVWHS UHVSHFWLYHO\ ,Q RWKHU ZRUGV WKH 0/ PHWKRG UHTXLUHV WRWDO LWHUDWLRQV DQG LWHUDWLRQV WR JHW DQ HPLVVLRQ LPDJH DQG DQ DWWHQXDWLRQ LPDJH UHVSHFWLYHO\ 7KH LQLWLDO HVWLPDWH IRU [VWHS LV brfM f§ = L MU IrU K ZKHUH FT A‘AWf 7KH LQLWLDO HVWLPDWH IRU VWHS LV [rf\ IRU DOO M 7KH LQLWLDO HVWLPDWH IRU WKH VXUYLYDO SUREDELOLW\ LV Dp ZKHUH Ut PLQXWHV DQG 7U PLQXWHV 7ZHQW\ LWHUDWLRQV RI WKH (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD PHWKRG LV LQLWLDOn L]HG XVLQJ [r A IRU DOO M ZKHUH )RU WKH /6 UHSURMHFWLRQ PHWKRG

PAGE 99

WR LQLWLDOL]H HT f ZH XVH D XQLIRUP LQLWLDO HVWLPDWH Lr IRU DOO M :H UXQ LWHUDWLRQV WR JHW KH /6 HVWLPDWH RI D IL 7KHQ ZH UXQ LWHUDWLRQV RI WKH DOJRULWKP VKRZQ LQ HT f 7KXV /6 UHSURMHFWLRQ PHWKRG UHTXLUHV WRWDO M2 LWHUDWLRQV WR JHW DQ HPLVVLRQ LPDJH 7KH WUDQVPLVVLRQ VFDQ LV D KLJK TXDOLW\ VFDQ EHFDXVH LW ZDV SHUIRUPHG ORQJ DIWHU WKH HPLVVLRQ VFDQ ZDV GRQH QR HPLVVLRQ FRXQWV ZLOO EH PLVUHJLVWHUHG DV WUDQVPLVVLRQ FRXQWVf 5HVXOWV 7KH HPLVVLRQ LPDJHV RI WKH ILOWHUHG EDFNSURMHFWLRQ PHWKRG WKH (00/ DOJRn ULWKP ZLWK FRUUHFWHG GDWD WKH /6 UHSURMHFWLRQ PHWKRG DQG WKH 0/ PHWKRG DUH VKRZQ LQ )LJ Df )LJ Ef )LJ Ff DQG )LJ Gf UHVSHFn WLYHO\ )URP WKLV ILJXUH ZH REVHUYH WKDW WKH (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD SURGXFHV DQ LPDJH ZLWK FOHDUHU OXQJV ERXQGDULHV RI WKH RWKHU PHWKRGV $OVR WKH 0/ PHWKRG SURGXFHV FRPSDUDEOH EXW VPRRWKHU LPDJH WKDQ WKH (00/ DOJRn ULWKP ZLWK FRUUHFWHG GDWD 7KH ILOWHUHG EDFNSURMHFWLRQ PHWKRG SURYLGHV D QRLV\ HPLVVLRQ LPDJH ZLWK ORWV RI DUWLIDFWV HYHU\ZKHUH ([FHSW WKH HPLVVLRQ LPDJHV RI WKH ILOWHUHG EDFNSURMHFWLRQ PHWKRG DOO HPLVVLRQ LPDJHV KDYH D fULQJf DUWLIDFW RXWVLGH WKH SKDQWRP VHH )LJ Ff DQG )LJ Gff ZKLFK LV QRW YLVLEOH LQ WKH LPDJH XVLQJ WKH (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD EHFDXVH WKH OLPLWHG JUD\ OHYHO VFDOHV RI WKH LPDJH WKH KHDUW DUHD LQ )LJ Ef KDV KLJK LQWHQVLW\ YDOXHV FRPSDUHG WR WKH EDFNJURXQG VR WKH fULQJf DUWLIDFWV RQ WKH EDFNJURXQG FDQQRW EH VHHQf

PAGE 100

)LJXUH (PLVVLRQ LPDJHV XVLQJ KLJK TXDOLW\ VPRRWKHG WUDQVPLVVLRQ VFDQ GDWD Df )LOWHUHG EDFNSURMHFWLRQ PHWKRG Ef (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD Ff /6 UHSURMHFWLRQ PHWKRG Gf 0/ PHWKRG

