<%BANNER%>

Non-Invasive Detection of Pre-Malignant Lesions Using Sub-Ablative, Deep Ultraviolet Laser-Tissue Interactions Coupled w...

Permanent Link: http://ufdc.ufl.edu/UFE0045034/00001

Material Information

Title: Non-Invasive Detection of Pre-Malignant Lesions Using Sub-Ablative, Deep Ultraviolet Laser-Tissue Interactions Coupled with Endogenous Tissue Fluorescence Proof-Of-Concept, Computational Modeling, and Detection-Theory Strategies for Clinical Instrument Design and Testing
Physical Description: 1 online resource (191 p.)
Language: english
Creator: Kozikowski, Raymond T
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2012

Subjects

Subjects / Keywords: biophotonics -- biosensing -- cancer -- excimer -- fluorescence -- spectroscopy
Biomedical Engineering -- Dissertations, Academic -- UF
Genre: Biomedical Engineering thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Half of all cancers are superficial in nature, originating in the stratified squamous epithelia including cervical, colorectal, skin, and oral cancer. Some of these diseases have seen a dramatic reduction in lethality due to the benefits of improved screening and diagnosis promoting earlier interdiction. Others, while treatable if caught at an early stage, have overall five year survival rates below 50% because they are challenging to identify early. These superficial carcinomas provide an opportunity for technological approaches to early detection due to the nature of their development: malignancy is invariably preceded by dysplastic precancerous cellular changes, which are often confined to the epithelial layer. These earliest changes are often not detectable visually, but are accessible using in vivo spectroscopy. Optical spectroscopies have been widely investigated as techniques for identifying pathological tissue; however, unrelated subject-to-subject variations in spectra complicate interpretation and consequently clinical adaptation has been limited. In this dissertation a new biosensing technique, differential laser-induced perturbation spectroscopy (DLIPS), is described and evaluated. This technique combines fluorescence probing (pre- and post-perturbation) with sub-ablative deep UV perturbation and difference spectroscopy to provide a new spectral dimension, facilitating two improvements over traditional techniques. First, this technique eliminates significant variations in absolute fluorescence within subject populations. Second, UV perturbations alter superficial protein layers, directly coupling the response to the spatio-biomolecular structure. This work is focused on the confluence of this innovation and the unmet clinical need for tools to detect pre-cancerous transformations earlier and with greater accuracy. In a proof-of-concept study, this technique is shown to work at least as well as fluorescence spectroscopy alone in detecting cancer but couples more specifically to changes induced by the dysplastic state, and putative relationships are drawn between tissues spectra and their biomolecular basis. To further understand the benefits of this technique, Monte Carlo modeling of DLIPS tissue spectroscopy is also presented, demonstrating that excimer perturbation enhances the returned molecular signatures from tissue layers and specifically emphasizes important disease biomarkers. Finally in a pre-clinical model of both oral dysplasia and benign pathology, detection performance and optimization of DLIPS and fluorescence technologies are considered from a detection-theory perspective.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Raymond T Kozikowski.
Thesis: Thesis (Ph.D.)--University of Florida, 2012.
Local: Adviser: Sorg, Brian.
Local: Co-adviser: Ormerod, Brandi K.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2012
System ID: UFE0045034:00001

Permanent Link: http://ufdc.ufl.edu/UFE0045034/00001

Material Information

Title: Non-Invasive Detection of Pre-Malignant Lesions Using Sub-Ablative, Deep Ultraviolet Laser-Tissue Interactions Coupled with Endogenous Tissue Fluorescence Proof-Of-Concept, Computational Modeling, and Detection-Theory Strategies for Clinical Instrument Design and Testing
Physical Description: 1 online resource (191 p.)
Language: english
Creator: Kozikowski, Raymond T
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2012

Subjects

Subjects / Keywords: biophotonics -- biosensing -- cancer -- excimer -- fluorescence -- spectroscopy
Biomedical Engineering -- Dissertations, Academic -- UF
Genre: Biomedical Engineering thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Half of all cancers are superficial in nature, originating in the stratified squamous epithelia including cervical, colorectal, skin, and oral cancer. Some of these diseases have seen a dramatic reduction in lethality due to the benefits of improved screening and diagnosis promoting earlier interdiction. Others, while treatable if caught at an early stage, have overall five year survival rates below 50% because they are challenging to identify early. These superficial carcinomas provide an opportunity for technological approaches to early detection due to the nature of their development: malignancy is invariably preceded by dysplastic precancerous cellular changes, which are often confined to the epithelial layer. These earliest changes are often not detectable visually, but are accessible using in vivo spectroscopy. Optical spectroscopies have been widely investigated as techniques for identifying pathological tissue; however, unrelated subject-to-subject variations in spectra complicate interpretation and consequently clinical adaptation has been limited. In this dissertation a new biosensing technique, differential laser-induced perturbation spectroscopy (DLIPS), is described and evaluated. This technique combines fluorescence probing (pre- and post-perturbation) with sub-ablative deep UV perturbation and difference spectroscopy to provide a new spectral dimension, facilitating two improvements over traditional techniques. First, this technique eliminates significant variations in absolute fluorescence within subject populations. Second, UV perturbations alter superficial protein layers, directly coupling the response to the spatio-biomolecular structure. This work is focused on the confluence of this innovation and the unmet clinical need for tools to detect pre-cancerous transformations earlier and with greater accuracy. In a proof-of-concept study, this technique is shown to work at least as well as fluorescence spectroscopy alone in detecting cancer but couples more specifically to changes induced by the dysplastic state, and putative relationships are drawn between tissues spectra and their biomolecular basis. To further understand the benefits of this technique, Monte Carlo modeling of DLIPS tissue spectroscopy is also presented, demonstrating that excimer perturbation enhances the returned molecular signatures from tissue layers and specifically emphasizes important disease biomarkers. Finally in a pre-clinical model of both oral dysplasia and benign pathology, detection performance and optimization of DLIPS and fluorescence technologies are considered from a detection-theory perspective.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Raymond T Kozikowski.
Thesis: Thesis (Ph.D.)--University of Florida, 2012.
Local: Adviser: Sorg, Brian.
Local: Co-adviser: Ormerod, Brandi K.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2012
System ID: UFE0045034:00001


This item has the following downloads:


Full Text

PAGE 1

NON INVASIVE DETECTION OF PRE MALIGNANT LESIONS USING SUB ABLATIVE, DEEP ULTRAVIOLET LASER TISSUE INTERACTIONS COUPLED WITH ENDOGENOUS TISSUE FLUORESCENCE: PROOF OF CONCEPT, COMPUTATIONAL MODEL ING AND DETECTION THEORY STRATEGIES FOR CLINICAL INSTRUMENT DE SIGN AND TESTING By RAYMOND THOMAS KOZIKOWSKI, III 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 2012

PAGE 2

2 2012 Raymond Kozikowski

PAGE 3

3 To my parents and sister

PAGE 4

4 ACKNOWLEDGMENTS I would like to take this opportunity to acknowledge and thanks those people that made this work possible. This undertaking was in no way a one person effort and without the support and help of these people, none of this would have been possible. First is my advisor, Dr. Brian Sorg. Even before coming to Florida, I heard great things about him, from his colleagues at Duke as well as other universities. His rese arch vision, passion for his work and his high level of regard in the research community were the preeminent factors in my decision to attend UF. Second is Dr. David Hahn Without his technical vision and his close collaborative relationship with Dr. Sorg this work never would have materialized. Despite how busy he always was, there was always a little bit extra time he could find to pro vide critical help and guidance through my time here. Additionally a big thank you to my committee members Dr. Brandi O rmerod and Dr. Benjamin Keselowsky, for helping me to understand how to take the raw output of my research efforts and mold it into something more meaningful. Without several key collaborators on this work this project never would have materialized, inclu ding Sarah Smith, Jennifer Lee, Se woon Choe, Dr. William Castleman, and Dr. Indraneel Bhattacharyya Thanks for all of the extra effort, I could not have done it without all of you. Finally, I would like to thank my family and friends: Mom, Dad, Teal, Lau ren, Olive, Joon, Jeff, Craig, Flo, Without their support and patience throughout the years, I never would have made it.

PAGE 5

5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 8 LIST OF FIGURES ................................ ................................ ................................ .......... 9 LIST OF ABBREVIATIONS ................................ ................................ ........................... 12 ABSTRACT ................................ ................................ ................................ ................... 14 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .... 16 Motivation ................................ ................................ ................................ ............... 16 Specific Aims ................................ ................................ ................................ .......... 17 Background ................................ ................................ ................................ ............. 18 Oral Epithelium Anatomy and Physiology ................................ ......................... 19 Progression to Malignancy ................................ ................................ ............... 21 Oral Cancer Detection ................................ ................................ ...................... 23 Tissue Optics Fundamentals ................................ ................................ ............ 26 Excimer Laser Tissue Interactions ................................ ................................ ... 30 DLIPS: State of the Technology ................................ ................................ ....... 32 DLIPS: A Signal to Noise Perspective ................................ ............................. 33 In vivo Optical Diagnostics: Problem Formalization ................................ .......... 35 Experimental Outline ................................ ................................ ............................... 37 2 COMPARATIVE EVALUATION OF DIFFERENTIAL LASER INDUCED PERTURBATION SPECTROSCOPY AS A TECHNIQUE TO DISCRIMINATE EMERGING SKIN PATHOLOGY ................................ ................................ ............ 47 Motivation ................................ ................................ ................................ ............... 47 Materials & Methods ................................ ................................ ............................... 51 DLIPS System ................................ ................................ ................................ .. 51 Che mical Initiation and Promotion of Mouse Skin Tumors ............................... 53 Histopathology ................................ ................................ ................................ .. 53 Collection of DLIPS and Fluorescence Spectra ................................ ................ 54 Multivariate Statistical Analysis ................................ ................................ ........ 55 Detection Theory Analysis of Classification Performance ................................ 56 Results ................................ ................................ ................................ .................... 58 Histopathology ................................ ................................ ................................ .. 58 DLIPS and Laser Induced Fluorescence Spectra ................................ ............ 59 Multivariate Statistical Analysis ................................ ................................ ........ 60 Pathology Detection Performance ................................ ................................ .... 62

PAGE 6

6 Discussion ................................ ................................ ................................ .............. 64 3 DLIPS SPECTRA OF ORAL MUCOSA PATHOLOGY: MONTE CARLO MODELING OF SITE SPECIFIC OPTICAL AND FLUORESCENT PROPERTIES AND THE CHANGES INDUCED BY 193 NM PHOTOCHEMICAL PERTURBATION ................................ ................................ .... 81 Motivation ................................ ................................ ................................ ............... 81 Materials & Methods ................................ ................................ ............................... 85 Simulation Workflow ................................ ................................ ......................... 85 Monte Carlo Model Construction ................................ ................................ ...... 86 Multi Layer Tissue Model ................................ ................................ ................. 87 Fluorescent Interaction s ................................ ................................ ................... 89 Photochemical Interactions ................................ ................................ .............. 90 Pathology Models ................................ ................................ ............................. 92 Simulatio n Details ................................ ................................ ............................. 93 Results ................................ ................................ ................................ .................... 93 Model Validation ................................ ................................ ............................... 93 Pre Perturbation Flu orescence ................................ ................................ ......... 93 Excimer Laser Perturbation Patterns ................................ ................................ 94 Post Perturbation Fluorescence ................................ ................................ ....... 95 Computed Spectral Remittance and Layer Specific Signals ............................ 95 Discussion ................................ ................................ ................................ .............. 98 4 DETECTION THEORY ANALYSIS OF DL IPS BASED SENSING OF ORAL DYSPLASIA: IMPLICATIONS FOR CLINICAL INSTRUMENT DESIGN AND TESTING ................................ ................................ ................................ .............. 117 Motivation ................................ ................................ ................................ ............. 117 Materials & Meth ods ................................ ................................ ............................. 119 Chemically Induced Cheek Pouch Pathology ................................ ................. 119 Collection of DLIPS and Fluorescence Spectra ................................ .............. 120 Histopathology ................................ ................................ ................................ 122 Dimensional Reduction and Subspace Analysis ................................ ............ 122 Classification Framewor k and Detection Performance ................................ ... 125 Results ................................ ................................ ................................ .................. 126 Histopathology ................................ ................................ ................................ 126 S pectral Analysis ................................ ................................ ............................ 127 Subspace Analysis ................................ ................................ ......................... 128 Measurement Classification ................................ ................................ ........... 131 Discussion ................................ ................................ ................................ ............ 132 5 CONCLUSION ................................ ................................ ................................ ...... 151 APPENDIX MATLAB CODE FOR MONTE CARLO SIMULATIONS ...................... 156 LIST OF REFERENCES ................................ ................................ ............................. 181

PAGE 7

7 BIOGRAPHICAL SKETCH ................................ ................................ .......................... 191

PAGE 8

8 LIST OF TABLES Table page 1 1 Common endogenous tissue fluorophores ................................ ......................... 43 2 1 Detection performance metrics using DLIPS, laser induced fluorescence, or a decision fusion approach. ................................ ................................ ................ 77

PAGE 9

9 LIST OF FIGURES Figure page 1 1 Commercially available systems for augmenting oral cancer detection .............. 40 1 2 Layers of epithelial tissue in the two general types of oral mucosa. A) keratinized ................................ ................................ ................................ .......... 41 1 3 Pictures and histology from patients presenting with different types of oral pathology ................................ ................................ ................................ ............ 42 1 4 Laser tissue interaction possible as a function of exposure time and irradiance ................................ ................................ ................................ ............ 44 1 5 Cleavage of peptide bonds due to direct ph otolysis induced by absorbance of 193 nm photons from an excimer laser source ................................ ................... 45 1 6 Signal propagation and transformation map for non invasive optical diagnosis of disease ................................ ................................ ................................ ........... 46 2 1 DLIPS acquisition scheme ................................ ................................ .................. 70 2 2 Schematic of the DLIPS system ................................ ................................ ......... 71 2 3 H&E sta ined histology sections of skin ................................ ............................... 72 2 4 Average spect ra at weeks two (n=6, 7), six (n=5, 4), and eight (n=7, 8) ............ 73 2 5 S pectra at weeks two, six, and eight ................................ ................................ .. 74 2 6 Score plots for the principal component analysis ................................ ................ 75 2 7 Proportion of variance in Principal Component 1 accounted for by each wavelength ................................ ................................ ................................ ......... 76 2 8 ROC curves for the detection of DMBA induced skin pathology using the DLIPS technique ................................ ................................ ................................ 78 2 9 ROC curves for the detection of DMBA induced skin pathology using laser induced fluorescence ................................ ................................ .......................... 79 2 10 ROC curves for the fused decision inco rporating both fluorescence and DLIPS spectra to boost the optimal performance in detecting DMBA induced ............................ 80 3 1 Simulation workflo w for modeling DLIPS fluorescence tissue spectroscopy .... 105 3 2 Tissue optical properties for layer specific modeling ................................ ........ 106

PAGE 10

10 3 3 Mode led layer specific optical properties ................................ .......................... 107 3 4 Bland Altman plots comparing the total remitted light from the experimental Monte Carlo model versus a previously validated Monte Carlo system ............ 108 3 5 Fluence (cm 2 ), in logarithmic units of photons through the tissue models ....... 109 3 6 Relative per layer absorption and fluores cence emission ................................ 110 3 7 Fluence (cm 2 ) of 193 nm light (logarithmic scale) through the superficial epithelium ................................ ................................ ................................ ......... 111 3 8 Photochemic al events stimulated by 19 3 nm irradiation as a fraction of the total number of incident photons ................................ ................................ ....... 112 3 9 Relative contributions to the remitted DLIPS signal from each layer of epithelial tissu e ................................ ................................ ................................ 113 3 10 Relative contributions to the remitted fluorescence signal from each layer of epithelial tissue ................................ ................................ ................................ 114 3 11 Simulated fl uorescence spectra for the three tissue models ............................ 115 3 12 Binned spectral DLIPS data from Monte Carlo simulations .............................. 116 4 1 Fiber pr obe used for DLIPS and fluorescence measurements in the hamster cheek pouch ................................ ................................ ................................ ..... 136 4 2 Pictures of everted cheek pouches displaying the model pathologies used in this study ................................ ................................ ................................ .......... 137 4 3 H&E stained histology sections of cheek pouches ................................ ........... 138 4 4 Average fluorescence spectra from 10 SLS treated benign irritation cheek pouches (blue ), 10 carrier only treated control cheek pouches after 8 or 9 weeks of application (green), and 10 DMBA treated dysplastic cheek pouches after 8 or 9 weeks of application (red) ................................ ................ 139 4 5 Average DLIPS spectra from 10 SLS treated benign irritation cheek pouches (blue), 10 carrier only treated control cheek pouches after 8 or 9 weeks of application (green), and 10 DMBA treated dysplastic cheek pouches after 8 or 9 weeks of application (red) ................................ ................................ .......... 140 4 6 Percent of variance explained by each of the first ten principal components, p values for the ANOVA between the three experimental groups (in each principal component), and post hoc analysis to determine which group means are significantly different for the laser induced fluorescence data ........ 141

PAGE 11

11 4 7 Percent of variance explained by each of the first ten principal components, p values for the ANOVA between the three experimental groups (in each principal component), and post hoc analysis to determine which group means are significantly different for the DLIPS data ................................ ......... 142 4 8 Data groups used to derive the six FLD transformations as well as the corresponding SNR between the H 0 and H 1 data group pre and post transformation ................................ ................................ ................................ ... 143 4 9 ANOVA and Tukey HSD post hoc analy sis of the transformed fluorescence measurements ................................ ................................ ................................ .. 144 4 10 ANOVA and Tukey HSD post hoc analysis of the transformed DLIPS measurements ................................ ................................ ................................ .. 145 4 11 Confusion matrix for the fluorescence data under transformation T1 ............... 146 4 12 Confusion matrix for the fluorescence data under transformation T2 ............... 147 4 13 Confusion matrix for the fluorescence data under transformation T4 ............... 148 4 14 Confusion matrix for the DLIPS data under transformation T4 ......................... 149 4 15 Confusion matrix for the DLIPS data under transformation T2 ......................... 150