PAGE 101

7KH DWWHQXDWLRQ LPDJHV RI WKH 0/ PHWKRG DQG WKH /6 UHSURMHFWLRQ PHWKRG DUH VKRZQ LQ )LJ Df DQG )LJ Ef UHVSHFWLYHO\ 7KH DWWHQXDWLRQ LPDJH RI WKH /6 PHWKRG LQ )LJ Ef FOHDUO\ GHOLQHDWHV OXQJV VSLQH DQG WLVVXH UHJLRQV ZKLOH WKH DWWHQXDWLRQ LPDJH RI WKH 0/ PHWKRG VHH )LJ Dff VWLOO VXIIHUV IURP OLQH DUWLIDFWV $GGLWLRQDOO\ WKH DWWHQXDWLRQ LPDJH RI WKH 0/ PHWKRG LV LQIOXHQFHG E\ WKH HPLVVLRQ GDWD QRWH WKH ZKLWH VSRW DW WKH FHQWHUf :H DOVR VKRZ WKH HPLVVLRQ LPDJHV RI WKH (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD DQG WKH 0/ PHWKRG XVLQJ WKH QRQVPRRWKHG WUDQVPLVVLRQ VFDQ GDWD LQ )LJ Df DQG )LJ Ef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

PAGE 102

)LJXUH $WWHQXDWLRQ LPDJHV XVLQJ KLJK TXDOLW\ VPRRWKHG WUDQVPLVVLRQ VFDQ GDWD Df 0/ PHWKRG Ef /6 UHSURMHFWLRQ PHWKRG

PAGE 103

)LJXUH (PLVVLRQ LPDJH XVLQJ KLJK TXDOLW\ QRQVPRRWKHG WUDQVPLVVLRQ VFDQ GDWD Df (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD Ef 0/ PHWKRG

PAGE 104

&+$37(5 &21&/86,216 ,Q WKLV GLVVHUWDWLRQV D 0/ PHWKRG DQG D /6 UHSURMHFWLRQ PHWKRG ZHUH SURn SRVHG IRU FRUUHFWLQJ DWWHQXDWLRQ LQ SRVLWURQ HPLVVLRQ WRPRJUDSK\ 7KH 0/ PHWKRG VLPXOWDQHRXVO\ HVWLPDWHV WKH HPLVVLRQ LQWHQVLW\ DQG WKH DWWHQXDWLRQ GHQVLW\ )RU WKH /6 UHSURMHFWLRQ PHWKRG WKH DWWHQXDWLRQ GHQVLW\ LV ILUVW HVWLPDWHG XVLQJ D OHDVW VTXDUHV HVWLPDWRU 7KHQ WKLV HVWLPDWH LV UHSURMHFWHG WR HVWLPDWH WKH VXUn YLYDO SUREDELOLWLHV ZKLFK DUH XVHG WR PRGLI\ WKH SUREDELOLW\ PDWUL[ )LQDOO\ WKH HPLVVLRQ LQWHQVLW\ LV HVWLPDWHG XVLQJ WKH (00/ DOJRULWKP ZLWK WKH PRGLILHG SUREDELOLW\ PDWUL[ ,Q VLPXODWLRQ VWXGLHV WKH SURSRVHG PHWKRGV ZHUH FRPSDUHG WR WKH (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD XVLQJ V\QWKHWLF GDWD DQG UHDO GDWD %HORZ ZH VXP XS WKH DGYDQWDJHV DQG GLVDGYDQWDJHV RI WKH SURSRVHG PHWKRGV f 7KH 0/ PHWKRG SURGXFHV HPLVVLRQ LPDJHV ZLWK EHWWHU FRQWUDVW DQG PRUH GHWDLOV ZLWK OHVV FRPSXWDWLRQDO FRPSOH[LW\ WKDQ WKH (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD DQG WKH /6 UHSURMHFWLRQ PHWKRG ZKHQ WKH VKRUW WUDQVPLVVLRQ VFDQ GDWD LV XVHG f 7KH 0/ PHWKRG SURYLGHV PRUH DFFXUDWH HPLVVLRQ LPDJHV WKDQ WKH (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD DQG WKH /6 UHSURMHFWLRQ PHWKRG