PAGE 12

12 LIST OF ABBREVIATION S ArF Argon Fluorine AUC Area under the curve DLIPS Differential Laser Induce d Perturbation Spectroscopy DMBA 7, 12 Dimethylbenz(a)anthracene DNA Deoxyribonucleic acid ECM Extracellular Matrix FAD Flavin adenine dinucleotide FDA Food and Drug Administration FPR False positive rate H&E Hematoxylin and eosin HPV Human Papilloma Virus ICCD Intensified Charge Coupled Device IEEE Institute of Electrical and Electronics Engineers IHC Immunohistochemical MAP Maximum a Posteriori MCC Matthews Correlation Coefficient NADH Nicotinamide adenine dinucleotide Nd:YAG Neodymium doped yttrium alumi num garnet NIR Near Infrared PCA Principle Component Analysis ROC Receiver Operating Characteristic RF Relative layer fluorescence RTE Radiative Transport Equation SCC Squamous cell carcinoma

PAGE 13

13 SLS Sodium Lauryl Sulfate SNR Signal to Noise Ratio TNM Tumor, N odes, Metastasis TOST Two one sided tests TPR True positive rate UV Ultraviolet UVA Ultraviolet A

PAGE 14

14 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Docto r of Philosophy NON INVASIVE DETECTION O F PRE MALIGNANT LESIONS US ING SUB ABLATIVE, DEEP ULTRA VIOLET LASER TISSUE INTERACTIONS COUPLED WITH ENDOGENOUS TISSUE FL UORESCENCE: PROOF OF CONCEPT, COMPUTATIONAL MODELI NG, AND DETECTION THEORY STRATEGIES FO R CLINI CAL INSTRUMENT DESIG N AND TESTING By Raymond Thomas Kozikowski, III December 2012 Chair: Brian Sorg Cochair: Brandi Ormerod Major: Biomedical Engineering Half of all cancers are superficial in nature, originating in the stratified squamous epithelia inc luding cervical, colorectal, skin, and oral cancer. Some of these diseases have seen a dramatic reduction in lethality due to the benefits of improved screening and diagnosis promoting earlier interdiction. Others, while treatable if caught at an early sta ge, have overall five year survival rates below 50% because they are challenging to identify early. These superficial carcinomas provide an opportunity for technological approaches to early detection due to the nature of their development: malignancy is in variably preceded by dysplastic precancerous cellular changes, which are often confined to the epithelial layer. These earliest changes are often not detectable visually, but are accessible using in vivo spectroscopy. Optical spectroscopies have been widel y investigated as techniques for identifying pathological tissue; however, unrelated subject to subject variations in spectra complicate interpretation and consequently clinical adaptation has been limited. In this dissertation a new biosensing technique, differential laser induced perturbation

PAGE 15

15 spectroscopy (DLIPS), is described and evaluated. This technique combines fluorescence probing (pre and post perturbation) with sub ablative deep UV perturbation and difference spectroscopy to provide a new spectral dimension, facilitating two improvements over traditional techniques. First, this technique eliminates significant variations in absolute fluorescence within subject populations. Second, UV perturbations alter superficial protein layers, directly coupling the response to the spatio biomolecular structure. This work is focused on the confluence of this innovation and the unmet clinical need for tools to detect pre cancerous transformations earlier and with greater accuracy. In a proof of concept study, this technique is shown to work at least as well as fluorescence spectroscopy alone in detecting cancer but couples more specifically to changes induced by the dysplastic state, and putative relationships are drawn between tissues spectra and their biomolecula r basis. To further understand the benefits of this technique, Monte Carlo modeling of DLIPS tissue spectroscopy is also presented, demonstrating that excimer perturbation enhances the returned molecular signatures from tissue layers and specifically empha sizes important disease biomarkers. Finally in a pre clinical model of both oral dysplasia and benign pathology, detection performance and optimization of DLIPS and fluorescence technologies are considered from a detection theory perspective.

PAGE 16

16 CHAPTER 1 IN TRODUCTION Motivation More than half of all cancers are superficial in nature, originating in the stratified squamous epithelia, including cervical, colorectal, skin, and oral. 1 Some of these diseases have seen a dramatic reduction in lethality or number of cases, due in a large part to the benefits of improved screening and earlier diagnosis promoting earlier interdiction. 2 To continue to succeed, primary c aregivers need new instruments and methods to continue pushing forward the time lines for cancer detection and diagnosis. Of particular utility are strategies that reduce the need for highly trained personnel and healthcare infrastructure to support screen ing and diagnosis, instead offering sample to answer capabilities at the point of care. As an example, oral cancer is the sixth most common cancer worldwide and in the year 2007 accounted for over 400,000 new cases. 3 The lethal nature of oral cancer is not related to an inability to treat the disease, but rather the fact that it is of ten not discovered until the cancer has metastasized. The 5 year survival rate has hovered around 50% for 50 years, but if oral cancers are diagnosed at stage I, the likelihood of survival after treatment improves to nearly 90%. 3 5 When caught early, oral cancers are often smaller and treatable with a far lower incidence o f morbidity. 4, 5 In some countries, oral cancers account for up to 40% of all cancers, further exacerbating the lethality. 6 Although this dramatically larger fraction is primarily a consequence of differential epi demiological risk factors, the fact remains that earlier detection and diagnosis are essential to driving down the death rate due to oral cancer.

PAGE 17

17 Despite the presence of advanced diagnostic aid s available to clinicians (Figure 1 1), there is no evidence to support the ability of these devices to detect dysplastic changes at any earlier stage than what a trained clinician can detect using an unaided visual and tactile exam. Because of the specifics of these systems, detection relies upon discovering changes that are characteristic of dysplasia, but not necessarily unique to the malignant or pre malignant state. As a result, these devices only serve to confirm what the physician already knows, adding limited new information regarding diagnosis or prognosis. Fu rther, these techniques exacerbate the burden of disease by erroneously detecting false cancerous transformations; leading to unnecessary biopsies, interventions, and stress for the patient. Specific Aims This dissertation describes the use of a new fluore scence spectroscopy technique, differential laser induced perturbation spectroscopy (DLIPS), as a potential technique for the detection and staging of epithelium derived dysplasia, like oral cancer. Here using both in vivo techniques and modeling it is dem onstrated that this technology realizes a powerful new spectral dimension that is strongly coupled to the underlying biology and was previously inaccessible using other optical techniques. Presented in this dissertation are results from a proof of concept animal study, Monte Carlo simulations, and a preclinical study of oral dysplasia that all support future work translating this technique into a clinical instrument for the enhanced detection of cancer. The long term vision for this work is to develop DLI PS technology into a clinical system that impacts cancer morbidity and mortality by facilitating earlier, non invasive detection and diagnosis of pre malignancies. The primary objectives of this work were to develop and test the first DLIPS tissue spectros copy system, continue to develop the

PAGE 18

18 understanding of the molecular mechanisms underlying this technique, and consider how to integrate this enhanced tissue information into a clinical detection scheme. To accomplish these goals, this work was divided into the following specific aims: Aim 1 Show proof of concept for using DLIPS to detect dysplasia against healthy skin in a murine model of emerging malignancy Aim 2 Use Monte Carlo simulations of tissue fluorescence and photochemical interactions to develop a biochemical and pathophysiological basis for the DLIPS spectral signature Aim 3 Investigate DLIPS in the hamster cheek pouch model for the detection of oral dysplasia, in contrast to both healthy buccal tissue and benign pathology Use these results to motivate strategies for clinical instrument design and testing In total, the results of these three invest igations show that the detection of dysplasia is possible using the DLIPS techniques, even in the presence of confounding benign pathology. These in vivo results are further supported by an exploration of the molecular signatures associated with the DLIPS signal. Further, by considering DLIPS both as a stand alone and adjuvant detection technique, the flexibility and interoperability of this technique is demonstrated. These results provide strong evidence for developing this technique further for detection of oral dysplasia in clinical studies as well as transferring the DLIPS platform to other pathologies that would benefit from improved non invasive diagnosis. This technology will have potential positive impacts by facilitating earlier cancer diagnosis (reducing mortality), showing proof of concept for the development of DLIPS sensing as a platform detection technology for other cancers and pathologie s, and bridging the technology gap between in vitro molecular diagnostic techniques and in vivo optical and fluorescence spectroscopy. Background To understand both the problem of detecting the pre cancerous state as well as the opportunities for technolog ical innovations that support enhanced cancer diagnosis,

PAGE 19

19 there are several areas of knowledge that must be considered foundational to understanding the biology of cancer detection. First are the fundamentals of epithelial physiology and the changes associa ted with benign and pre malignant pathology of the oral mucosa. Building off of these topics a treatment of current available diagnostic aids for enhanced oral cancer detection is discussed to round out the definition of the opportunity available for innov ating in oral cancer diagnostics. To support the development of biophotonics based strategies in approaching solutions to this challenging diagnostic setting, a review of tissue optics concepts and current approaches is covered. From there, support is buil t for the scientific basis of the DLIPS technique by discussing the laser tissue interactions and the photochemistry induced by excimer lasers. Additionally, the details of the DLIPS system are discussed from a signal to noise perspective, highlighting the advantages and susceptibilities associated with the DLIPS technique. Finally, the problem of optical cancer diagnostics is considered from a signal processing perspective, critical for properly framing the problem, understanding the needs and opportunitie s for innovation, and identifying rational approaches to solving these problems. A summary of these topics is presented here to motivate and support the experimental work discussed in the subsequent chapters of this dissertation. Oral Epithelium Anatomy an d Physiology The structure of epithelial tissues is roughly conserved throughout the body, consisting of thin, superficial, avascular layers of epithelial cells, supported by a basement membrane. Below this is a thicker stromal layer that supports the supe rficial cellular layers structurally with a dense collagen fiber matrix and nutritionally as it is vascularized. This deeper, thinker stromal layer attaches the epithelium to underlying

PAGE 20

20 tissues. This structural archetype extends across skin, colon, cervix, and the oral cavity. 7 However, these tissues can vary greatly in the composition and function of the cells, altering the presentation of these tissues dramatically at both the macr oscopic and molecular levels. Even in the oral cavity there are at least two structurally unique epithelial tissues types, mastica tory mucosa and lining mucosa (F igure 1 2 A B ). The superficial layer of stratified squamous epithelium in the lining mucosa i s the most apparently differentiator from the masticatory mucosa because it is non keratinized. Despite not having a stiff, tough top layer of keratin, these superficial cells keratinized misnomer) as they push towards the surface. However these cells retain nuclei and organelles as they flatten and join this protein cross linked network. 7 These non keratinize d tissues are softer and more flexible than there keratinized counterparts and most closely resemble the epithelial lining of the cervix. 7 In the intermediate and basal epithelial l ayers, cells grow and differentiate in as they push up from the basal layer into spinous intermediate layer. Here the cells form a coherent network of cells attached by spiny cellular processes and desmosomes. 8 Ultimately, these cells will flatten and increase their intracellular keratin density as they are pushed towards the superficial layer by cell from the basal layer. Between the basal epithelium and the lamina propria, the b asement membrane is a flat interface, unlike in the masticatory region. 7 This supporting region of the oral mucosa is composed primarily of fibroblasts, collagen (20 25% by volume) and capillary networks (0.25% blood by volume). 7, 9 The extracellular matrix of this region is primarily collagen type I, with significant fractions of type III fibers

PAGE 21

21 and elastin. 10 Lining mucosa covers oral regions including the cheek and the floor of the mouth. Most noticeably, the masticatory region is covered by a tough keratinized, cornified layer that provides protection from mechanical injury, similar to the role of the keratinized surface of the epidermis. 7 These cornified cells are flat and hexagonal in shape, packed with keratin, and bound by a lipi d matrix derived from interactions as cell fuse into this layer from the intermediate region. 7 Unlike the flat basement membrane junction of the non keratinized epithelium, in the m asticatory region there are ridges and connective tissue papillae into the epithelial layers; adding extra mechanical stability to these regions. 7 In between, the basal layers and s pinous layers are the primary layer of cellular differentiation from basal cells into cornified superficial cells. This variety of tissue structure can make diagnosing disease challenging as the gross morphological changes between regions and those induced by pathology may be hard to deconvolve, there is thus a need for technologies to address subsurface sensing approaches to the in vivo detection and discrimination of pathology at the point of care. Progression to Malignancy Still another layer of ambiguit y exists in detecting carcinomas, cancers derived from the epithelium, or more importantly those earlier changes that will lead to a malignancy. This is because in back tracing the progression to a carcinoma, the pattern of behaviors leading to the emergen ce of a malignant neoplasm are a mix of prognostic pre neoplastic identifiers as well as many other morphological change highly non specific to the cancerous transformation. In the case of oral cancer, definite visual identification is often not possible u ntil a tumor actually begins to form at the latest stages of localized disease: the malignant neoplasm. 11 Although diagnosis at this stage

PAGE 22

22 is unambiguous, the efficacy of intervention is limited because the cancer has disrupted and penetrated the basement membrane of the epithelium, providing access to the ly mphatic and circulatory systems for the cancer to metastasize systemically. In the epithelium of the oral mucosa, this stage of disease is preceded by a period of disease in which the neoplastic cells are confined locally above the basement membrane, carci noma in situ (TNM stage 0). 11 At this point the diseas e presents as an incipient cancer that if left untreated will definitely become a malignant neoplasm. The abnormal cells contained in this area have progressed into a stage of neoplastic growth; unchecked proliferation and an ambiguous state of cellular di fferentiation, among other abnormal cellular features. 1 1 This definite pathological state is brought about only following a much longer pre interventional state of dysplasia, where both the presentation of disease (visually and histopathologically) as well as the fate of the lesion are unclear. Oral dysplasia is defined clinically using a cohort of abnormal cellular features including: 12 loss of basal cell polarity loss of specific functions, shape or structures uni que to the apical or basal faces of basal cells parabasilar hyperplasia abnormally large numbers of the smallest oral epithelial cells increased n u clear:cytoplasmic ratio abnormally enlarged nuclei drop shaped rete ridges abnormally large regions of epi thelial thickening extending down into the stroma abnormal epithelial maturation highly differentiated cells observed in basal layers and non differentiated cells observed in the superficial layers increased mitotic activity mitosis in the superficial ep ithelium cellular pleomorphism abnormally shaped cells

PAGE 23

23 nuclear hyperchromaticity abnormal increase in the density of nuclear material in the nucleus enlarged nucleoli loss of cellular cohesiveness decrease in the organization and packing of the cells w ithin and between epithelial layers cellular keratinization in the spinous epithelial layer abnormal keratin density in the cytoplasms of non superficial epithelial cells It is at this point where an ability to identify and assess the malignant potentia l of lesions would add a clear benefit to patients, improving 5 year survival rates to over 90%. 4, 5 Diagnosis prior to this point is likely not possible given the current understanding of these earlier stag es of neoplasm development as well as the development of benign pathologies. Still there is opportunity in continuing to identify pre malignant conditions, morphologies with an associated risk of malignant tumor formation (eg: chronic oral ulceration), as a means of screening the population for high risk patients, but that is not within the scope of this work. With the current state of disease understanding and non invasive technologies, one is limited to attempting to evaluate and discover pre malignant le sions, those morphologies containing dysplastic cells that will definitely progress to malignancy (eg: cervical epithelial dysplasia, adenomatous polyps). 13 15 Fundamentally, a new detection technique for oral cancer must not just detect dysplasia, but must be able to further determine if that lesion has malignant potential or will simply persistent as a benign lesion. Oral Cancer Detection Oral cancer provides an excellent example of both a problematic diagnosis as well as an opportunity in that early diagnosis and intervention can have a dram atic impact on survival. 3 5 Because of the high value associated wit h improved diagnosis of oral

PAGE 24

24 cancer, there are multiple strategies both in the laboratory and in the marketplace to address this need, but with uniformly disappointing results. In review articles published in 2007, 2008 and 2010, the authors all reached si milar conclusions regarding these systems: there is insufficient evidence to support claims of improved oral cancer diagnosis over conventional oral examination. 3, 13, 15 Principally, the problem is an inability to sort dysplastic from benign disease. This is because the preclinical phase of the dys plastic transformation into oral squamous cell carcinoma is varied, often asymptomatic, and easily misdiagnosed as benign inflammation, hyperkeratosis, or ulceration. 16 Conversely, several benign oral abnormalities are often mistaken for oral cancer including morsicatio buccarum, melanotic macules, or leukodema. 16 And these benign diseases are in fact more prevalent than oral cancer, as up to 15% of the population presents with oral mucosal disease (non malignant or pre malig nant) at any given time. 17, 18 Further complicating the issue, dysplastic changes may be associated with some other oral pathologies (proliferative verrucous leukoplakia, lichen planus), poor oral hygiene (leading to chronic inflammation), or c hronic mucosal infections (HPV and herpes simplex). 19, 20 These co morbidities mask the slow pre malignant transformations from both the dental prof essional and currently available diagnostic aids. As a result, most dental professionals choose to follow suspicious lesions over time, watching for clear signs of the malignant transformation before opting for biopsy to confirm diagnosis or begin treatmen t Figure 1 3 illustrates this, showing cases investigated by Dr. Indraneel Bhattacharyya (Department of Oral & Maxillofacial Diagnostic Sciences, University of Florida, Gainesville, FL), where visual inspection alone would miss dysplastic changes,

PAGE 25

25 another where a benign lesion resembles an oral pre malignant lesion (but histology confirms the lack of cancer), and a final example where a suspicious lesion is confirmed as cancerous after histopathological analysis. Figure 1 1 covers the most pervasive detect ion technologies for oral cancer and compares them to unaided examination. A common sentiment within the scientific and medical community is there is a lack of rigorous clinical testing of these devices, with manufacturers instead relying on studies of hig h risk (or previously diagnosed) populations where the incidence of cancer was often above 20% and in some cases as high as 100% or anecdotal evidence from individual case studies. 3, 13 This introduces a large spectrum bias between the diagnostic performance of these devices on the tested (virtual certainty in presence of cancer) and target population (uncertainty in clinical presentatio n). Further, many of the studies either point out directly or it has been noted by others, that the investigational devices were not able to perform better than unaided visual/tactile examination by a physician; likely due to the obvious presentation of d ysplastic transformation or outright squamous cell carcinoma. 3, 13 Compounding this issue is that many of the studies have a limited d ata set to compare against the gold standard, biopsy. This is not a shortcoming of the studies but rather due to the fact that it would be unethical to biopsy tissue that did not present with clinical indications warranting such an invasive procedure. Howe ver this inadvertently screens out many potential false negatives from the data collected on the performance of these devices. Therefore, the diagnostic performance numbe rs reported in the literature ( Figure 1 1 ) should be understood to represent the best c ase performance metrics and ultimately do not reflect how well this devices perform at the point of care. Despite that, all of these

PAGE 26

26 systems are highly susceptible to reporting false positives; detecting oral cancer when in fact there is no dysplasia prese nt. The explanation is a complex mix of the following factors: The high cost associated with missing dysplastic lesions (false negatives) leads to an unacceptable amount of false positives The sources of tissue contrast employed in these devices is not un ique to the molecular changes of the oral dysplastic transformation (erroneous detection of benign conditions) High degree of intra patient variation in tissue properties (source of noise in non referential detection schemes) This low specificity, regardl ess of the tradeoff for improved sensitivity to finding pre malignant lesions or cancer, is troubling for a number of reasons. Tests with low specificity have little diagnostic merit due to the high prevalence of disease finding when none is present. This nullifies the utility of the test as a positive result will always require a confirming follow up test. In the case of oral cancer, this gold standard for follow up would be biopsy. However, when these techniques are tested in independent clinical trials o r by practitioners, there is little evidence to support the claims of improved sensitivity; the devices do not show improved ability to detect dysplasia and additionally report unacceptably high numbers of false positives. 3, 13 Tissue Optics Fundamentals In this work, well established relationships between changes in tissue structure and disease state are exploited with a novel light bas ed sensing scheme to enhance discrimination of dysplastic tissue in vivo. The field of tissue optics is concerned with a fundamental understanding of these associations, and these principles form the basis for the application of light based strategies to d isease detection and diagnosis. As a gross over generalization, when a photon interacts with matter it is either absorbed by

PAGE 27

27 that material or it is not. 21 When the light is not absorbed it will exit the material, with its trajectory potentially altered by sca ttering. On the other hand, once the light is absorbed, there are multiple possible fates. To connect this basic idea with biological tissue, tissue optical analysis is concerned with two basic ideas. First is an understanding of the light energy per time interacting with a chromophore at a fixed position in the tissue, known as the fluence rate. 21 Unfortunately, this information is not directly observable; however by modeling the optical properties of tissues and incorporating information about the radiance a nd remitted/transmitted light, it can be approximated. Second, and more relevant to the task of applying biomedical optics to disease diagnosis, is to measure the optical properties of tissue. 21 This is essentially the inverse problem of the first task, in th e sense that with an accurate picture of the fluence in tissue and incident radiance, the optical properties can be inferred. Although the absorption and scattering properties of tissue are a direct result of the molecular structures and the chemical envir onment surrounding them, in tissue regions are often assumed to have bulk scattering and absorption properties, described by a scattering coefficient, s where the inverse of this parameter describes the average distance between scattering events, and an absorption coefficient, a where the inverse of this parameter describes the average distance a photon travels between absorption events. 21 Empirically, these values represent the superimposed effects of all the constituent chromophores in the tissue, but re alistically calculations are often not possible and these bulk optical properties must be measured or estimated using less direct approaches. 22 Absorption in tissue around the visible light range is highly

PAGE 28

28 wavelength dependent and driven by several key diagnostic chromophores with high extinction coefficients including hemoglobin, myoglobin, and melanin. 23 Scattering in in biological tissues is equally important for tissue optical diagnostics, as scatte ring effects tend to dominate absorption in the UVA, visible, and near infrared portions of the optical spectrum. 21 This will determine the volume of interaction for incident light as well as how well a beam is able to maintain its original collimated structu re. 2 1 Elastic scattering, the redirecting of light without a loss of energy, is the most common effect observed in tissue and thus inelastic scattering will not be considered here. Sources of scattering arise at three size levels, macroscopic from muscle fibe rs and tissue layers, microscopic from cells and organelles, and submicroscopic from macromolecules and their structures (protein fibers and lipid membranes). 24 In tissue, the scattering of light is often approximated by combining the effects from a Rayleigh description of scattering as well as Mie Theory. The combination is harmonious as each describes scattering events at a different scale, for structures much smaller than a wavelength of light and for structures on the scale of or larger than a wavelength respectively. Around the visible portion of the spectrum, Mie Theory tends to contribute substantially, including d ictating the angle through which the incident light is scattered. This is defined as the anisotropy factor, g. Using the Henyey Greenstein function, this parameter can be used to approximately describe the probability of light be scatter off of a particle at a particular angle. 24 Tissue, as a generally rule, is highly forward scattering, with anisotropy factors in the ra nge of 0.65 to 0.95. 23 When light is absorbed that energy must be dissipated as the excited absorber returns to its ground state from its excited molecular state. Typically, this energy is

PAGE 29

29 dissipated through collision quenching or converted in heat, but for certain molecular systems, more complex photochemical interactions are possible. The most well known photochemical reaction is that of fluorescence generation. This occurs when the excited electron moves from one energy level to a lower energy level within the excited state. Upon relaxing back to the ground state, the emitted photon is at a low er energy (longer wavelength) than the excitation photon. 21 Although this is not the only photochemical process possible from the excited state, it is the most common in biomolecules and the most germane to this particular optical diagnostic technique. Genera lly, fluorescence is a nondestructive and sensitive technique that allows in vivo detection of organic and biological materials under different environmental conditions in very short times. Because the excitation/emission wavelength pair is a combination of the molecular structure and the overall molecular environment (e.g. biological matrix), fluorescence emission may provide discrimination among emitting materials. Tissue autofluorescence is primarily derived from several major endogenous ti ssue fluoroph ores are shown in T able 1 1 with their most prominent excitation and emission ranges listed as well 25 It has been shown that many of these molecules are correlated with certain pathological conditions and have unique distributions in tumors, and that these fluorescent molecules can be used to distinguish benign from malignant tumors. 26, 27 Light propagation in tissue can be co nsidered analytically with the R adiative T ra nsport E quation (RTE) shown in E quation 1 1. 22 (1 1) This equation relates both the goals of tissue optical diagnostics (understanding light propagation in the context of the optical properties of tissue) into a single unifying

PAGE 30

30 relationship that is driven by a source term and the losses of light energy from that incident beam in terms of divergence and extinction, considers the light incident on and propagating from infinitesimally small spherical region of the medium, and describes the absorption and scattering properties of the medium. 22 However, there is often no closed form solution to this equation without introducing approximations such as the diffusion approximation and these approximations only apply un der highly contrived situations. 21 A more flexible method for estimating the fluence and remission of light from tissue are Monte Carlo based numerical simulations. These methods have the added benefit of permitting extension of the model to other photo effec ts including fluorescence and photochemical reactions. 9, 28 30 Although still emerging as a platform for disease diagnostics, the systematic study of light tissues interactions both in vivo and in vitro has yielded a wealth of information regarding the optical properties of biological samples at nearly all scales: individual molecules, macromolecular structures, cells and organelles, as well as whole tissues and entire organs. As an example, a summary of research into the optical properties of the oral m ucosa is given in Figure 3 2 to drive model development for studying the DLIPS system. This has allowed for refinement of optical disease detection strategies, designed to probe these properties and the changes associated with disease, into clinical instru detect and diagnose disease. Excimer Laser Tissue Interactions Generally, tissue optical properties for diagnostic applications are only considered in the range of visible and near in frared frequencies that have relatively low absorbance by water, nucleic acids, and most proteins (aside from those with transition metal

PAGE 31

31 complexes such as hemoglobin). However, for the application presented here, a discussion of the light tissue interacti ons for coherent deep ultraviolet light is necessary as well. For the wavelength range of ArF excimer lasers there is considerable absorption of light in tissue by peptide bonds, aromatic compounds, and nucleic acids. 31, 32 Although there are few studies measuring the optical properties of tissue for the deep UV range estimates range from 1000 c m 1 up to 20,000 cm 1 with these properties varying greatly depending upon the biomolecular target tissue composition. 24, 31, 32 While there are no studies exploring the scattering properties of light in tissue in the deep UV range, it is expect ed that scattering should increase by a factor of 16 from that observed at 400 nm because of the dominance of Rayleigh scattering at shorter wavelengths. 24 It is further assumed that the scattering of light will be fairly isotropic as well given the heavy weighting towards Rayleigh scattering. It is also known that excitation of some amino acids at 193 nm can induce fluo rescence, but this effect is minimal when compared to the range of other energy emissive effects possible at this wavelength. Figure 1 4 summarizes the possible effects of tissue exposure to laser irradiation, as a function of both pulse duration and inten sity. The perturbation pulses from the deep UV excimer laser (193 nm, 6.4 eV) are strongly absorbed by biological tissue and used to cleave molecular bonds within the extracellular matrix (ECM) as shown schematically in Figure 1 5 Irradiation of biologic al matrices at 193 nm can cause photoionization, including strand breakage, locally denatured sites, interstrand cross linking, reactions via photo dimers, and other products. 33 The primary absorbers of 193 nm light in vivo are peptide bonds and aromatic amino acid side chains with respective extinction coefficients around 5.5 x 10 3 M 1 cm 1 and 2 x 10 4 5 x

PAGE 32

32 10 4 M 1 cm 1 and are roughly found at concentrations of 1 M and 0.1 M in the cytosol of epithelial cells. 32 Although nucleic acid s also have high absorbance cross sections for 193 nm light, it has been shown that due to the high concentration of peptide bonds in the cytosol, limited photochemical reactions are realized on the nuclear material contained within the nucleus. 31 While no direct ablation is realized for the low energy densities used in this application, around 2 3 mJ/cm 2 (see Figure 1 4 ), a single photon of 193 nm radiation exceed nearly all bo nd energies (around 4 eV for peptide bonds and aromatic ring systems) in the biological matrix and for cytosolic proteins, hence permanent photochemistry is induced despite being below the critical photon flux to affect material removal. In some studies th e quantum yields for the scission of peptide sc has been found to vary from about 2% to 3% while the quantum yield for the d, was reported to be in the range of 10% to 12%. 34, 35 DLIPS: State of the Technology The DLIPS sensing scheme incorporates two complementary techniques to improve upon previous biosensing strategies: ult raviolet laser perturbation of tissue, and difference spectroscopy (F igure 1 5 ) 36 This technique is flexible in that it can be applied to multiple spectral domains, creating a means to extending existing eq uipment and analytical techniques into a new domain of biochemical analysis. These domains could include fluorescence spectroscopy, Raman spectroscopy, or even non optical techniques such as nuclear magnetic resonance. In this work, fluorescence is used to measure endogenous tissue fluorescence before and after the tissue is laser perturbed. Despite the excimer laser intensity being well below the threshold for tissue ablation, permanent alteration of the underlying tissue structure is induced, with resulti ng

PAGE 33

33 changes within the fluorescence spectrum, specifically with respect to photoreactive biomolecules. The proposed cancer detection technology is rooted in previous research demonstrating that the biological matrix (collagen and amino acids) may be altered by low intensity (i.e. subablative) coherent UV light such that the resulting fluorescence properties are perturbed. 36, 37 Initial proof of concept was shown using dye fluorophores, collagen, and bovine corneas. In testing th is technique using peptides, a photochemical efficiency of around one peptide bond cleavage event per 50 incident photons was reported. 36, 37 These initial tests with excised tissue confirmed the proof of concept for coupling this technique with both fluorescence and Raman spectroscopy. In these experiments, the excimer laser was used at a fluence more than 20 times below the fluence used for refrac tive surgery (marked on Figure 1 4 ). Further, using this technique in collagen f ilms, the fluorescence signal was found to actually be enhanced by photochemical perturbation, suggesting some mechanism for releasing the fluorescence crosslinking compounds (hydroxylysyl pyridinoline and lysyl pyridinoline) from the collagen matrix as a result of perturbation. 10, 25 Details of computing the DLIPS spectru m as well as the system used for subsequent in viv o applications can be found in C hapters T hree and F ive of this document. 38 DLIPS: A Signal to Noise Perspectiv e Approaching the DLIPS technique and fluorescence spectroscopy from a stochastic signals in noise perspective obviates several important considerations for actual applications of this technique. Equations 1 2 through 1 8 in the scheme below describe a sin gle observation of a signal (DLIPS or fluorescence), in the presence of additive white Gaussian noise, here assumed to be due to the measurement technique alone, and not capturing signal variability between targets. In estimating the signal to

PAGE 34

34 noise ratio (SNR) for each technique, the SNR is defined as the average observed signal, divided by the standard deviation for multiple observations. Since the DLIPS measurement is a differential measurement, the signal strength scales to the size of that difference, in this case to between 10% and 1% of the original fluorescence signal. This large difference in signal magnitude between DLIP S and fluorescence is shown in E quation 1 8. The white noise from the background does not scale and in fact the variance is margin ally amplified during the computation of the DLIPS signal from the par ent fluorescence measurements (E quations 1 2 and 1 3). The very best case scenario then is that the noise is only amplified by a factor of root two, but is likely worse for low signal le vels as the contribution from the denominator must be factored into the equation as well and can no longer be approximated as having no contribution. This implies that the very best case scenario is that the DLIPS SNR is approximately 14 times lower than m easurements made using fluorescence alone. The reality however is that the measurement SNR is possible more than 140 times lower when low signal fluorescence measurements are made. (1 2) (1 3 ) (1 4) (1 5)

PAGE 35

35 (1 6) (1 7) (1 8) To complete this description of measurement SNR for DLIPS versus fluorescence alone, the effect of repeated mea sures must be considered as well. For the fluorescence measurements alone, the SNR will improve directly as a factor of the number of repeated observations. This is analogous to a maximum likelihood estimation of the fluorescence signal. But for the DLIPS SNR, this scaling is reduced during the computation of the DLIPS signal, since repeated measures can only be made for the Em pre Em post and the Dark signals but not the DLIPS signal directly. So the DLIPS SNR will improve by averaging repeated individua l measurements, but the effect will be diminished during the computation. In vivo Optical Diagnostics: Problem Formalization From a signal processing or information propagation perspective, the problems of understanding the experimentally accessible optica l properties of disease, and making a non invasive diagnosis can be described as a cascade of change of bases; propagating information from the actual disease state to the dete cted disease state as shown in F igure 1 6 As the goal is detection of a disease state, the underlying information begins as a (in this case) one dimensional representation of the true state, with examples including: healthy, benign, pre malignant, or malignant. This information some subspace change of basis transform which converts the information encoded in the disease state into an n dimensional vector of all of the features that describe any tissue disease state. Features

PAGE 36

36 in this case might include nuclear cytoplasmic ratio, shape, or basement membrane. A priori knowledge or accurate modeling of this projection for each of the disease state is critical for both traditional diagnostic me dicine or to understand the characteristics to exploit in designing optical diagnostic systems. In the case of a pathologist examining histological slides, that physician computes an approximation of the inverse transformation between the disease state and the observable features; to arrive at an estimate of the disease state, his diagnosis. If one is interested in building a device for detecting the disease state, then understanding the layer of detectable feature is critical for tuning that sensing system to use the optimal combination of those features to estimate the disease state. If a non invasive optical sensing scheme is the desired sensing modality, then this feature space is further transformed via a change of basis into a matrix of spatio optical properties, where in the first dimension, the relevant optical properties such as absorbance, scattering, and fluorescence are defined, and the second the unique position in the tissue for those properties are defined. When a measurement is actually made, this information is then transformed into the subspace of remitted light, a matrix of the spatial position of the light versus the wavelength of the light. This is the state of information readily accessible for non invasive optical diagnostics. In the fi nal transform, this information is transformed into an estimate of the disease state, the diagnosis. Again this hinges on being able to accurately approximate the inverse transformation to when considering this transformation cascade, no assumptions of linearity or

PAGE 37

37 uniqueness were made, making a closed form solution to the inverse transform impossible. Further, for an in vivo experiment or clinical measurement, many o f these information representations are inaccessible, essentially contained within a black box. By framing the problem in this way, there are two appropriate approaches to understanding the remitted spectral signature in the context of detecting the diseas e state: developing model based approximations to the transfer functions contained within the tissue black box, or defining an arbitrary transform for the remitted light that minimizes the error between the estimate of the disease state and the actual dise ase state. Using this framework for the problem space of oral cancer detection, the experiments presented here seek to define this black box of disease for DLIPS tissue spectroscopy through both experimental approaches as well as modeling. In combining the se two approaches, a deeper understanding of these transformations is realized with the potential for improving the state of the technology as a future high impact clinical diagnostic tool. Experimental Outline After this chapter the remainder of this dis s ertation is organized into four subsequent sections: Chapter T wo is adapted from a paper published in Journal of Biomedical Optics that demonstrates the proof of concept for using DLIPS in vivo. This work represents the first use of DLIPS in vivo, specifi cally monitoring the progression of pre cancerous skin lesions. Using multivariate statistical methods and optimal detection theory, the DLIPS technique is shown to provide a more robust measure of pre neoplastic tissue differentiation in this model. Furth er, the unique molec ular underpinning of the source of

PAGE 38

38 tissue contrast detected by DLIPS is considered and shown to vary spectrally from that of tissue fluorescence alone The t hird c hapter is a pap er that is being prepared for submission to IEEE Transacti ons in Biomedical Engineering where Monte Carlo modeling of DLIPS tissue interactions is employed to consider the molecular underpinnings of this technique as a disease detection technology. This work represents the first attempt to model both the DLIPS t echnique as well as excimer laser tissue photochemistry using Monte Carlo methods. The results demonstrate that the nature of the DLIPS technique permits specific coupling to the most superficial epithelial tissue layers, accessing molecular information th at is obscured when analyzing normal tissue fluorescence signatures. This unique access to complimentary molecular information further promotes previous finding that DLIPS may compliment fluorescence spectroscopy using orthogonal sensing methodologies. Cha pter F our presents a study of both dysplastic changes and benign pathology in the hamster cheek pouch using DLIPS that is being prepared for publication in Optics Letters This work explores the inherent contrast between oral dysplasia and a confounding no n cancerous oral pathology for both DLIPS and fluorescence spectroscopy; particularly germane as these non cancerous inflammatory lesions are most commonly mistaken for dysplasia (and vice versa). Using discriminant analysis and optimal detection theory, t his work demonstrates that clinical detection systems must be trained to recognize the contrast between dysplastic lesions and benign lesions; a non trivial result as this is not a common practice for either the development of clinical testing of commercia lly available devices.

PAGE 39

39 The final chapter summarizes these finding, considers the implications of this work, and provides future directions to build off of these results.