PAGE 105

f 7KH /6 UHSURMHFWLRQ PHWKRG SURGXFHV FRPSDUDEOH HPLVVLRQ LPDJHV EXW LV OHVV DFFXUDWH WKDQ WKH 0/ PHWKRG ZKLOH WKH WUDQVPLVVLRQ VFDQ GDWD DUH VPRRWKHG f 7KH /6 UHSURMHFWLRQ PHWKRG SURGXFHV EHWWHU DWWHQXDWLRQ LPDJHV WKDQ WKH 0/ PHWKRG IRU FDUGLDF DWWHQXDWLRQ SKDQWRP f 7KH 0/ PHWKRG SURGXFHV EHWWHU DWWHQXDWLRQ LPDJHV WKDQ WKH /6 UHSURMHFn WLRQ PHWKRG IRU FKHVW DWWHQXDWLRQ SKDQWRP f 7KH (00/ DOJRULWKP ZLWK FRUUHFWHG GDWD SURGXFHV PXFK EHWWHU HPLVVLRQ LPDJHV ZKHQ WKH WUDQVPLVVLRQ VFDQ GDWD LV VPRRWKHG *LYHQ WKH SHUIRUPDQFH RI WKH SURSRVHG PHWKRGV LW LV IHOW WKDW WKH\ ZLOO EH PRVW XVHIXO LQ VKRUW WUDQVPLVVLRQ VFDQ SURWRFROV )XWXUH :RUN 7KH ZRUN LQ WKLV WKHVLV SURYLGHV D QHZ DSSURDFK WR FRUUHFW WKH DWWHQXDWLRQ LQ 3(7 :LWK WKH VLPXODWLRQV XVLQJ WKH UHDO GDWD REWDLQHG IURP WKH (PRU\ 8QLYHUn VLW\ WKH 0/ PHWKRG VWHSV LQ WKH UHDO OLIH DSSOLFDWLRQ UDWKHU WKDQ WKH WKHRUHWLFDO ZRUN ,W SURYLGHV DQ DOWHUQDWH WRRO IRU WKH PHGLFDO VRFLHW\ WR XVH WKH UHVHDUFK UHVXOW LQ FOLQLFV :LWK PRUH UHVHDUFK GRQH RQ VHDUFKLQJ GLIIHUHQW HVWLPDWRUV DQG RSWLn PL]DWLRQ DOJRULWKPV VXFK DV WKH F\FOLF FRRUGLQDWH GHVFHQW JROGHQ VHFWLRQ PHWKRG WKH 0/ PHWKRG FDQ EH H[SHFWHG WR LPSURYH WKH HPLVVLRQ LPDJHV PRUH ,Q DGn GLWLRQ ZH PD\ XVH WKH LQIRUPDWLRQ IURP WKH KLVWRJUDP RI WKH UHFRQVWUXFWLRQ RI DWWHQXDWLRQ GHQVLW\ WR VHJPHQW WR WKH GLIIHUHQW PHGLXPV 2QFH ZH NQRZ ZKHUH WKH GLIIHUHQW PHGLXPV GLVWULEXWH ZH FDQ DVVLJQ WKH FRPPRQ DWWHQXDWLRQ GHQVLW\

PAGE 106

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VWHS WKH FKRLFH RI WKH QXPEHU RI WKH JOREDO LWHUDWLRQ WR REWDLQ WKH EHVW HPLVVLRQ LPDJHV WKH FRQYHUJHQFH RI WKH DOJRULWKP DQG WKH VWDWLVWLFDO DQDO\VLV RI WKH HVWLPDWH DUH RSHQ LVVXHV )RU /6 UHSURMHFWLRQ PHWKRG D VWRSSLQJ UXOH DQG WKH DQDO\VLV RI WKH DOJRULWKP DOVR QHHGV PRUH LQYHVWLJDWLRQ

PAGE 107

$33(1',; $ (00/ $OJRULWKP IRU 3(7 6KHSS DQG 9DUGL XVHG WKH (0 DOJRULWKP >@ WR REWDLQ WKH 0/ HVWLPDWH RI WKH HPLVVLRQ LQWHQVLW\ IRU HUURU IUHH GDWD >@ ,Q WKHLU IRUPXODWLRQ WKH LQFRPSOHWH GDWD LV WKH REVHUYHG GDWD \ DQG WKH FRPSOHWH GDWD LV ^$AM` ZKHUH $UM LV WKH QXPEHU RI HPLVVLRQV LQ YR[HO M WKDW DUH GHWHFWHG E\ WXEH L 8QGHU WKH DVVXPSWLRQ WKDW ^1LM` DUH LQGHSHQGHQW UDQGRP YDULDEOHV DQG WKH DVVXPSWLRQV LQ 6HFWLRQ WKH (0 DOJRULWKP FRQVLVWV RI WKH IROORZLQJ VWHSV ,Q WKH (VWHS WKH H[SHFWHG YDOXH RI WKH ORJ OLNHOLKRRG IXQFWLRQ IRU WKH FRPSOHWH GDWD JLYHQ WKH LQFRPSOHWH GDWD DQG WKH FXUUHQW HVWLPDWH RI WKH HPLVVLRQ LQWHQVLW\ LV FRPSXWHG /HW / EH WKH ORJ OLNHOLKRRG RI WKH FRPSOHWH GDWD /\Q\L\LM\LL\L\LMf ORJ31Q \Q Q \X $Lf 1X \X1Q ML1L \Xf $f 8[[Nf (>/1X $L 1?M 1Q $ 1cMf?\ [N [@ M -LSLM[1D LrJ3MrMf`_\ [N[? L L M L $f $f e=^3LL;L 3LM;M SrfL \L ORJ 3M ;Mf` $f ,Q WKH 0VWHS D QHZ HVWLPDWH IRU WKH HPLVVLRQ LQWHQVLW\ LV REWDLQHG E\ PD[LPL]LQJ 8[[Nf ZLWK UHVSHFW WR [ [ IFO eL DUJ 8[ [Nf $f

PAGE 108

,WHUDWLQJ EDFN DQG IRUWK EHWZHHQ WKH (VWHS DQG 0VWHS OHDGV WR WKH IROORZLQJ LWHUDWLRQ N B [M [M = M f§ $f W 3rNfL ZKHUH [?f LV WKH NWK HVWLPDWH RI ;M 7KLV LWHUDWLRQ LV FDOOHG WKH H[SHFWDWLRQ PD[n LPL]DWLRQ PD[LPXP OLNHOLKRRG (00/f DOJRULWKP 'HWDLOV RQ WKH (00/ DOJRn ULWKP FDQ EH IRXQG LQ >@

PAGE 109

5()(5(1&(6 >@ / 3 $GOHU 3 ) )DXOKDEHU & 6FKQXU 1 / $O.DVL DQG 5 5 6KHQN $[LOODU\ O\PSK QRGH PHWDVWDVHV VFUHHQLQJ ZLWK ) GHR[\IOXRUR' JOXFRVH )'*f 3(7 5DGLRORJ\ f 0D\ >@ 0 0 $QGHUVRQ % $ 0DLU DQG 0 5DR 3RVLWURQ HPLVVLRQ WRPRJUDSK\ $ UHYLHZ &RQWURO DQG &\EHUQHWLFV f f§ >@ 0 0 $QGHUVRQ % $ 0DLU 0 5DR DQG &KHQ+VLHQ :X :HLJKWHG OHDVW VTXDUHV UHFRQVWUXFWLRQ PHWKRGV IRU SRVLWURQ HPLVVLRQ WRPRJUDSK\ ,((( 7UDQV RQ 0HGLFDO ,PDJLQJ f $SULO >@ 0 0 $QGHUVRQ & @ 0 %DULQDJD %UDLQ UHVHDUFKHUV VSHDN D FRPPRQ ODQJXDJH 6FLHQFH 'HF >@ 0 6 %D]DUDD + 6KHUDOL DQG & 0 6KHWW\ 1RQOLQHDU SURJUDPPLQJ WKHRU\ DQG DOJRULWKPV FKDSWHU -RKQ :LOH\ DQG 6RQV ,QF HGLWLRQ >@ 6KDURQ %HJOH\ +RZ WR EXLOG D EDE\f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