PAGE 40

40 Figure 1 1. Commercially available systems for augmenting oral cancer detection

PAGE 41

41 Figure 1 2 Layers of epithelial tissue in the two general t ypes of oral mucosa. A) keratinized. B ) non 15 (2001)] A B

PAGE 42

42 Figure 1 3 Pictures and histology from patients presenting with different types of oral pathology. Visual and cellular/molecular presentation often do not agree, thus the need for new diagnostic tools to detect oral dy splasia with a bio psy. A) Significant leukoplakia in a low risk site without dysplastic changes. This would often register as cancer using currently avai lable diagnostic aids B) Significant verrucoid leukoplakia in a low risk site without visible dysplasia, but a very worrys ome histologic profile. This normally wou ld not be biopsied C) Subtly visible lesion, but severe dysplasia rev eal in the biopsy. D) Severe verrucoid/papillary hyperkeratosis. Keratin layer is equal in thickness to the rest of the epithe lial layers. Severe dysplasia with increased nuclear/cytoplasmic rations, nuclear pleomorphism and basilar hyperplasia. [Photos and histology courtesy of Indraneel Bhattacharyya, 2012] A C B D

PAGE 43

43 Table 1 1 Co mmon endogenous tissue fluorophores Chromophore Excitat ion Min Excitation Max Metabolism Markers NADH FAD 290 nm 340 nm 440 nm 450 nm Peptides Tryptophan Tyrosine Phenylalanine 280 nm 275 nm 260 nm 350 nm 300 nm 280 nm Structural Proteins Collagen Elastin Hydroxylysyl Pyridinoline Lysyl Pyridinoline 28 0 nm 265 nm 330 nm 450 nm 350 nm 410 nm 450 nm 325 nm 325 nm 310 nm 385 nm 390 nm 530 nm 420 nm 500 nm 520 nm 400 nm 400 nm Lipid Degradation Products Ceroid Lipofuscin 340 395 nm 410 470 nm 430 460 nm 540 640 nm 500 695 nm Vitamin B6 and derivatives 310 340 nm 375 430 nm Porphyrin derivatives ~400 nm 600 650 nm

PAGE 44

44 Figure 1 4 Laser tissue interaction possible as a function of exposure time and irradiance [Modified from Boulnois JL. Photophysical processes in recent medical laser developments: a r eview. Laser Med Sci. 1986; 1:47 66]

PAGE 45

45 Figure 1 5 Cleavage of peptide bonds due to direct photolysis induced by absorbance of 193 nm photons from an excimer laser source [Unpublished figure, used with permission, David Hahn, 2012]

PAGE 46

46 Figur e 1 6 Signal propagation and transformation map for non invasive optical diagnosis of disease

PAGE 47

47 CHAPTER 2 COMPARATIVE EVALUATION OF DIFFERENTIAL LASER INDUCED PERTURBATION SPECTROSCOPY AS A TECHNIQUE TO DISCRIMINATE EMERGING SKIN PATHOLOGY 1 Motivation Biophotonics, the study of interactions between light and biological samples had its earliest applications in diagnostic medicine with the use of optical microscopy to study tissue samples and histological specimens. 39 Since that time, advances in optical technologies have spurred the creation of increasingly sophisticated instruments to st udy tissue for research and in the laboratory, with a particular focus on the early detection of cancer. Fluorescence based techniques have been developed to probe the endogenous fluorophores of tissue including collagen, nicotinamide adenine dinucleotide (NADH), flavins, and porphyrins. 40 Through the study of these endogenous markers, it has been shown that these molecules are correlated with certain pathological conditions and have unique distributions in tumors, and that these fluorescent molecules can be used to distinguish benign from malignant tumors. 26, 27 Generally, fluorescence spectroscopy is a nondestr uctive and sensitive technique that allows in vivo detection of organic and biological materials under different environmental conditions in very short times. Because the excitation/emission wavelength pair is a combination of the molecular structure and the overall molecular environment (e.g. biological matrix), fluorescence emission may provide discrimination among emitting materials. Rep rinted with permission from: R. T. Kozikowski, S. E. Smith, J. A. Lee, W. L. Castleman, B. S. Sorg, and D. W. Hahn, "Differential laser induced perturbation spectroscopy: demonstration of a new spectroscopic technique in a murine model of emerging skin pat hology," Journal of Biomedical Optics 17 067002 (2012).

PAGE 48

48 Despite the long standing utility of biophotonics based strategies in the laboratory, these advanced technologies are onl y beginning to be applied in the clinic. Epithelial tissues, often easily accessible or with minimal invasiveness via endoscopy, have been targets for the majority of clinical techniques employing tissue autofluorescence imaging or spectroscopy. This inclu des the detection of gastrointestinal cancers and diseases, colorectal cancer screening, and oral oncology. 41 43 Additionally, there are ongoing clinical trials evaluating novel fluorescence techniques such as fluorescence lifetime imaging, multip hoton imaging, hyperspectral imaging, enzyme activity imaging, and single molecule detection for in vivo diagnostic applications. 39, 44 Each of these strategies leverages measurement of intrinsic optical properties from superficial tissue layers to assay the presence of disease, or the potential for development. The promise of these optical diagnostic stra tegies, however, has been marred by high false positive rates when applied to in vivo clinical screening and diagnosis. 45 Consequently, following screening, biopsy of suspect les ions followed by histopathological analysis under visible light microscopy remains the gold standard for disease state confirmation in a large proportion of cancers. 46 48 This has resulted in a slow maturation process for point of care biophotonics and can be attributed to several limiting fa ctors inherent to the complexities of the clinical setting. Common to spectra taken from complex samples, high intensity, broadband fluorescence responses from tissue often obscure the rich biomolecular information found in lower intensity fluorescence em ission bands. As this broadband response could potentially be from a tissue fluorophore unrelated to disease, targeting specific biomarkers of disease has been challenging using these techniques. However, the

PAGE 49

49 largest hurdle to surmount in translating these spectral strategies to the clinic has been the significant patient to patient variation in fluorescent properties. 49, 50 This has been shown to be associated with race, age, sex, air temperature, and even deformation of the tissue when applying the probe. 49, 51 57 The results of this patient and sampling variability includes: fluctuations in absolute emission in tensity, emission peak shifts, and changes in the scattering and absorption properties of the tissue, among other effects. 49, 57 Brookner 49 for example, reported variations in peak fluorescence intensities varying by more than a factor of five between patients while the intra patient coefficie nt of variation was less than 25%. While fluorescence signals are directly linked to the local molecular structure of the targeted system, there clearly remains a need for further improvement in optical based sensing schemes to specifically address the va riations realized with absolute fluorescence intensity. In this paper we present the first application of a new spectral technique, differential laser induced perturbation spectroscopy (DLIPS), to the in vivo detection of epithelial pathology in an animal model. 36, 37 The DLIPS sensing scheme incorporates three complementary techniques to improve upon previous fluorescence based biosensing strategies: laser induced fluorescence emission, ultraviolet (UV) laser perturbation of tissue, and difference spectroscopy. 36 In this instance, fluorescence is used to measure the response of tissue fluorophores before and after the tissue is laser perturbed. The perturbation pulses from the d eep UV excimer laser (193 nm, 6.4 eV) are strongly absorbed by biological tissue and used to cleave molecular bonds within the extracellular matrix (ECM) as shown schematically in Figure 2 1. Irradiation of biological matrices at 193 nm can cause photoion ization, including strand breakage,

PAGE 50

50 locally denatured sites, inters trand cross linking, reactions via photo dimers, and other products. 33 In the current work, despi te being well below the intensity threshold for tissue ablation, permanent alteration of the underlying tissue structure is induced, with resulting changes within the fluorescence spectrum, specifically with respect to photoreactive biomolecules, as made a pparent with the DLIPS scheme. We note here that while no direct ablation is realized, a single photon of 193 nm radiation exceed nearly all bond energies in the biological matrix; hence permanent photochemistry is induced despite being below the critical photon flux to affect material removal. Additionally, because the pre and post perturbation spectra are combined (see Equation 2 1) into a difference spectrum the DLIPS technique mitigates unwanted contributions from unperturbed tissue fluorophores, br oadband fluorescence, and importantly, variations in fluorescence emission bands which are unique to the patient, but not necessarily to the targeted pathology. Equation 2 1 shows the DLIPS spectral response, namely: (2 1) where E m pre ( ) and Em post ( ) represent the fluorescence emission intensity recorded at each wavelength before (pre) and following (post) perturbation by the UV excimer laser, respectively. As defined, a negative DLIPS signal corresponds to a reduction in fluoresc ence intensity following the photo perturbation step, which is generally attributed to the destruction of a corresponding fluorophore. In contrast, a positive DLIPS signal corresponds to an increase in fluorescence intensity following perturbation, which may indicate destruction of a fluorescence quenching species and/or the destruction of a concomitant absorbing compound, thereby allowing more

PAGE 51

51 light to reach the actual fluorophore. Overall, the complexity of the local fluorescence environment provides th e opportunity for the perturbing UV radiation to affect a unique change to the resulting fluorescence response. Therefore, this combination of fluorescence, photochemical perturbation, and differential spectroscopy creates a completely unique spectral sig nature from targeted tissue. The result is a technique that specifically couples to important photosensitive tissue biomarkers of early pathological changes and that has promise to mitigate the apparent noise sources due to inter patient variations. The p rimary goal of this study is to characterize and evaluate the DLIPS biosensing technique using an animal model of epithelial pathology. Using principal 2 test, we demonstrate that this technique can detect chang es in epithelial tissue in vivo Further, we show via component loading analysis that this technique couples specifically to morphological changes in the ECM, which is unique compared to laser induced fluorescence. Using detection theory analysis of unproc essed principal component data we show that DLIPS is an inherently more robust technique for assaying the presence of pathology in this animal model compared to conventional laser induced fluorescence alone. Finally, to demonstrate the utility of the DLIP S technique in complementing existing diagnostic techniques, we employ decision fusion with laser induced fluorescence to further boost performance when detecting the pathological state. Materials & Methods DLIPS System The DLIPS system is shown schematica lly in Figure 2 2. For all measurements, fluorescence excitation was accomplished using a Q switched, frequency tripled

PAGE 52

52 Nd:YAG laser, while the UV perturbation was accomplished using a 193 nm ArF excimer laser. The 355 nm fluorescence excitation beam was aligned coaxially with the 193 nm perturbation laser using a dichroic mirror (193 nm) as a beam combiner. Prior to the beam combiner, a variable attenuator was used to reduce the excimer laser to the desired perturbation intensity. The 355 nm beam diame ter was about 30% less in diameter than the 193 nm beam (top hat beam profile) at the target plane, to ensure complete perturbation of the entire fluorescence probe volume. The co linear laser beams were passed through a pierced mirror positioned at 45 o t o the target plane. With the 355 nm beam incident on the target, fluorescence emission was redirected by the pierced mirror and then focused onto a fiber optic bundle using a combination of two 50 mm diameter UV grade lenses, where it was passed to a 0.3 m Czerny Turner spectrometer and recorded with an intensified CCD (ICCD) array detector. Prior to entering the fiber optic, two sharp edge filters were used in series: a high pass filter designed to block the residual 355 nm light, and a low pass filter d esigned to block any residual 532 nm light remaining from the frequency tripling process. With this system, fluorescence emission was successfully collected in the spectral window between about 390 500 nm. A digital delay generator was also used to synch ronize and control both the ICCD and the 355 nm laser, and a 3 s detector gate was used to collect the fluorescence, set to the correspond to the leading edge of the 355 nm laser pulse (20 ns full width); hence all prompt and delayed fluorescence was reco rded. Additional experimentation revealed that the majority of fluorescence emission corresponded to the fir st 100 ns of the detector gate.

PAGE 53

53 Chemical Initiation and Promotion of Mouse Skin Tumors Protocols were approved by the University of Florida Institu tional Animal Care and Use Committee. Tumor formation on female athymic nude mice (Hsd:Athymic Nude Foxn1 nu Harlan Laboratories, Indianapolis IN), six to eight weeks old, was induced and promoted by 7,12 dimethylbenz(a)anthracene (DMBA) (Sigma Aldrich, S t. Louis MO) in mineral oil (Fisher Scientific, Pittsburgh PA) at a concentration of 0.5% w/w applied topically to the dorsal skin according to previously described methods. 58 64 Application was repeated two to three times per week th roughout the 11 week course of the experiment. DMBA application was discontinued if mice began exhibiting signs of systemic toxicity, particularly weight loss. In all cases, cessation of topical application resulted in alleviation of toxicity symptoms. Further, all of these mice showed visible lesions during week eight, consistent with those mice who continued receiving topical DMBA. Topical application of mineral oil alone was used on the control mice. The protocol was terminated at 11 weeks, prior to the emergence of squamous cell carcinomas (SCCs) as the goal of the study was not to investigate the biological difference between cancerous and normal tissue but rather to investigate DLIPS potential as a precancerous diagnostic technique. Thus tissue c omparisons were made between healthy tissue and tissue during the course of SCC formation due to DMBA initiation and promotion. 59, 60 Histopathology At weeks four and eight, one mouse was selected from the control group as well as the DMBA treated group and euthanized using an intraperatoneal injection of Euthasol (Virbac AH, Ft. Worth TX). Likewise, the remaining mice were euthanized at the end of the study. Immediately following euthanasia, the dorsal skin was excised all

PAGE 54

54 the way down to the anteroposterior axis an d fixed in formalin (Sigma Aldrich, St. Louis MO). Fixed samples were paraffin embedded, sectioned at 5 m, and stained with hematoxylin and eosin (H&E). Tissue preparation was performed at the Cell and Tissue Analysis Core, McKnight Brain Institute, Uni versity of Florida (Gainesville FL). The sections were analyzed by a board certified veterinary pathologist. Collection of DLIPS and Fluorescence Spectra Starting in week two of the study, two control mice and two DMBA mice were selected at the beginning of each week for DLIPS and laser induced fluorescence spectral analysis of the skin. Prior to all DLIPS spectral measurements, each mouse was wiped clean with an alcohol wipe and anesthetized using a ketamine/xylazine solution (Phoenix Pharmaceutical, St Joseph MO; Lloyd Laboratories, Shenandoah IA) at 10 ml/kg of body weight. Each mouse was then positioned in an alignment holder and spectra were collected from the back of each mouse, two spots from each side of the mouse. For each spot, a fluorescence spectrum was recorded using a 200 shot ensemble average recorded at 5 Hz laser repetition rate, referred to as the pre perturbation spectrum. Immediately following the pre perturbation fluorescence measurement, the 193 nm excimer laser was used to delive r a total of 2500 perturbation pulses to the target spot with laser energy of 100 J/pulse (fluence = 3 mJ/cm 2 ) and a laser repetition rate of 50 Hz. Following perturbation, a delay of 30 seconds was introduced, after which a second fluorescence spectrum was recorded using an identical 200 shot ensemble average, referred to as the post perturbation spectrum. These two spectra were then subtracted to generate the difference spectrum at each location, with the difference divided by the pre perturbation spec trum to generate the DLIPS response for the given surface site. In this manner, a total of eight DLIPS

PAGE 55

55 spectra (4 spots x 2 mice) were recorded each week from each group (control and DMBA). Despite the emergence of some lesions on the treated mice by wee k eight, all spectra were collected from skin regions that appeared to be pathology free under visual inspection (i.e. visible lesions were avoided). Finally, the pre perturbation spectra alone were retained and averaged separately for each group (control and DMBA) and week, providing traditional fluorescence spectral data to be analyzed independently of the DLIPS approach. Multivariate Statistical Analysis At the conclusion of the study, all spectra were visually inspected and 22 individual spectra (~14% of the total collected) were rejected due to (1) excessive spectral noise and/or no observable difference in the pre and post perturbation spectra; indicative of low Nd:YAG laser and/or low perturbation laser intensity during the collection process, or (2 ) due to atypical spectral appearance (e.g. unusually large negative or positive perturbation) which is indicative of mouse movement (e.g. twitching) between the pre and post perturbation measurements. Following this process, the remaining spectra (pre p erturbation fluorescence and DLIPS) for both the control and DMBA groups were imported into MATLAB (The Mathworks, Natick MA) for processing. Difference spectra and fluorescence spectra were preprocessed using a five point moving average filter to remove acquisition noise. Fluorescence and DLIPS spectra were then analyzed separately, but using parallel methods, to facilitate side by side comparison at each analytical level. Spectra from the treated and control groups were grouped together in several diff erent ways and analyzed using principal component analysis (PCA). Clustering within the data sets was visualized using score plots and the validity of using only the first two principal components for analysis was verified using a scree plot of the

PAGE 56

56 princi pal components (cumulative variance explained greater than 95%). Sub groups from the entire data set were selected for individual PCA to highlight the emergence of differences between the DMBA treated skin and the untreated skin. For each PCA analysis, H 2 test was used to verify that the clusters from the treated and control samples were statistically different. 65 The component loadings from the first principal component were used to assess which spectral regions best explain variations within and between data sets. 66 Component loadings were squared to facilitate direct comparison of loading across wavelength regions between sub populations of the data set (this casts each ue in terms of percent variance explained in the particular component). The regions of the spectrum at higher component loading levels were considered to have the highest impact on the observed differences between DMBA treated and control mice. Detectio n Theory Analysis of Classification Performance To investigate the ability of the DLIPS technique to discriminate between healthy and pathological tissue, a likelihood ratio classifier was implemented as this has been shown to be the optimal detector to di scover a signal in noise, irrespective of the type of data passed into it. 67, 68 Further, this classifier implementation facilit ates direct performance comparison between laser induced fluorescence and DLIPS without the need for the subjective evaluation of classifier appropriateness for one spectral technique versus another. The spectral data set was first processed using PCA to reduce the dimensionality of the data and consolidate the majority of the variation in the first few resolved components. PCA was used because this change of basis maximizes the variance

PAGE 57

57 across all of the data and does not require a priori knowledge of c lass associations. Due to the large percentage of variance explained by the first principal component for both the DLIPS and fluorescence spectra, as well as the limited size of the data set, detection theory modeling was restricted to decisions made usin g this first component. In this study, the null hypothesis (H 0 ) was considered to be the absence of DMBA induced pathology plus some noise (e.g. biological background, shot noise in the ICCD), while the alternate hypothesis (H 1 ) was that a signal due to p athology is present in a noisy background, as shown in Equation 2 2. 67 no pathology markers (2 2) pathology markers present However, the overall inter sample variance is assumed to be dominated by the presence or absence of the pathological state, not the noise. No prior knowledge of the data distribution under H 0 or H 1 was assumed and the distributions were approximated using kernel density estimation (bandwidth set using a Gaussian approximation) as P(X|H 0 ) and P(X|H 1 ) respectively. 69 The probability density functions were combined the receiver o perating character (ROC) curves, as shown in Equation 2 3. For each operating point, the null hypothesis was rejected if the likelihood ratio was less than or considerati on to compute the false positive rate (FPR) and true positive rate (TPR) for the corresponding operating point, given by: (2 3 )