PAGE 110

DFWLYLW\ FRHIILFLHQWV ,((( 7UDQV RQ 0HGLFDO ,PDJLQJ 16f $SULO >@ 1 + &OLQWKRUQ &RQVWUDLQHG OHDVWVTXDUHV YV PD[LPXP OLNHOLKRRG UHFRQn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n FRPSOHWH GDWD YLD WKH (0 DOJRULWKP -RXUQDO RI WKH 5R\DO 6WDWLVWLFDO 6RFLHW\ 6HULHV % >@ $ )HVVOHU 3HQDOL]HG ZHLJKWHG OHDVWVTXDUHV LPDJH UHFRQVWUXFWLRQ IRU SRVLWURQ HPLVVLRQ WRPRJUDSK\ ,((( 7UDQV RQ 0HGLFDO ,PDJLQJ >@ $ )HVVOHU 1 + &OLQWKRUQH DQG : / 5RJHU 2Q FRPSOHWHGDWD VSDFHV IRU 3(7 UHFRQVWUXFWLRQ DOJRULWKPV ,((( 7UDQV RQ 1XFOHDU 6FLHQFH ff§ >@ + )LVFKHU :LN DQG 0 )UHGULNVRQ )XQFWLRQDO QHXURDQDWRP\ RI UREn EHU\ UHH[SHULHQFH DIIHFWLYH PHPRULHV VWXGLHG ZLWK 3(7 1HXURUHSRUW f 6HSWHPEHU >@ 6 *HPDQ DQG ( 0F&OXUH %D\HVLDQ LPDJH DQDO\VLV $Q DSSOLFDWLRQ WR VLQJOH SKRWRQ HPLVVLRQ WRPRJUDSK\ ,Q 3URF 6WDWLVW &RPSXW 6HFW $PHU 6WDWLVW $VVQ SDJHV :DVKLQJWRQ '& >@ 7 +HEHUW DQG 5 /HDK\ $ JHQHUDOL]HG (0 DOJRULWKP IRU WKH %D\HVLDQ UHFRQVWUXFWLRQ IURP 3RLVVRQ GDWD XVLQJ *LEEV SULRUV ,((( 7UDQV RQ 0HGLFDO ,PDJLQJ

PAGE 111

>@ ( +RIIPDQ HW DO 4XDQWL]DWLRQ LQ SRVLWURQHPLVVLRQ FRPSXWHG WRPRJUDSK\ HIIHFW RI DFFLGHQWDO FRLQFLGHQFHV RI &RPSXWHU $VVLVWHG 7RPRJUDSK\ >@ 6 +XDQJ HW DO 4XDQWL]DWLRQ LQ SRVLWURQHPLVVLRQ FRPSXWHG WRPRJUDSK\ HIIHFWV RI LQDFFXUDWH DWWHQXDWLRQ FRUUHFWLRQ RI &RPSXWHU $VVLVWHG 7RPRJn UDSK\ >@ / .DXIPDQ 0D[LPXP OLNHOLKRRG OHDVW VTXDUHV DQG SHQDOL]HG OHDVW VTXDUHV IRU 3(7 ,((( 7UDQV RQ 0HGLFDO ,PDJLQJ >@ %HWW\DQQ .HYOHV 1DNHG WR WKH ERQH PHGLFDO LPDJLQJ LQ WKH WZHQWLHWK FHQn WXU\ FKDSWHU SDJHV 5XWJHUV 8QLYHUVLW\ 3UHVV 1HZ %UXQVZLFN 1>@ ( 5 .RSV + +HU]RJ $ 6FKPLG 6 +ROWH DQG / )HLQHQGHJHQ 3HUIRUPDQFH FKDUDFWHULVWLFV RI DQ HLJKWULQJ ZKROH ERG\ 3(7 VFDQQHU RI &RPSXWHU $VVLVWHG 7RPRJUDSK\ f 0D\-XQH >@ 6 0 .RVVO\Q HW DO ,GHQWLI\LQJ REMHFWV VHHQ IURP GLIIHUHQW YLHZSRLQWV $ 3(7 LQYHVWLJDWLRQ %UDLQ >@ /DQJH DQG 5 &DUVRQ (0 UHFRQVWUXFWLRQ DOJRULWKPV IRU HPLVVLRQ DQG WUDQVPLVVLRQ WRPRJUDSK\ RI &RPSXWHU $VVLVWHG 7RPRJUDSK\ >@ /DQJH DQG $ )HVVOHU *OREDOO\ FRQYHUJHQW DOJRULWKPV IRU PD[LPXP D SRVWHULRU WUDQVPLVVLRQ WRPRJUDSK\ ,((( 7UDQV RQ ,PDJH 3URFHVVLQJ f f§ 2FWREHU >@ ( /HYLWDQ DQG 7 +HUPDQ $ PD[LPXP D SRVWHULRUL SUREDELOLW\ H[SHFn WDWLRQ PD[LPL]DWLRQ DOJRULWKP IRU UHFRQVWUXFWLRQ LQ HPLVVLRQ WRPRJUDSK\ ,((( 7UDQV RQ 0HGLFDO ,PDJLQJ >@ = /LDQJ DQG 5 -DV]F]DN 2Q ED\HVLDQ LPDJH UHFRQVWUXFWLRQ IURP SURMHFWLRQV ,((( 7UDQV RQ 0HGLFDO ,PDJLQJ >@ =KLGRQJ /X $ VWDWLVWLFDO PHWKRG RI UHFRQVWUXFWLQJ ERWK DFWLYLW\ DQG DWWHQXn DWLRQ LQ HPLVVLRQ FRPSXWHG WRPRJUDSK\ 3K' WKHVLV %URZQ 8QLYHUVLW\ 0D\ >@ $ 0DXUL & ,QLJXH] -HULFR 6 'LD],QVD ) $EDG DQG ) 0RUDOHV (SLOHSWLF VHL]XUHV GXULQJ VOHHS 5HY 1HXURO f f§ 2FWREHU