PAGE 58

58 To compare the inherent performance of the two spectral techniques, the area under the ROC curve (AUC) was used. 70 The maximum Matthews correlation between the tw o techniques as it provided equal weight to minimizing both false positive and false negative results. Finally, a multi modal pathology detector was designed using decision fusion in order to show the ability of DLIPS to complement traditional spectral tec hniques and boost the overall detection performance. Liao 71 and Jesneck 67 provide excellent reviews of the details for applying this technique to fuse multiple detectors as well as the benefits and limitations. The implementation employed here follows directly from Liao 71 without significant modification. Operating points for the local detectors were optimized by choosing the threshold that maximized the MCC. In this case, only the decisions from the DLIPS detector and the laser induced fluorescence detector were used to create a fused detector. Performance was evaluated by calcul ating the AUC of the ROC for the fusion processor and comparing that to the performance of a single detector. Results Histopathology Figures 2 3 ( A D ) are representative photomicrographs of the H&E sections of the skin at four, eight, and eleven weeks of th e study. The top left panel (Figure 2 3 ( A ) ) is from a control mouse after week four of the study. There is normal epidermal thickness with several layers in the stratum corneum. Hair follicles and sebaceous glands are in normal density within dermal col lagen and subcutis. Skin samples from control mice in subsequent weeks were comparable to the sample at week four. Figure 2 3 ( B ) shows

PAGE 59

59 the skin after four weeks of DMBA treatment which displayed moderate epidermal changes including acanthosis and orthoker atotic hyperkeratosis, characterized by thickening of the stratum spinosum, stratum granulosum, and stratum corneum. Hair follicles were reduced in density and the dermis in treated areas was almost devoid of sebaceous glands compared to skin from control mice. At the same time dermal collagen density was increased. After eight weeks of topical DMBA application, skin was characterized by more severe acanthosis, orthokeratotic hyperkeratosis and dermal fibrosis than in skin after four weeks, as shown in F igure 2 3 ( C ). Hair follicles were infrequently found in treated areas, and sebaceous glands were almost completely absent. Squamous cell papillomas were occasionally present in the epidermis (although not shown in this section). At the conclusion of the study, week 11, t he DMBA treated skin (Figure 3 ( D )) showed more severe epidermal acanthosis and hyperkeratosis. There was also an increased, but still low density, scattering of squamous cell papillomas throughout the epidermis. Dermal fibrosis was slig htly more severe than in samples from mice at eight weeks of DMBA treatment. DLIPS and Laser Induced Fluorescence Spectra Figure 2 4 ( A C ) shows the average DLIPS spectra of the DMBA treated skin as compared to the control samples at weeks two, six, and eig ht. Figure 2 4 ( D F ) shows the average of the laser induced fluorescence spectra at the same time points, again comparing DMBA and control groups. Over the weeks of DMBA application, noticeable changes in shape of DLIPS spectra emerge (DMBA versus control ), while these changes are not directly observable in the fluorescence spectra alone. Figure 2 5 shows individual (i.e. single mouse) spectral samples at the same time points of the study, illustrating the ability of the DLIPS technique to exploit interna l normalization of

PAGE 60

60 the spectra relative to the tissue, and thus revealing information about the underlying pathology. Both figures show the emergence of a noticeable band in the lower wavelengths of the DLIPS spectra throughout the pathological progressio n of the DMBA model. This information is either obscured or not detected using fluorescence alone, as the spectra from the skin (both control and DMBA treated) are dominated by the broad, correlated fluorescence peak centered on 460 nm. Multivariate Stati stical Analysis Figure 2 6 ( A B ) shows the score plots from the entire data set for the DLIPS data as well as the laser induced fluorescence spectra. In both cases, 68 individual spectra were used from the DMBA treated mice and 80 from the control mice thr oughout the eleven weeks of the study. For the entire DLIPS spectral data set the cumulative variance explained by PC1 and PC2 was 95% while these components explained 99% 2 te st returned a p value less than 0.001, confirming that the spectra collected from untreated mice are statistically unique from those collected from the DMBA treated mice. As an extension of this dimension reduction and remapping of the spectral data set i nto principal component space, the component loading were used to identify the important spectral regions in explaining the variance across the entire data set as well as several sub populations of the data set. Figure 2 7 shows the proportion of variance in a given data set as a function of wavelength. PC1 was the only component considered since greater than 90% of the total variance in principal component space was accounted for in this component for all cases. Figure 2 7 ( A C ) shows the proportion of v ariance using the DLIPS technique for the entire data set, the DMBA data subset, and the control data subset, respectively. Across all acquired spectra, the most

PAGE 61

61 important regions in explaining the differences are: below 400 nm, a peak at 420 nm, and a re gion of increasing importance at wavelengths above 460 nm. Additionally, when the control spectra and DMBA treatment spectra are analyzed separately, unique band regions emerge to potentially explain the observed pathological progression to lesion formati on in the DMBA model as opposed to the skin from the control mice. Primarily, the DMBA treated skin has an important band between 400 and 420 nm, populations shows much loading above 460 nm, the entire data set shows significant variability above this wavelength. This implies that this region may specifically account for variability between control and DMBA treated skin using the DLIPS technique. Remarkably, the region around th e peak fluorescence emission is of minimal importance using the DLIPS technique, a dramatic contrast to the component loadings for the traditional laser induced fluorescence spectra across the same data sets (Figure 2 7 ( D F )). The fact that the shape of t he component loading map closely resembles that of the fluorescence spectra is indicative of a strong dependence on intensity fluctuations at the peak emission wavelength in describing variations across the data set. Further, there are no unique spectral regions within either the DMBA treated or control groups when compared to the loadings from the entire data set. This sensitivity to signal intensity highlights an important limitation of traditional in vivo fluorescence techniques, namely, that subject t o subject variability can often obscure the relatively much smaller variability due to changes in tissue autofluorescence. 49 Finally, F igure 2 7 ( G I ) shows the same technique applied to the spectra collected after the perturbatio n laser pulse. This analysis was applied to this data to assess whether any new

PAGE 62

62 information gained from applying the differential technique is unique to DLIPS, or merely inherent to the post perturbation fluorescence spectra. Not unlike the component loa dings for laser induced spectra, these resemble the shape of the acquired fluorescence spectra, indicating that the most significant feature is the variability of the intensity of the broad fluorescence peak fluorescence centered at 460 nm. Again, the com ponent loadings are the same across all three figures, indicating limited coupling of these spectra to underlying tissue biomolecular structure. Pathology Detection Performance Using the same compressive transformation as above, sub populations of the data set were compared to describe the ability of each technique to discriminate the pathological (or pre pathological) state from normal skin. In this case, only the first principal component was considered due to the small size of the sub populations used i (week 2 11) segments, based upon the visual observation of similar pathological states within those time periods for the DM BA treatment group was considered pre pathological as both observational and histological evidence showed only the earliest stages consider ed definite pathological because dysplastic lesions and papillomas were visible during those periods of the study. PCA was performed independently on all of the data (DMBA treated and control) for the given time periods. Figure 2 8 shows the ROC curves de rived from the likelihood ratios computed using the DLIPS technique, and Figure 2 9 shows the curves for laser induced fluorescence. With the exception the

PAGE 63

63 better than u sing laser induced fluorescence, as measured by calculating the AUC for note is considered pre pathological, the DLIPS technique only lagged fluorescence by 2.5%. However, d 5.01% improvement in detection performance, respectively. In addition, Table 2 1 detectors, assumin g an equal weighting for false negatives and false positives. This table shows that for DLIPS, as the pathology matures, the specificity of detection stays at a high level while the sensitivity improves. On the other hand, when laser induced fluorescence is used the sensitivity still improves, but at the expense of specificity. Thus, even when the detection performance of these techniques is optimized using an off the shelf classification scheme (i.e.: artificial neural network, linear discriminant analy sis, etc.), the information encoded using DLIPS is more robust than traditional fluorescence spectroscopy and offers improved discrimination when detecting pathology. As evidenced above using component loadings, the DLIPS spectra and fluorescence spectra emphasize unique spectral regions and features. Assuming this means that these spectra reflect different underlying biological changes, or at the very least some amount of statistical independence between the two techniques, the DLIPS technique was combin ed with laser induced fluorescence to improve the overall detection performance. 67, 71 Figure 2 1 0 shows the ROC curve for this fused classifier as well as the ROC curves for the local detectors with the operating points used to

PAGE 64

64 is improved by 11.70% over detectio n using laser induced fluorescence alone. Based upon these results, DLIPS may serve as a new dimension in the composite evaluation of pathology for in vivo detection and diagnosis applications. Discussion In this study, we investigated a new spectral tech nique, DLIPS, and evaluated its applicability to detecting pathological changes in a murine model. Our findings show preliminary support for applying this technique as a stand alone medical diagnostic tool or as a complementary technique to traditional fl uorescence spectroscopy for the detection of pathology in vivo Of particular note were the findings that the DLIPS spectra emphasize different spectral regions from the laser induced fluorescence spectra and that the endogenous pathology detection potent ial for DLIPS shows superior performance when compared to laser induced fluorescence alone in this study. As has been shown before, the strong fluorescence peak centered on 460 nm for the fluorescence spectra (Figures 2 4 and 2 5) can most reasonably be a ssigned to the tissue fluorophore NADH. 26, 72 74 While this is an important tissue fluo rophore, and is a primary target of fluorescence tissue assessment, it serves as a marker of increased cellular metabolism, which is a hallmark of dysplasia, but not necessarily unique to it. In conjunction with increased cellular metabolism, one of the ea rliest markers of pre cancerous progression is cellular infiltration and the release of growth factors and cytokines. This unchecked cellular signaling results in the proliferation of fibroblasts, increased collagen synthesis, and suppression of collagenas e production, with the overall effect of restructuring the local ECM. 75 These effects suggest that techniques which target collagen remapping might provide direct diagnostic coupling of the spectra to the pathology. Figure 2 7 ( A ) sho ws that this spectral region, up to 420 nm, was the

PAGE 65

65 most prominent region in explaining the variance when using DLIPS, and is in the range for emission due to collagen. 26, 72 74 This agrees with previous work showing that excimer laser pulses can break collagen amide bonds, resulting in a pronounced difference spectrum band. 36, 37 Further, when PCA was performed independently on the control spectra and the DMBA treated spectra (Figure 2 7( B C )), a unique collagen region emerges in each case, at lower wavelengths for control skin and at higher wavelengths for DMB A treated skin. This region of higher variability is likely due to the progression of ECM remodeling associated with pre cancerous lesion formation throughout the 11 weeks of the study for the DMBA treated skin. As for the control skin, this region of hi gher variability could be due to the aging related remodeling of skin or the natural turnover of skin ECM proteins. 49, 53 In this regard, DLIPS serves as a convenient probe for ECM structure that was previously inaccessible when exciting natural fluorescence at 355 nm, as it has been no ted that at this excitation wavelength it is not possible to resolve collagen fluorescence from that of NADH. 76 Our component loading plots for the laser induced fluorescence spectra and the post perturbation spectra (Fig ure 2 7( D I )) illustrate this effect as the variation is most strongly tied to the intensity of the broad NADH peak at 460 nm. The other important spectral region in distinguishing DLIPS spectra is above 460 nm. This band of increasing importance up to 5 00 nm is not clearly assignable to metabolic or structural fluorophores, but is definitely a point of contrast between DMBA treated and control skin as it only shows up in the combined spectral pool (Figure 2 7( A )). The proximity to the 460 nm NADH peak c ould indicate it is simply a residual region from that emission but further work will be

PAGE 66

66 required to isolate the molecular source of this DLIPS region and assess its diagnostic strength. The DMBA model of pathogenesis has been well characterized, particula rly with regard to in vivo models for the evaluation of fluorescence based techniques to detect and diagnosis cancerous and pre cancerous lesion formation. 59, 60, 74 While our study found improved performance using the DLIPS technique, other studies have reported accuracy rates up to 90% for detecting precancerous morphological changes and sensitivities and specificities ranges of 76 95% and 83 95% respectively when using fluorescence spectroscopy. 59, 74, 77, 78 However, a major difference of these studies was the focus on an algorithmic approach to boosting detection performance using techniques like partial least squares discriminant analysis or a support vector machine to develop a maximized basis on which to separate the data and subsequently only resolve a single det ection operating point. 60, 74, 79 In diagnostic medicine, there is often a much greater cost associated wit h a false negative result as opposed to a false positive, as evidenced by the high sensitivities and low specificities reported for physician driven diagnostic techniques. 47 Thus, an understanding of the entire performance curve is critical as different applications may require operating far from the optimal decision threshold in order to boost either sensitivity or specificity. In our work, a technique driven investigational study, PCA was employed allowing the inherent variance of the da ta to be the factor across which the dimensionality was reduced and the data were separated. This permitted a probing of the primary source of variation; the emerging pathology or noise. Since the tissue under investigation in this study was not actual s pontaneous human lesions ( in vivo or

PAGE 67

67 otherwise), development of a discriminant model to improve detection was not considered, as it must be derived from the intended target, but will be implemented when appropriate in future work. Despite this significant difference in change of basis methodology, the perform ance metrics reported in Table 2 1 indicate that this technique has at least comparable detection performance to previously reported fluorescence studies in DMBA induced model pathology. However, when translating results to actual clinical use, performance of fluorescence spectroscopy is reported to drop to 70 82% sensitivity and 68 92% specificity. 79, 80 This is most often attributed to patient to patient variability and is a known confounding factor for clinical in vivo fluorescence measurements. Although mu ltiple standardization and normalization techniques have been proposed to combat this issue, it necessarily comes at the cost of lost information from the collected spectra. 60 Our results in Figure 2 9 support this notion that without a normalization strategy, emission spectra can vary widely due to factors other than changes in the underlying pathological condition. The severe dr op in performance for identifying pathology during weeks five through eight is likely due to emission fluctuations associated with the data collection process or the particular spots being interrogated on the back of particular mice. In the case of DLIPS, the spectra represents a differential response measurement from the same location and is in that sense self referencing in order to specifically remove variation associated with inter sample differences. Figure 2 8 supports this stabilization of measurem ent as it shows the detection performance improving uniformly with the maturation of the DMBA induced pathology.

PAGE 68

68 Irrespective of the diagnostic performance of this technique using a model of pathology, or how that compares to other techniques, DLIPS still provides a new spectral dimension with which to investigate epidermal pathology. As more diagnostic tests and techniques become available, there is an increasing trend in medicine toward boosting the performance of pathology detection by combining dispara te data sets from multiple unique sources. Although this can be a challenging problem, techniques from the bioinformatics and machine learning communities can be applied to optimally fuse this information. 67, 81 83 In the case of DLIPS, its largest impact may be in offering a new source of unique information about the ECM to be processed along with multiple other markers of disease. O ur preliminary findings (Figure 2 10) fusing this technique with fluorescence spectroscopy shows that improvements in diagnosis can be realized by combining this technique with other independent biomarkers for the target pathology. We note an additional p oint in regard to the DLIPS technique in the context of orthogonal sensing, namely, that the scheme is readily implemented with Raman spectroscopy as the primary optical scheme. 36 In other words, Raman spect ra are recorded pre and post perturbation, and the difference Raman DLIPS spectrum is then calculated in the same manner as done with a fluorescence probe. In summary, our work here presents a promising preliminary evaluation of DLIPS for the in vivo de tection of disease. We succeeded in revealing the underlying sources of physiological contrast to which DLIPS is sensitive and also established the baseline performance of DLIPS in detecting pathological changes. Applying identical data collection and ev aluation techniques, DLIPS showed superior detection performance when compared to laser induced fluorescence spectroscopy. Future work will include

PAGE 69

69 the refinement of this technique in subsequent animal studies, application of the DLIPS technique to other spectral domains, further investigation of the endogenous chromophores responsible for the spectral changes under DLIPS detection, with the long term goal of translating this technique to the clinic.

PAGE 70

70 Figure 2 1. DLIPS acquisition scheme

PAGE 71

71 Figure 2 2. Schematic of the DLIPS system

PAGE 72

72 Figure 2 3 .H&E stained histology sections of skin A ) C ontrol mouse, week 4 B ) DMBA mou se, week 4 C ) DMBA mouse, week 8 D denotes thickend regions of the stratum corneum, symptomatic of orthokeratotic hyperkeratosis. ( X ) hig hlights the thickening of the stratum spinosum, characteristic of acanthosis. ( ) denotes areas of increased d ermal collagen density. D ) a section through a developed papilloma A B C D

PAGE 73

73 Figure 2 4. Average spect ra at weeks two (n=6, 7), six (n=5, 4), and eight (n=7, 8) A C ) DLIPS D F ) normalized fluorescence denotes DMBA teated skin and untreated skin spectra are unlabeled. A B C D E F

PAGE 74

74 Figure 2 5. S pectra at weeks two, six, and eight A C ) DLIPS D F ) fluorescence All measurements from the same spot on the backs of DMBA untreated mice A B C D E F

PAGE 75

75 Figure 2 6 Score plots for the pri ncipal component analys is A ) DLIPS B ) L aser induced fluorescence All spectra from week two to eleven of the study. Control ( ), DMBA treated weeks 2 & 3 ( ), weeks 4 & 5 ( ), weeks 6 & 7 ) A B

PAGE 76

76 Figure 2 7. Proportion of variance in Principal Component 1 accounte d for by each wavelength A C ) DLIPS spectra D F ) L aser indu ced fluorescence spectra G I ) P ost perturbation spectra. In each row, the leftmost plot considers the variation across the entire data set, the middle plot repre sents the variation from only from DMBA treated mice and the right plot represents variation within the control data. A B C D E F G H I

PAGE 77

77 Table 2 1 Detection performance metrics using DLIPS, laser ind uced fluorescence, or a decision fusion approach. Early (wk 2 4) Middle (wk 5 8) Late (wk 9 11) DLIPS Fluorescence DLIPS Fluorescence DLIPS Fluorescence Fused AUC 0.734 0.755 0.835 0.643 0.913 0.869 0.971 Sensitivity 0.429 0.524 0.783 0.870 0.833 1.00 0 Specificity 0.957 0.913 0.815 0.593 0.955 0.727

PAGE 78

78 Figure 2 8.ROC curves for the detection of DMBA induced skin pathology using the DLIPS technique. The data set was partitioned into early (weeks 2 4), middle (weeks 5 8), and late (weeks 9 11) time periods of papilloma development resulting from topical DMBA application.

PAGE 79

79 Figure 2 9. ROC curves for the detection of DMBA induced skin pathology using laser induced fluorescence. The data set was partitioned into early (weeks 2 4), middle (weeks 5 8) and late (weeks 9 11) time periods of papilloma development resulting from topical DMBA application.

PAGE 80

80 Figure 2 10 ROC curves for the fused decision incorporating both fluorescence and DLIPS spectra to boost the optimal performance in detecting DMBA in duced s nine to eleven. The other ROC curves are from the DLIPS and fluorescence based techniques alone, with the selected operating points noted ( ).

PAGE 81

81 CHAPTER 3 DLIPS SPECTRA OF ORAL MUCOSA PATHOLOGY: MONTE CA RLO MODELING OF SITE SPECIFIC OPTICAL AND FLUORESCENT PROPERTIES AND THE CHANGES INDUCED BY 193 NM PHOTOCHEMICAL PERTURBATION Motivation Superficial carcinomas like oral cancer provide an opportunity for technological approaches to earlier screening and di agnosis due to the nature of their development. These carcinomas are invariably preceded by dysplastic precancerous cellular changes, in which histological changes associated with malignancy are often confined to the epithelial layer. 84 Dysplasia, defined as unequivocal neoplastic epithelium, is at present the m ost reliable biomarker of malignancy. Many of these earliest changes (excellent targets for diagnosis and screening) are not detectable visually, but these molecular changes are often detectable using fluorescence or optical spectroscopy both in vitro and in vivo As a result, there has been an explosion of proposed optical approaches to addressing the problem of oral cancer diagnosis. These technologies have primarily attempted to translate successful in vitro techniques directly into tools for evaluating cancer in vivo The oral mucosa specifically, because of the ease of access, has been targeted for a large number of clinical techniques employing tissue autofluorescence imaging or spectroscopy. 41 43 The promise of these strategies however, has b een marred by high false positive rates when applied to clinical screening and diagnosis. 45 This is because the preclinical phase of the dysplastic transformation into oral squamo us cell carcinoma is varied, often asymptomatic, and easily lumped into two generic oral pathology categories: leukoplakia or erythroplakia. 17, 18 Leukoplakia is describes a white patch of oral mucosa that cannot be removed through gentle conta ct.

PAGE 82

82 The lack of color is often due to the hyperkeratosis of the superficial epithelium and is responsible for the non 85, 86 Erythroplakia presents with a red, inflamed, velvety surface 85, 86 These terms are generally terms of exclusion, reserved for when the underlying cause is undetermi ned. Most often, these lesions are the result of benign inflammation, hyperkeratosis, or ulceration and may persist or self resolve. And these benign diseases are in fact more prevalent than oral cancer, as up to 15% of the population presents with oral mu cosal disease (non malignant or pre malignant) at any given time, despite the much lower incidence of oral cancer in the population. 17, 18 These generic characteristics of oral pathology often mask the slow pre malignant transformations from bo th the dental professional and currently available diagnostic aids. It has been reported that variations of up to a factor of five have been observed in the endogenous fluorescence signals between patients. 49 Recently, a new tissue fluorescence probing technique, Differential Laser Induced Perturbation Spectroscopy (DLIPS), has been proposed to address some of these limiting factors for traditional tissue optical diagnostics as well as open another optical sensing dimension to augme nt tissue spectroscopy techniques. 38 The DLIPS sensing scheme is based on recent research showing that the biological matrix may be altered by low intensity, su b ablative deep ultraviolet radiation (193 nm) such that the intrinsic fluorescence response is perturbed. 36, 37 It has been demonstrated that this novel sequential combination of fluorescence probing, UV photochemical perturb ation, and repeat fluorescence probing, realizes a powerful new spectral dimension based on difference spectroscopy that is strongly coupled to the local biomolecular matrix. 36, 37

PAGE 83

83 Because of the intimate coupling of the pertu rbing laser to the biological matrix, molecular structures of abnormal tissue (e.g. dysplastic precancerous cells) have been shown to respond differently than normal tissue. 38 This technique examines differences in fluorescence response; hence many of the significant variations in the absolute optical response, generally observed in patient to patient populations, can be overcome. Preliminary pre clinical investi gation of DLIPS as a diagnostic modality for the detection of pre malignant lesions has shown promising results, however the physical and biochemical underpinnings of this technique have received no consideration aside from drawing speculative connections between common tissue fluorophores and the diagnostic regions of the DLIPS spectrum. 38 Understanding the relevance of DLIPS to in vivo diagnostics requires a f undamental understanding of how this spectral dimension responds to the biochemical and morphological changes associated with disease. Further, as the 193nm laser perturbation interacts photochemically with the tissue, a thorough understanding of this phen omenon is essential as well. Monte Carlo methods have previously been applied to the study of light tissue interactions and in the present study are extended to consider the propagation and photochemical activity of coherent 193nm light as well as the impa ct of these interactions on the tissue optical and fluorescent properties. Previous work with Monte Carlo models of multi layered tissues has shown success in transforming experimental measurements into accurate representations of the underlying optical a nd fluorescence properties, and these models have been used to understand the hierarchal structure of epithelial tissue in the context of the remitted light signal. 9, 28 30, 87 Further, sensitivity analyses of these models have demonstrated

PAGE 84

84 that the remitted light is extremely sensitive to changes in the optical properties and morphology of the target tissue. 29 In the case of DLIPS based tissue measurements, these changes in optical p roperties are not just due to pathological changes but molecular changes induced by the perturbation laser. 34, 36 38 Here we demonstrate that by using model based approaches, the diagnostic potential of DLIPS can be a scribed to specific biochemical changes of both disease and photochemical perturbation. The purpose of this study was to develop a model of the DLIPS tissue interactions, within the context of the ultimate sensing modality, fluorescence spectroscopy. Beca use of the high degree of variability in the fluorescent properties of the individual layers of buccal epithelium, as well as the limited penetration depth and photochemical potential of 193 nm light, this was built around multi layered epithelial tissue m odel as described previously. 9, 29 The optical properties and gross morphologies of these tissue simulations were varied to describe healthy, benign, and dysplastic oral mucosa. The model was populated with tissue opt ical properties from measurements of these properties reported in the literature. 9, 28 31, 33, 34, 87 Using this model, we were able to draw connect ions between the remitted fluorescence from tissue and the contributions of individual tissue layers and fluorophores to that signal, in addition to showing how excimer perturbation affects this signal. As supporting evidence, we present outputs of Monte C arlo simulations before and after photochemical perturbation as well as for models of healthy oral tissue, dysplastic (pre malignant) oral tissue, and benign oral pathology. Specifically, we examine changes total remitted fluorescence as well as from indiv idual tissue layer of the oral mucosa. Further we examine the computationally modeled DLIPS spectra and consider

PAGE 85

85 the origins of this spectral signature in the tissue using the remitted fluorescence spectra. Finally, we compare these spectra to measurements made in vivo. 38 In presenting these outputs from the Monte Carlo model, this study confirms previous empirical observations that using laser perturbation with difference spectroscopy does change the direct coupling of the returned spectral signature to the target tissue. While further validation will need to be performed for this technique using excised tissue layers, this work presents a mechanistic framework f or understanding the complex spectral signatures generated by this technique; a crucial component of understanding the future impact of this technique as a clinical detection modality. Materials & Methods Simulation Workflow As the goals for this simulatio n exceed that obtainable through a traditional Monte Carlo modeling approaching to light propagation in tissue, Monte Carlo methods were embedded within the overall simulation scheme (F igure 3 1). In a simple approach to modeling a DLIPS measurement, one e nvision the linear experimental workflow (fluorescence measurement, perturbation, second fluorescence measurement) as being modeled as sequential Monte Carlo simulations of the fluorescence measurement, excimer laser perturbation, and a final fluorescence measurement. In addition however, this workflow takes into consideration the photochemical interplay between excimer laser perturbation and tissue optical properties, tradeoffs between mo deling 193 nm light using Monte Carlo simulations and the Beer Lamber t Law, the layer dependence of tissue optical and fluorescence properties, and the changes in optical properties due to disease. The global analysis centers around the information embedded in the remitted

PAGE 86

86 light; how both disease and photochemical perturbat ion (separately and jointly) influence that collected spectrum. Monte Carlo Model Construction A basic Monte Carlo framework for simulating light propagation in tissue was constructed in MATLAB (The Mathworks, Natick MA) for the purpose of this study, base d upon previously developed models. 21 The rational for developing the modeling software in house was to have the flexibility to model photochemical and fluorescent interactions, as well as the impact of the former effect on the tissue optical properties withi n the model. The source was modeled either as a tophat beam profile from free laser beams or from direct contract of a fiber optic bundle with the tissue, again with a tophat beam profile. Each beam profile was modeled with a diameter of 1 mm. The free bea m was assumed to have no divergence at the point of tissue irradiation while the photons emitted from the fiber were emitted at an angle of up to 12, the maximum emission angle calculated from the numerical aperture of the fiber. Additionally, the refract ive index of the window for the fiber probe was considered at the boundary between the probe and tissue. The basic functionality of the model was validated by comparing the diffuse reflectance from this model versus a free, peer reviewed Monte Carlo model. 88 The two models were compared under identical conditions for two sets of tissue optica l properties. Each simulation was run five times using 50,000 photon packets. The results for these simulations were tested for equivalence using a modified t test in the two one sided tests (TOST) framework and using Bland Altman analysis, with a differen ce of up to 5% in the total remitted light accepted as equivalent.

PAGE 87

87 Multi Layer Tissue Model The buccal oral mucosa was modeled as a five distinct tissue layers, based off of histological and anatomical data as well as previous epithelial tissue optics mode ls. 9, 29 The l ayers were defined as shown in Figure 3 2 with the thickness of each layer used as reported in the literature. 9 There were three defined layers of epithelium above the basement membrane, as wel l two distinct layers of lamina propria below the membrane. The fluorescent properties of each layer were modeled by using the dominant fluorophore in that tissue layer as discovered using confocal microscopy and in vitro immunohistochemical techniques, an d as used in other tissue Monte Carlo models to study fluorescence alone. 9, 29 The major fluorophores used here were keratin, FAD, NADH, and collagen. The shape of the emitted light spectrum for each of these fluoroph ores was found in the literature. 29, 89, 90 Quantum yields for each of the five layers of tissue were adapted from a model of the optical properties of skin. 91 While skin is analogous to oral mucosa in its general structure, it varies greatly in the abs orption and scattering properties in the superficial layers due to different cellular structures and the presence of melanin at much higher concentrations. 23 Despite this, the dominant fluorophores are generally considered to be consistent on a layer by layer basis between epithelial tissues. 90, 92 Variability in the fluorescence in different types of tissue is accounted for b y the relative fluorescence; the product of the quantum yield (a fixed property), the fluorophore extinction coefficient (a fixed property), and the fluorophore concentration (variable for different tissue models). The anisotropy and refractive index for each layer were considered to be wavelength independent and used as described previously. 9, 23, 91 A summary o f these properties is given in Figure 3 2

PAGE 88

88 Models of the spectral properties of absorption coeffici ents and scattering coefficients of the tissue were built from aggregated, experimentally collected values from various sources. 9, 28, 30, 93 The spectral response for scattering was developed as a linear combination of Mie scattering and Rayleigh sc attering for the UV Vis range (E quation 3 1) and extrapolated into the deep UV range using a Rayleig h scatteri ng with constant offset model (E quation 3 2). The exponent used for the Mie theory term was adopted from Jacques. 94 In both the Mie and Rayleigh case s, the parameters were lumped together into coefficients in order to facilitate fitting to exp erimental data (E quations 3 1 and 3 2). The spectral variation in the absorption coefficient was likewise considered and modeled after data collected directly fro m tissue samples and constituent chromophores. 9, 23, 29, 30, 93 In this case both the exponent and coefficient were f it to the availa ble data using E quation 3 3. The fits to this data are shown in F igure 3 3 and the values for s and a at 355 nm are given in Figure 3 2 The notable exception is for layer of tissue containing hemoglobin, where the extinction coefficient was used as repo rted in the literature and the hemoglobin saturation was fixed at 80% function for the wavelength dependence of the background tissue absorbance was used. 94 (3 1) (3 2) (3 3)

PAGE 89

89 Fluorescent Interactions In addition to the consideration of absorption and scatter ing of photons at each step of the Monte Carlo simulation, generation (and consequently propagation of fluorescence light) was treated as follows, developed loosely from. 9, 29, 91 In this model, the excitation wavelength was fixed at 355 nm, and the range of possible emission wavelengths was from 356 nm to 58 5 nm. At each interaction site in the tissue, the was considered by converting the steps of fluorescence generation into probabilities of each event in the cascade happ ening and then taking the product to find the relative fluorescence efficiency over the emissio n range (E quation 3 5). In this equation, the ratio of absorption coefficient to transport coefficient represents the probability that a photon will be absorbed, quantum yield represents the ratio of absorbed photons that generate fluorescence from a single fluorophore. The re lative layer fluorescence, RF (E quation 3 4), is an experimentally derived combination of absorbance and quantum yield for each tissue mode l and layer described in Figure 3 2 29 (3 4) This combined parameter was estimated previously using co nfocal fluorescence microscopy. 9 experimental data collected from excised tissue layers as well as from solutions of the relevant constituent fluorophores. 9, 23, 87, 90, 92, 93 This data was fit to a Weibull function ( E quation 3 6) using a non linear solver in MATLAB When fluorescent photons were generated, no additional photon weight was deposited in the tissue and all of the weight

PAGE 90

90 of the excitation photon was transferred to the emitted photon. Figure 3 3 shows the f luorescence efficiency, calculated from the quantum yield measured in the given layer of tissue where the fluorophore dominates and the emission spectral shape modeled after measurements made on exogenous fluorphores. 89 (3 5) (3 6) Photochemical Interactions Biochemically, peptide bonds, aromatic amino acids, and nucleic acids all stron gly absorb 193 nm. However, it has been shown that in cultured cells, 193 nm light induces far less photochemical DNA damage when compared to middle UV light. 31 This is due to the high absorption coefficients of peptide bonds and aromatic amino acids, both present in high concentrations in the extracellular matrix and the cytosol effectively blocking the light from interaction with the genetic material contained in the nucleus. For the purposes of this model, absorption by protein was the only interaction considered. In the case of 193 nm light, the energy of a single photon 6.4 eV, exceeds that of peptide bonds as well as the aromatic ring structures of characteristically fluore scence amino acids. As the energy exceeds the bond energy in all cases considered, this model assumes that if a photon is absorbed, then the bond breaking reaction will proceed in accordance with the quantum yield for that reaction. The probability of a ph oton being absorbed in the Monte Carlo model was given as in the fluorescence model, the absorbance fraction versus scattering. Multiplying that by the effective cross section, the net concentration and the net quantum yield for all peptide dissociation r eaction, the

PAGE 91

91 net ph otodissociation is calculated (E quation 3 7). This is the generic propensity for absorbed light to beak some bond in a protein, not specifically a peptide bond. The net photodissociation reaction is related to all of the possible specifi c bond breaking mec hanisms by the relationship in E quation 3 specific mechanism. In this case since both the quantum yield and extinct ion coefficients are considerably higher for aromatic amino acids as opposed to aliphatic ones, the contribution from aliphatic amino acids is assumed to be negligible. Using this relationship, the photodissociation reactions can then be grouped into two m ain types, t hose that break peptide bonds (E quation 3 8) and those th at break aromatic side chains (E quation 3 9). 34 (3 7) (3 8) (3 9) (3 10) In the most superficial lay er of buccal epithelium, keratin is both the dominant fluorophore as well as the dominant protein, and has an estimated peptide bond density of 1 M and an estimated aromatic ring density of approximately 0.065 M. 32 Unlike collagen in which the source of fluorescence is the cross linking molecules, keratin fluorescence arises primarily from the aromatic amino acids nested into the protein

PAGE 92

92 coiled supra molecular structure. Therefore it is reasonable to assume that photolysis of the aromatic residue will result in both a loss of fluorescence as well as a decrease in the absorbance cross section at 355 nm. However, if the bond cleaved is a pep tide bond, there is no change to the absorbance cross section at 355 nm but there is the potential to disrupt the supra molecular structure responsible for the unique fluorescence character of keratin. It was assumed in this case that the closer the peptid e bond was to an aromatic residue, the higher the likelihood it would disrupt the local structure critical to fluorescence generation. Thus the likelihood of peptide bond cleavage inducing loss of fluorescence was given as the fraction of aromatic amino ac ids in a typical keratin fiber to the total number of peptide bonds. Similar treatment was given to deeper tissue layers but due to the limited penetration depth of 193 nm light, these layers were not modeled as no photons propagated into any layer beyond the superficial epithelium and using the Beer Lambert law, the incident photon energy drops by an order of magnitude approximately every 5 m. Pathology Models Models for benign pathology and dysplastic pathology were modeled by combining histopathology d ata, confocal microscopy data, optical coherence tomography data, and tissue spectroscopy studies, as well as clinical observations available in the literature. 9, 23, 29, 87, 89, 90, 92, 93, 95 To model our benign pathology, a pe rsistent inflammatory lesion with hyperkeratosis was used as a model of a generic benign leukoplakia. Dysplasia, was modeled by taking into account increased cellular metabolism, loss of cellular differentiation, and a thickening of the superficial epithel ial layer among other characteristics as previously d esc ribed and as shown in Figure 3 2 9, 96, 97 Further, both

PAGE 93

93 are often characterized by a loss of fluores cence upon UV excitation, which was only addressed by using previously reported experimental measurements of layer fluorescence. 9 Figure 3 2 summarizes the differences from the baseline model for each of the model pathologies as well as the biolog ical justification for the previously Simulation Details Masks of the photochemical damage and the resulting modifications to the local absorption and fluorescence properties of the tissue were created by running simulations of the ArF laser perturbation five times using 10 6 photons for each trial and then averaging the results. Simulations of pre and post perturbation tissue fluorescence for each of the tissue models described w ere performed with 10 6 photon packets each. Results Model Validation For both sets of tissue parameters tested in the Monte Carlo system presented here versus a previously validated model, the null hypothesis (ie: that the data sets are from different dis tributions) was rejected with a p < 0.01 using the TOST methodology and accepting a up to 5% difference between the model o utputs as equivalent. Figure 3 4 shows the Bland Altman plots for the two Monte Carlo frameworks, with the bias and limits of agreeme nt noted for each set of model parameters. Pre Perturbation Fluorescence Figure 3 5 ( A C ) shows the fluence, in logarithmic units, of photons through each of the three tissue models: healthy cheek, dysplastic oral mucosa, and a benign buccal lesion respec tively. Comparison between these three plots demonstrates that even