PAGE 112

>@ 6 5 0HLNOH 0 'DKOERP DQG 6 5 &KHUU\ $WWHQXDWLRQ FRUUHFWLRQ XVn LQJ FRXQWOLPLWHG WUDQVPLVVLRQ GDWD LQ SRVLWURQ HPLVVLRQ WRPRJUDSK\ 7KH -RXUQDO RI 1XFOHDU 0HGLFLQH f f§ -DQ >@ :LOOLDP : 0RVHV )XQGDPHQWDOV RI PHGLFDO LPDJLQJ FKDSWHU SDJHV 3(73(7 ,((( 1XFOHDU 6FLHQFH 6\PSRVLXP DQG 0HGLFDO ,PDJLQJ &RQn IHUHQFH 6DQ )UDQFLVFR &$ 2FWREHU >@ ( 0XPFXRJOX 5 /HDK\ DQG 6 &KHUU\ $ VWDWLVWLFDO DSSURDFK WR WUDQVn PLVVLRQ LPDJH UHFRQVWUXFWLRQ IURP ULQJ VRXUFH FDOLEUDWLRQ PHDVXUHPHQWV LQ 3(7 ,Q PLF YROXPH SDJHV 2UODQGR )/ >@ 0 2OOLQJHU DQG $ )HVVOHU 3RVLWURQHPLVVLRQ WRPRJUDSK\ ,((( 6LJQDO 3URFHVVLQJ 0DJD]LQH ff§ -DQXDU\ >@ $WKDQDVLRV 3DSRXOLV 3UREDELOLW\ UDQGRP YDULDEOHV DQG VWDWLVWLF SURFHVVHV FKDSWHU SDJHV 0F*UDZ +DOO >@ 0 ( 3KHOSV ( +RIIPDQ 1 $ 0XOODQL DQG 0 0 7HU3RJRVVLDQ $Sn SOLFDWLRQV RI DQQLKLODWLRQ FRLQFLGHQFH GHWHFWLRQ WR WUDQVD[LDO UHFRQVWUXFWLRQ WRPRJUDSK\ -RXUQDO RI 1XFOHDU 0HGLFLQH >@ $ 5 'H 3LHUUR 2Q WKH FRQYHUJHQFH RI WKH LWHUDWLYH LPDJH VSDFH UHFRQVWUXFn WLRQ DOJRULWKP IRU YROXPH (&7 ,((( 7UDQV RQ 0HGLFDO ,PDJLQJ -XQH >@ 3ROLWWH DQG / 6Q\GHU &RUUHFWLRQV IRU DFFLGHQWDO FRLQFLGHQFHV DQG DWn WHQXDWLRQ LQ PD[LPXPOLNHOLKRRG LPDJH UHFRQVWUXFWLRQ IRU SRVLWURQHPLVVLRQ WRPRJUDSK\ ,((( 7UDQV RQ 0HGLFDO ,PDJLQJ ff§ 0DUFK >@ ( 6DOPRQ HW DO 'LIIHUHQWLDO GLDJQRVLV RI $O]KHLPHUfV GLVHDVH ZLWK 3(7 RI 1XFOHDU 0HGLFLQH ff§ 0DUFK >@ .KDOLG 6D\RRG ,QWURGXFWLRQ WR GDWD FRPSUHVVLRQ FKDSWHU SDJH 0RUn JDQ .DXIPDQQ 3XEOLVKHUV ,QF >@ /RXLV / 6FKDUI 6WDWLVWLFDO VLJQDO SURFHVVLQJ GHWHFWLRQ HVWLPDWLRQ DQG WLPH VHULHV DQDO\VLV FKDSWHU $GGLVRQ :HVOH\ >@ / $ 6KHSS DQG % ) /RJDQ 7KH )RXULHU UHFRQVWUXFWLRQ RI D KHDG VHFWLRQ ,((( 7UDQV RQ 0HGLFDO ,PDJLQJ >@ / $ 6KHSS DQG < 9DUGL 0D[LPXP OLNHOLKRRG UHFRQVWUXFWLRQ IRU HPLVVLRQ WRPRJUDSK\ ,((( 7UDQV RQ 0HGLFDO ,PDJLQJ