PAGE 94

94 subtle changes to the tissue can greatly influence how and where the light interacts with tissue. What is not apparent here is how these properties influence the generation and propagation of fluorescence light specifically. Despite the tissue depth in the figure cutting off at 1,600 um, the simulation actually ran using a depth of over 10,000 um, long enough to consider the deep stroma an infinite medium for this model. Figure 3 6 however considers this full depth of tissue for each o f the models. This figure shows the fraction of the light absorbed by each layer as compared to the fluorescence generated in that tissue layer, and also shows the fractional efficiency of fluorescence generation for each layer. More important however is the light that is actually remitted from the tissue, shown in (d f) for the healthy cheek, benign leukoplakia, and a dysplas tic lesion respectively. Again there are obvious differences between the model s, particularly in the loss of total fluorescence observed for the pathological states. It is interesting to note however that despite losing fluorescence intensity in these two cases, the layer by layer breakdown indicates different reasons for this obser vation. Also of note is the small signals from the superficial epithelium and the intermediate epitheliu m for healthy tissue and the marked increase in fluorescence signal from these layers for either benign or dysplastic pathology models Additionally, th ere is considerable fluorescence information lost from the collagen network of the stroma, likely due to the high absorbance of the stromal layer as well as the tissue depth through which that emitted light must travel in order to reach the surface of the oral mucosa. Excimer Laser Perturbation Patterns Figure 3 7 shows the fluence of 193 nm light through the superficial epithelium. In the deep UV range the absorbance properties of tissue dominate the scattering

PAGE 95

95 properties, explaining the high degree of pre servation of the collimated beam profile and the extremely low penetration depth. Figure 3 8 shows the fractions of photons converted into different bond breaking reactions in the tissue. The slightly elevated levels of loss of fluorescence reactions over aromatic amino acid interactions is explained by the fact that the model takes into account some local area influence of peptide bond cleavage on the fluorescent properties of keratin. The slight drop in the middle of the fluence pattern is an artifact of the Monte Carlo method; it is often hard to pass photons through the small volume of voxels on the axis of radial symmetry. This effect is particularly prominent when absorbance dominates scattering, as is the case for 193 nm light (the opposite effect is in fact observed when scattering dominates absorbance). 21 Post Perturbation Fluorescence Post perturbation patterns of fluorescence, as assessed using fluence patterns and remitted light signatures, looked similar to pre perturbation patterns, as expected fro m previous experimental studies of DLIPS. 38 Computed Spectra l Remittance and Layer Specific Signals Figure 3 9 shows sources of contrast in the DLIPS spectrum o n a per layer basis for healthy tissue, pre malignant lesions, and benign lesions. Interestingly, despite modeling hyperkeratosis in the superficial epithelium for both benign and pre malignant disease, the DLIPS signal from this layer is much weaker when compared to healthy oral mucosa. Additionally, despite the small fluorescence footprint of FAD when excited at 355 nm, the DLIPS spectral characteristics integrate significantly different levels of signal from the intermediate epithelium for the different oral mucosa states considered here. Although there is some change observed in the DLIPS signals from basal

PAGE 96

96 epithelium and the stroma, it is relatively small which is not surprising given the depth of these layers in the tissue when compared to the penetrat ion depth of 193 nm light. In contrast to the changes observed in for the DLIPS spectrum as a function of tissue layer and pathology, F igure 3 10 shows the signal contributions from each tissue layer for fluorescence alone. Based upon the large degree of e xperimental work in fluorescence tissue spectroscopy assigning epithelial fluorescence at 355 nm primarily to collagen, it is not surprising to find that the most prominent feature (in terms of signal strength) is from the stroma. 87, 92, 95 Surprising however, is the convergence in t erms of stromal signal strength, approximately equal for both pathologies, despite the fact that the models vary in the particulars of changing the optical properties of the stromal layer and the more superficial layers. Also of note are the extremely low fluorescence signals from the superficial and epithelial layers, essentially biasing the returned information towards the basal epithelium and stromal layers exclusively. Figure 3 11 shows computed fluorescence spectra from each of the three oral tissue ty pes. Similar to previously reported experimental data, all of the spectra are dominated by a combination of collagen and NADH fluorescence, obscuring the weaker signals from keratin and FAD. 87, 90, 92, 95 Further validating this model are the observations of a loss of fluorescence for both the leukoplakia model and the dysplasia model, again characteristic of experimental obse rvations. 3, 9, 17, 20 Finally, a shift of the signal peak is observed as the dominant fluorophore shifts from collagen in healthy oral mucosa to NADH in the dysplastic tissue. Figure 3 1 2 shows a computed DLIPS spectrum from healthy or al mucosa pre malignant, and benign pathologies Due to the low amount of DLIPS signal recovered in

PAGE 97

97 this model, the spectral resolution is limited to bins of 20 nm in size. In a similar way to what is observed experimentally, because of the high degree of sensitivity to changes between measurements across the DLIPS spectrum, it is challenging to visually inspect the spectra and assign features to changes in the tissue optical and fluorescent properties. Previously, decomposition using principal component a nalysis has been shown to provide some insight into the origins of contrast implicit in the DLIPS signatures of contrast in epithelial pathologies, but these relationships did not consider a model of the fluorescent properties of tissue, only the most like ly explanations for the remitted spectral signatures. Wh ile in the fluorescence spectra it is not possible to ascribe spectral regions to each tissue la yer or particular fluorophore, F igure 3 1 2 demonstrates this is possible for DLIPS and further that ther e are unique changes for the pathology models, not just the arbitrary loss of fluorescence observed in traditiona l tissue fluorescence spectra (F igure 3 11 ). In shorter wavelength range of the DLIPS spectrum, the dominant tissue feature should be keratin, and there are dramatic changes in the DLIPS signal for this range between the healthy tissue model and either of the pathology models. This is both because of the specificity of interaction of the perturbing excimer laser with this layer as well as the dra matic changes in fluorescent and morphological properties associated with either the dysplastic transformation or persistent inflammatory lesions. At the long end of the DLIPS spectrum there are also pronounced changes that can be ascribed to changes in ac cessibility to FAD fluorescence information in the intermediate epithelial layer due to the optical clearing effect on the superficial epithelium of the perturbing excimer laser.

PAGE 98

98 Discussion In this paper we presented a Monte Carlo modeling approach to stu dying a new tissue fluorescence spectroscopy technique, DLIPS. The model was based upon previous clinical, histopatholgical and optical spectroscopy studies of both healthy and diseased tissue. By using disease modeling approaches previously used to study tissue fluorescence, instead of developing the modeling uniquely for studying DLIPS, this approach minimizes the risk of confirmation bias in support of previously proposed mechanisms and advantages hypothesized for the diagnostic potential of DLIPS. The M onte Carlo model presented here was motivated by a need to explore three fundamental phenomena related to DLIPS tissue spectroscopy. First we explored the supposed biochemical changes induced by 193 nm laser perturbations and the impact on the remitted fl uorescence signal. Second we connected in vivo observations of previous DLIPS studies with biological changes associated with both diseased and healthy tissue, noting the biochemical differences and the changes in the associated DLIPS spectra. And finally we assessed what features of pathology DLIPS preferentially enhances as opposed to fluorescence spectroscopy alone. As with any model, this represents an imperfect approximation of the reality of DLIPS and fluorescence tissue spectroscopy. However, as m uch of the model was developed around actual measured tissue optical properties, with pieces validated in other peer reviewed studies, it is presumed to at least show relationships within the model. This includes exploring how signals change in different p athological states as well the tissue source s of signal for DLIPS and fluorescence spectroscopy. The notable deviations from actual tissue are catalogued and explained further here. The first is the simplification of each tissue layer to only be represente d by a single major fluorophore.

PAGE 99

99 In reality, there is likely some cross over between layers, but in other cases experimental evidence has shown that these fluorescent molecules are not present in a particular tissue layer. In the former case, for any layer that contains actively dividing cells, NADH and FAD are likely to both be present as fluorophores 92 But in the latter case, IHC stains have shown that there is no extracellular collagen pres ent in the epithelial layer of the oral mucosa above the basement membrane. 29, 98 In other studies addressing this topic, it was found that the fluorescence from individual tissue layers was dominated by a particular fluorophore, although the dominant fluorophore could change as a result of pathology. 92 This effect is modeled in this system and is based upon previously reported findings for oral dysplasia and chronic inflammation. Another limitation is the approximation of layer boundaries as flat interfaces. In reality these interface can range from small undulation to regions of repeated large invagination between the layers to outright breakdown at the boundary whe re the interface is not well defined. 92 These effects have previously been modeled for skin Monte Carlo simulation and it was shown that the effect is essentially to increase the scattering of light at the boundaries, reducing the amount of incoming light that penetrates these layers and reducing the amount of remitted light. 99 Since these boundary conditions are likely to change for a given pathological state, this could impact differential signals when using optical parameter estimation as the source of pathological contrast but is unlikely to have a significant eff ect when fluorescence is used. The final two important approximations deal with the treatment of 193 nm light for Monte Carlo simulations. Due to the limited number of studies available estimating the

PAGE 100

100 tissue optical properties in this wavelength range, som e had to be inferred from a higher wavelength range, like the refractive index. It was assumed to be constant over all wavelengths in this model but that is likely not the case at 193 nm. This could dramatically impact the amount of specular reflection at the surface of the oral mucosa and have a significant impact on the penetration depth and photochemical activity. The final generalization was regarding dynamic updating of the absorption coefficient during the modeling of the excimer perturbation process. Photolysis not only affects fluorescence and absorbance properties of tissue at 355 nm, but also at 193 nm as the cleaved bonds are the source of the high absorbance cross section for this wavelength. Figure 3 7 shows this effect is not trivial and in fac t saturation of the bond breaking process is observed in this model for the most superficial tissue when the model is extended to the full photon flux used experimentally, about 2.4 x 10 17 photons. This type of process is tough to implement in a Monte Carl o simulation as it propagates only a single photon at a time. It could potentially have a sizable impact on the penetration depth for the observed photochemical changes, although it is unclear what the implications would be for the DLIPS signal. Despite t hese limitations, this work represents a step forward for tissue optics modeling using Monte Carlo methods due to the incorporation of photolytic effects and further this framework can serve as a virtual testbed for optimizing the DLIPS measurement configu ration to recover more DLIPS signal or explore how to access more information in the tissue using DLIPS. The latter is considered through the remaining discussion of results.

PAGE 101

101 In F igure 3 6 the remitted fluorescence is considered with respect to the fluor escence generated in that layer and the amount of photons absorbed. Not surprisingly, stromal absorbance is the dominant effect in terms of photons deposited because this is the first tissue layer where hemoglobin is present. This effect is magnified in th e pathological case because of the increase vessel density due to local angiogenesis as the emerging neoplasm establishes a hospitable environment for growth. In terms of fluorescence however, emission from NADH tends to provide a nearly equal amount of fl uorescence, with the relative levels varying widely between the disease states. From this it is also clear that any changes associated with superficial and intermediate epithelial layers are overshadowed by the signals from collagen and NADH. There is howe ver potentially valuable information about the pathological state contained in these layers that needs to be assayed using a different technique, like DLIPS. Based upon this fact that the relative levels of NADH and collagen fluorescence remittance change so dramatically between the three states of oral tissue modeled here, this would seem to be an appropriate biomarker for detecting pathology. This is problematic at an excitation wavelength of 355 nm, as the spectral overlap makes decomposition of the two signatures hard. 87 Although the DLIPS technique is supposed to remove inter subject sources of variability, it is extremely sensitive to measurement noise, due to the differential nature of the signal while keep th e same noise level associated with the individual fluorescence measurements. As a consequence, both the experimentally collected spectra and those derived from this model show a large noise level relative to the size of the signal level. It is likely this can be corrected experimentally by making repeat pre and post

PAGE 102

102 perturbation measurements and averaging those to reduce the noise level somewhat. By looking at the relative stability of the per layer DLIPS signals over repeated measurements, this confirms t hat by collecting more DLIPS measurements, a more stable measurement can be made. The other possible approach is to increase the amount of perturbation realized between fluorescence measurements, presumably increasing the differential signal. In F igures 3 9 3 10 3 11 and 3 1 2 the sources of contrast between the healthy cheek and the two model pathologies are explored. This is the most important information recovered from the model because it allows for the formulation of a biological theory supporting a ny detection contrast observed between these states experimentally and also can suggest if the information contained in the fluorescence spectrum is redundant with the DLIPS information or if they are unique and complimentary. In F igure 3 9 the largest si gnals are derived from the top two epithelial layers for DLIPS, which is in opposition to the fluorescence levels in F igure 3 10 which heavily emphasize basal layer and stromal fluorescence. This may be an important source of contrast inherent to the DLIPS technique because research has shown that quantitative measurements of keratin fluorescence and density from histopathology can provide early diagnostic biomarkers; despite often presenting clinically in both benign and dysplastic disease as hyperkeratosi vivo. When considering the difference in signals between the healthy and pathological states, this fluorescence contrast for detection clearly favor the basal layer (NADH) as the changes to collag en are approximately the same (F igure 3 11 ). This figure further confirms that the FAD and keratin signals are essentially washed out as noise for

PAGE 103

103 fluorescence spectroscopy, but when considering the DLIPS spectra in F igure 3 1 2 all of the epithelial layers provi de contrast, while the stromal layer does not show much difference from healthy tissue, or a clear differentiator between benign and dysplastic pathology. This is somewhat surprising for fluorescence spectroscopy as one of the earliest pre neoplastic chang es is collagen remodeling near the basement membrane. 75 This could however also occur in benign chronic inflammation as immune cells invade the upper stroma and basal layer of the epithelium. Biomolecular changes of collagen were not considered here but there is evidence to support the presence of at least three unique types of collagen found in the epithelial stroma based upon fluorescence emission. 10, 25, 28 The relative levels of these three collagen structures could provide the contrast between different pathologies for fluorescence as well as provide a unique signature in the DLIPS spectrum. What was unexp ected in the DLIPS signature was the contribution due to FAD fluorescence, particularly since the emission is so weak at 355 nm excitation. 29, 89 Since the DLIPS is a differential technique, small signals are accessible assuming they are not completely buried in noise. In this case FAD can be an important marker of cellular metabolism, particularly as cell s undergo the dysplastic transformation and leveraging this as a source of contrast directly couples the DLIPS spectrum to a cellular process underlying pathological changes. Accurate modeling of tissue spectroscopy is a complicated undertaking, but can be simplified by using iterative methods and only considering the most important tissue factors impacting it. Adding to that the layers of complexity inherent in addressing disease and population variability makes it even tougher to extract a working unders tanding of the biochemical basis of tissue spectroscopy. Using experimental

PAGE 104

104 studies to derive site specific and dise ase specific optical properties mitigates these uncertainties and improves the confidence in associating model details and experimental DLIP S and fluorescence spectroscopy observation with biochemical and morphological changes. Still this work motivates and directs future experimental work to optimize the DLIPS technique for clinical detection as well as how to further explore the fundamental mechanisms which support using this technique to differentiate dysplastic pathology.

PAGE 105

105 Figure 3 1. Simulation workflow for modeling DLIPS fluoresce nce tissue spectroscopy. Blue indicates model outputs used to relate remitted light signal characteristics w ith aspects of pathology or the DLIPS collection scheme

PAGE 106

106 Figure 3 2 Tissue optical properties for layer specific modeling

PAGE 107

107 Figure 3 3 Modeled layer specific optical properties. A) A bsorpti on. B) Scattering. C) Fluorescence. Measurements for fitti ng the model were found in multiple literature sources as noted in the text (c) (b) (a) A B C

PAGE 108

108 Figure 3 4 Bland Altman plots comparing the total remitted light from the experimental Monte Carlo model versus a previously validated Monte Carlo system. Each system was validated using two sets of model parameters. The blue line represents the bias between the two models and the red lines are one and two standard deviations from the bias value, a measure of the limits of agre ement

PAGE 109

109 Figure 3 5 Fluence (cm 2 ), in logarithmic units of phot ons through the tissue models. A) H ealthy bu ccal tissue. B) D ysplastic oral mucosa. C) B enign oral pathology. Horizontal line s denote boundaries and vertical lines indicate width of the coherent excitation source. (a) (c) (b) A B C

PAGE 110

110 Figure 3 6 Relative per layer absorption and fluorescence emission. A C) A bsorption of photons, generation of fluorescence, and the layer effici ency for the healthy, benign, an d dysplastic oral epithelium m odels respectively. D F) F luorescence generated from each layer as compared to the amount of that light that is remitted from the surface of the tissue to be collected. (a) (f) (e) (d) (c) (b) A B C D E F

PAGE 111

111 Figu re 3 7 Fluence (cm 2 ) of 193 nm light (logarithmi c scale) through the superficial epithelium

PAGE 112

112 Figure 3 8 Photoc hemical events stimulated by 19 3 nm irradiation as a fraction of the total number of incident photons

PAGE 113

113 Figure 3 9 Relative contributions to the remitted DLIPS signal from each layer of epi thelial tissue

PAGE 114

114 Figure 3 10 Relative contributions to the remitted fluorescence signal from each layer of epithelial tissue

PAGE 115

115 Figure 3 11 Simulated fluorescence spectra for the three tissue models

PAGE 116

116 Figu re 3 12 Binned spectral DLIPS data from Monte Carlo simulations. Each bin is 26 nm in width and the values represent the summed DLIPS signal over that range (Averaged across three simulations)

PAGE 117

117 CHAPTER 4 DETECTION THEORY ANALYSIS OF DLIPS BASED SENSING OF ORAL DYSPLASIA: IMPLICATIONS FOR CLINICAL INST RUMENT DESIGN AND TESTING Motivation Enhanced, early detection of oral pre malignancies is well established as a critical component of strategies to improve the 5 year survival rate. 17 Historically, screening patients for oral dysplastic lesions has relied upon a convention al tactile and visual exam, with diagnosis con firmed via biopsy. 3 Over the past 20 years, several adjuvant technologies have e merged to address the problem of better screening for oral dysplasia or benign pathologies. 13 Despite these advances in technology, it has yet to be demonstrated that use of any of these adjuvant techniques results in a meaningful reduction in morbidity and mortality. 13 The fundamental fl aw with these devices is a lack of specificity of the technique for changes associated with cancer progression. Instead these devices are developed and trained to provide contrast for obvious pathological conditions, but are then applied in the clinic to s creen for pre disease state changes. 16 This pre supposes that the earlier pre malignant stages of disease resembl e advanced disease Unfortunately, early pre cancerous lesions are generally subt le and do not resemble advanced dysplasia or oral malignancies. 3 In addition, pre malignant lesions are highly heterogeneous in presentation, often mimicking common benign pathologies. 3 There is also a growing body of evidence that indicates many oral pre malignancies are not detectable to the unaided eye at the earliest stages 3 New technologies to detection oral dysplasia must consider these issues from the fi rst generation of diagnostic adjuvants, with strategies to specifically assay biomarker of early disease.