PAGE 113

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n YHUVLW\ +RVSLWDO >@ 0LFKDHO :HOFK 5HVHDUFK LQ WKH GHYHORSPHQW RI 3(7 UDGLRSKDUPDFHXWLFDOV RI &RPSXWHU $VVLVWHG 7RPRJUDSK\ f1 DQG 1 $SULO >@ :LHQKDUG / (ULNVVRQ 6 *URRWRRQN 0 &DVH\ 8 3LHWU]\N DQG : +HLVV 3HUIRUPDQFH HYDXODWLRQ RI WKH SRVLWURQ VFDQQHU (&$7 (;$&7 RI &RPSXWHU $VVLVWHG 7RPRJUDSK\ ff§ 6HSW2FW > : =DQJZLOO 1RQOLQHDU 3URTUDPPLQT FKDSWHU SDJHV 3UHQWLFH +DOO 1HZ
PAGE 114

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c:D\ %HUQDUG $ 0DLU 3URIHVVRU RI 0DWKHPDWLFV FHUWLI\ WKDW KDYH UHDG WKLV VWXG\ DQG WKDW LQ P\ RSLQLRQ LW FRQIRUPV WR DFFHSWDEOH VWDQGDUGV RI VFKRODUO\ SUHVHQWDWLRQ DQG LV IXOO\ DGHTXDWH LQAVFRSA DQG TXDOLW\ DV D GLVVHUWDWLRQ IRU WKH GHJUHH RI 'RFWRU RI 3KLORVRSK\ nQQSSH RI (OHFWULFDO DQG &RPSXWHU SQHHULQJ FHUWLI\ WKDW KDYH UHDG WKLV VWXG\ DQG WKDW LQ P\ RSLQLRQ LW FRQIRUPV WR DFFHSWDEOH VWDQGDUGV RI VFKRODUO\ SUHVHQWDWLRQ DQG LV IXOO\ DGHTXDWH LQ VFRSH DQG TXDOLW\ DV D GLVVHUWDWLRQ IRU WKH GHJUHH RI 'RFWRU RI 3KLORVRSK\ 9r 0XUDOL 5DR 3URIHVVRU RI 0DWKHPDWLFV

PAGE 115

7KLV GLVVHUWDWLRQ ZDV VXEPLWWHG WR WKH *UDGXDWH )DFXOW\ RI WKH &ROOHJH RI (QJLQHHULQJ DQG WR WKH *UDGXDWH 6FKRRO DQG ZDV DFFHSWHG DV SDUWLDO IXOILOOPHQW RI WKH UHTXLUHPHQWV IRU WKH GHJUHH RI 'RFWRU RI 3KLORVRSK\ 0D\ :LQIUHG 0 3KLOOLSV 'HDQ &ROOHJH RI (QJLQHHULQJ .DUHQ $ +ROEURRN 'HDQ *UDGXDWH 6FKRRO

PAGE 116

81,9(56,7< 2) )/25,'$


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 EBI62KCZC_MA70C2 INGEST_TIME 2014-05-14T22:36:07Z PACKAGE AA00021354_00001
AGREEMENT_INFO ACCOUNT UF PROJECT UFDC
FILES