PAGE 118

118 Here we present a new biosensing technique Differential Laser Induced Perturbation Spectroscopy (DLIPS) The technique combines sequential fluorescenc e probing (pre and post perturbation) with sub ablative UV photochemical perturbation and difference spectroscopy to provide a new spectral dimension, faci litating two improvements over fluorescence spectroscopy. 36, 37 First the differential technique eliminates significant variations in absolute fluorescence response within subject populations. Second, UV pertur bation directly couples th e DLIPS response to the superficial biomolecular structure. 38 Further we develop a pre clinical animal model framework for testing this device, as well as conventional fluorescence biosensing, which better approximates the observation of both benig n, dysplastic, and healthy tissue in the clinic. The goal of this study was to apply the DLIPS fluorescence technique to specifically detecting an earlier and often ambiguous state of pre malignant oral l esion development in the presence of benign patholog y, a well known confounding factor in clinical detection. Unfortunately, due to low signal to noise levels for the collected DLIPS spectra, it was not possible to conclusively evaluate the performance of this technique in this study. However, despite these negative findings with DLIPS, the structure of the study facilitated critical evaluation of an underlying iss ue in oral cancer detection: there may be a principled way to design detections systems as well as training and clinical testing to improve diagno stic capabilities. Because of the di sappointing performance of commercial ly available devices for detecting oral neoplasia and pre malignant le sions and that critical analysis of these devices indicates a lack of testing and optimization prior to reaching the marketplace,

PAGE 119

119 here we present an experimental framework for an optimal detection theory approach to analyzing the detection performance of these devices. Specifically, using the hamster cheek pouch model for dysplastic lesion development as well as ben ign lesion development and healthy oral mucosa, the ambiguity of the clinical scenario was roughly approximated. Using laser induced fluorescence (with no pre defined detection algorithm) as a general surrogate for all fluorescence and light based technolo gies in the marketplace, the framework is constructed to evaluate one specific question endemic to all of these devices: If the goal is to find oral cancer and reduce the false positive rate, how can the detection problem be framed to optimize these outcom es? By applying optimal detection theory to several different embedded detection problems and comparing those detectors with currently available analogs, insight into how detector engineering, algorithm training, and clinical testing may all contribute to improved diagnostic performance at the point of c are is provided. T his model shows that by re framing the detection problem in terms of detecting pre malignant lesion over benign lesions (as opposed to the more commonly used healthy tissue), substantial ga ins can be realized in terms of enhanced accuracy and anticipated improved clinical performance as this strategy directly addresses the most common confounding source of false positives. Materials & Methods Chemically Induced Cheek Pouch Pathology Male Gol den Syrian hamsters, ten in total were used in this study. Pre cancerous dysplasia was induced by application of 7,12 dimethylbenz(a)anthracene (DMBA) (Sigma Aldrich, St. Louis MO) in acetone (Fisher Scientific, Pittsburgh PA) at a concentration of 0.5% w/ w to the right cheek pouch of five hamsters. Application was

PAGE 120

120 repeated three times weekly for up to nine weeks. The DMBA hamster cheek pouch model and the application timeline were selected because of previous work showing that it closely resembles the dysp lastic transformation sequence to malignancy observed in the human oral cavity. 96, 97 Particularly, by eight to nine weeks, the p ouches show definite signs of the dysplastic transformation leading to cancer, without any malignancies which often take up to 20 weeks to form in this model. 58 64 The left cheek pouch was coated three times weekly for the same time pe riod with acetone alone to serve as an internal control. Five additional hamsters had measurements made in both cheek pouches prior to and immediately following four days of applying 0.5% SLS in DMSO in order to induce benign pathological changes. 100, 101 None of the hamsters in this study displayed signs of systemic toxicity during the study and all animal protocols were approved by the University of Florida Institutional Animal Ca re and Use Committee. Collection of DLIPS and Fluorescence Spectra Spectra were collected using a fiber optic probe coupled to a DLIPS system as previously described ( Chapter T wo). 38 The fiber probe (Ocean Optics, Dunedin FL) used in these experiments consists of a ring of seven angled solarization resistant fibers to direct both the 193 nm perturbation beam and the 355 nm excitation beam toward the sample. Betwe en fiber tips and the tissue is a 2.0 mm thick window of UV grade fused silica, to permit relatively undiminished transmission of the excimer laser light. Fluorescence emission from the tissue is collected with a single center fiber. Using this configurati on, the excimer perturbation pattern the same size or larger than the surface area sampled for emitted light. The emission is then focused from the fiber probe onto a fiber optic bundle using a combination of two 50 mm diameter UV grade lenses, where

PAGE 121

121 it is passed to a 0.3 m Czerny Turner spectrometer and recorded with an intensified CCD (ICCD) array detector. Prior to entering the fiber optic, two sharp edge filters are used in series: a high pass filter designed to block the residual 355 nm light, and a l ow pass filter designed to block any residual 532 nm light remaining from the frequency tripling process. This allows spectra from the tissue to be collected from about 390 nm to 500 nm. At weeks eight and nine of the study, all five of the hamsters had l aser induced fluorescence measurements made in each of their cheek pouches for a total of ten measurements in control tissue and ten measurements of dysplastic tissue. Prior to application of the SLS solution, measurements were made in each cheek pouch of the five animals in this group to give ten additional control measurements. After the four day course of SLS treatment, these hamsters again had measurements made in each cheek pouch, serving as measurements of benign pathological changes. Before each meas urement, the probe was wiped clean with methanol and rinsed with water, and the pouch was inspected for debris. Hamsters were anesthetized using a ketamine/xylazine solution (Phoenix Pharmaceutical, St. Joseph MO; Lloyd Laboratories, Shenandoah IA) at 10 ml/kg of body weight. The fiber optic probe was inserted into the pouch of the anesthetized animal and a fluorescence spectrum was recorded across a 1 mm diameter tissue area using a 200 shot ensemble average recorded at 5 Hz laser repetition rate (F igure 4 1). Immediately following this pre perturbation fluorescence measurement, the 193 nm excimer laser was used to deliver a total of 2500 perturbation pulses to the target spot with laser energy of 100 J/pulse (fluence = 2 mJ/cm 2 ) and a laser repetition r ate of 50 Hz. Approximately 30 seconds

PAGE 122

122 after perturbation, a second fluorescence spectrum was recorded using an identical 200 shot ensemble average to collect a post perturbation spectrum. These two spectra were then subtracted to generate the difference spectrum at each location, with the difference divided by the pre perturbation spectrum to generate the DLIPS response for the given surface site. Histopathology After week nine, the DMBA treated hamsters were euthanized and the cheek pouches were excised and fixed in formalin (Sigma Aldrich, St. Louis MO). Likewise, immediately after measurements were made in the SLS cheek pouch, that tissue was harvested and fixed. Fixed samples were sectioned at 5 um and stained with hematoxylin and eosin (H&E). Tissue preparation was performed at the Cell and Tissue Analysis Core, McKnight Brain Institute, University of Florida (Gainesville FL). The sections were read by a board certified oral pathologist, and diagnosed based upon established criteria for dysplasia. 12 Dimensional Reduction and Subspace Analysis During the collection of spectra, the shape and signal strength were actively monitored, and measurements were rejected and repeated if the fluorescence measurement was well out of the expected range (approaching an order of magnitude increase or decrease). In these cases, the probe was re cleaned, the cheek of the hamster was wiped and the probe was repositioned to take a nother measurement. In another case, the root cause behind a set of low signal to noise measurements turned out to be low excimer laser fluence, so the entire set of measurements was repeated. Following this process, the spectra (pre perturbation fluoresc ence and DLIPS) for all four groups were imported into MATLAB (The Mathworks, Natick MA) for

PAGE 123

123 processing. Difference spectra and fluorescence spectra were preprocessed using a twenty five point moving average (a range of less than 5 nm in wavelength) filte r to remove acquisition noise. Fluorescence and DLIPS spectra were then analyzed separately, but using parallel methods, to facilitate side by side comparison at each analytical level. Spectra from the treated and control groups were grouped together and decomposed using principal component analysis (PCA) for up to ten components for each technique. Clustering within the data sets was visualized using score plots of the first two components and the validity of using the PCA basis for further analysis was determined by considering both the percent of variance explained by each component and a one way Analysis of Variance (ANOVA) in each component to test for statistically significant differences between the three experimental groups. Individual group dif f erences were assessed using Tu key HSD post hoc analysis, and two groups were considered different at a 95% confidence level. To consider the different clinical testing, algorithm training, and contrast mechanisms possible for DLIPS based diagnostic devices or fluorescence based devices, data from the three experimental groups (dysplastic pathology, benign pathology, and no pathology) was clustered into six binary detection groupings. For each of these six binary groupings, a transformation of the spectral d ata was performed to reduce the dimensionality of the spectrum down into the optimal one dimensional subspace for that binary detection problem. This was done by applying th e change of basis described in E quations 4 2 and 4 3 using Fisher's Linear Discriminant, B, where H 0 and H 1 represent the two data classes being used to construct the linear transform, mu is the mean vector for each data set and sigma is the covariance matrix. 102 This

PAGE 124

124 maximizes the separation between the two classes while at the same time reducing the dimensionality of each observation from y x to b x A limitation of this approach is that the dimensionality of the original data s et must be less than or equal to the smallest number of observations with a group. 102 In this study, the dimensionality of the original spectral data was reduced to 10 dimensions (the number of observations in each experimental Discriminant. Using this framework, the signal to noise ratio (SNR) for detection in that basis wa s computed using E quation 4 4 below and as a reference the SNR prior to the transform was also calculated by substituting out Fisher's vec tor B for a one's vector, J (E quation 4 1). 102 (4 1) (4 2) (4 3) (4 4) While the change of basis transformation and the SNR are only dependent upon the two data groups se lected to compute the transformation, the transform can be applied to the entire data set, independent of class association or use of the data when computed the transformation, which occurred for this study.

PAGE 125

125 Classification Framework and Detection Performan ce To assess the appropriateness for detecting oral pre malignancies, benign pathology or healthy oral mucosa in each of the six detection bases, a maximum a posteriori (MAP) test was used to predict the class associations. 103 In this case, for each measurement taken, class association was predicted by selecting the most probable class, as estimated from the a posteriori probability of that measurement coming from each class (ie: a test of the likelihood of each class given the observation). 103 The conditional probab ilities of each of the 30 observations were estimated by a multi parametric fit of mean, standard deviation, skewness, and kurtosis to find the most appropriate parametric distribution for the observations from that class. The Johnson System of distributio ns was used to for fitting the conditional probability distribution and estimates to the parameters (mean, standard deviation, etc.) were made directly from the observations using a maximum likelihood estimation least squares approach. 104, 105 In total, three condi tional probabilities, P(y|H x ), were estimated in each basis, supported in all cases by ten observations of that class. Equations 4 5 and 4 6 below summarize the MAP classification criteria for the three groups. (4 5) (4 6) Where P(H x |y) represents the a posteriori probability of class H x given the observation y, and P(H x ) is the prior probability of that class occurring in the population. 103 This test was used to assign predicted class to the data points in all cases.

PAGE 126

126 To explore the improved discrimination in each of the six bases, a simple one way ANOVA with Tukey HSD post hoc analysis (as above for the PC transformed data) was performed on the transformed data to determine if the groups were statistically sign ificantly different following application of the Fisher transformation. For each of the six detection bases, a confusion matrix of the three class problem was constructed. Both row and column statistics within the confusion matrix were computed, as analog s of sensitivity and specificity, for the multi class problem. The metrics used were Precision, the fraction of observations that were correctly assigned to that class, and Recall, the fraction of the members of a particular class correctly assigned to it. 102 To collapse these values into a concise representation of the performance of DLIPS and fluorescence to accurately detect each disease state, the over all Accuracy of correct classification was computed. Unfortunately, for multi class problems there is no well established framework for describing the detection performance across all decision thresholds (ie: ROC curve and AUC), and as a consequence does n ot permit tuning of the detection threshold to explore the optimal operating points and compare this to estimated operating points for clinical devices. 106 Results Histopathology Figures 4 3 ( A C ) are representative photomicrographs of the H&E sections of hamster cheek pouches fro m th e control group SLS treated gr oup an d DMBA treated gro up respectively The completely untreated cheeks and those treated with carrier (acetone only) for eight weeks appeared histologically indistinguishable, and a representative s ection of control chee k i s shown in F igure 4 3 (A) The epidermis was of normal thickness and consistent throughout the sections. The layers were well defined,

PAGE 127

127 with a thin, consistent layer of keratinization on the surface and a clearly define d, smooth basement membrane layer separating the basal cells and the lamina propia. Figure 4 3 ( B ) shows the cheek pouch after four days of SLS application. These pouches displayed moderate epithelial changes including hyperkeratosis, acanthosis, and hyper granulosis. These finding were consistent with other studies using this a model of inflammatory benign pathology. 100, 101 Visual and tactile examination of pouches prior to excisio n showed evidence of increased toughening of the oral mucosa, as assess by some loss in elasticity and a cloudiness beginning to obscure visualization of the sub epithelial vessels (F igure 4 2). Figure 4 3 ( C ) shows a section from a cheek pouch after 8 to 9 weeks of DMBA application. This particular section is through a verrucoidpapillary mass, consistent with pre neoplastic lesion found in human with a high incidence of transforming into an invasive carcinoma. 107 Obvious morphological dysplastic characteristics of this section include: extreme hyperkeratosis, drop shape d rete ridges, abnormal epithelial maturation, and a loss of cellular cohesiveness, in addition to cellular and nuclear changes. Visual and tactile exam showed a low incidence of papillomas (less than one per pouch) and no obvious ulcerations in the pouch despite clear visual evidence of oral pathology including large patches of leukoplakia and small les ions resembling erythroplakia (F igure 4 2) as well as a clearly abnormal stiffness associated with this generally elastic tissue. Spectral Analysis Figure 4 4 shows the average fluorescence spectra and Figure 4 5 shows the average DLIPS spectra of the DMBA treated cheek pouches versus SLS treated and the control pouches. The fluorescence spectra collected from the cheek all resemble

PAGE 128

128 each other in shape but ar e scaled differently for the three groups. The most prominent spectral features are a peak around 460 nm as expected from NADH, but also a surprising peak at 395 nm. Despite the averaging, the spectra are quite noisy, likely due to the low levels of light collected through the collection fiber of the probe as well as the added noise as the light propagates through the fibers to the detector. The DLIPS spectra on the other hand, show both a change in signal intensity as well as a change in shape between grou ps; despite the apparently low sig nal to noise ratio. T he control and dysplasia spectra resemble one another in shape but vary greatly in signal intensity. The SLS spectrum, while overlapping the control spectrum, shows pronounced shape difference above 43 0 nm unique from the other groups. Subspace Analysis Figure 4 6 shows the percent of variance explai ned by each principal component. Correspondingly, Figure 4 7 shows identical analysis of the DLIPS data. In both cases 30 total spectra were used, 10 from t he DMBA treated hamsters, 10 from the SLS treated hamsters, and 10 from the acetone treated hamsters. In addition to assessing each component in terms of the amount of total data variance encoded in that dimension, an equally important metric is to determ ine whether the variance is specifically representative of the between group variance (ie: the particular principal component dimension preserves of enhances difference between observations of dysplastic tissue versus benignly inflamed tissue versus health tissue). Using one way ANOVA, it was discovered that the group means are significantly different in components one, two, and three for the fluorescence measurements but only in components two and eight for the DLIPS measurements. To further explore the d ifference between the groups, Tukey HSD post hoc analysis was performed. For

PAGE 129

129 fluorescence measurements, differences between dysplastic and benign pathology data dysplasti c groups were significant under PC three. For DLIPS measurements, differences between control and benign pathology data groups were significant under groups. Differences b etween benign pathology and control groups were significant under PC eight. The lack of a convergence in the principal component space in terms of finding a PC dimension that both describes a large proportion of the variance as well as having significant d ifferences between the three measurement groups indicates that the PC basis is not an appropriate basis for describing the inherent differences between groups that are captured when making either fluorescence or DLIPS measurements The studies in this work however are not designed to evaluate whether this is inherent to the particular of the detection problem, technique, or the noisiness of the data. To find a more appropriate basis for describing the differences between the three groups, six transformation s for both fluorescence and DLIPS were constructed from the collected cheek pouch measurements. Figure 4 8 shows the six binary classification strategies possible for the three class problem. In particular the data contained in these six partitions formed the basis for the transformation into the optimal detection dimension using Fisher's Linear Discrim i nant (FLD). It should be noted however, that this change of basis is only optimal for the two gro ups being compared in columns H 0 and H 1 there is no condit ioning for a third data class. Further, in each of the six scenarios, the SNR reported both pre and post transformation only represent s the distance between groups H 0 and H 1 In comparing the DLIPS data and the fluorescence

PAGE 130

130 data, the fluorescence data has a higher SNR for all cases, except when comparing in the basis of benign lesions versus healthy oral mucosa; a rather trivial basis for comparison or construction of a detection device. For fluorescence, the largest values of SNR always come from a detect ion basis that has the dysplastic tissue as a specific point of contrast with respect to the rest of the data, not unlikely previous finding where fluorescence spectroscopy has been shown to have a high sensitivity and specificity for detecting oral pre ma lignancies against healthy oral mucosa in the hamster cheek pouch. 97 Further, based on analysis of the histopathological section of the dysplastic cheek pouches, a large change in the optical and fluorescent properties should be expected. For the transformations of the DLIPS data however, all of the SNRs after applying the transformations are fairly consistent. This may indicate that the DLIPS measurements are more balanced in detecting points of contrast among all three of the tissue groups, or this may indicate that high noise levels in the experimental measurements are obscuring any dramatic contrast between the three target tissue types. Unlike the SNR, an ANOVA test accounts the differences between all three of the group means but unfortunately only considers if at least one mean is statistically different from the others. Figures 4 8 and 4 9 show the p values for an ANOVA test of the three groups under each transformation for fluorescence a nd DLIPS measurements respectively. As expected, in all cases at least one group measure was found to be significantly different from the other because the nature of the FLD transformation is to maximize the distance between two groups while minimizing the variance within each of the two groups. Ideally however there exists an FLD transformation that also preserves

PAGE 131

131 or enhances the differences between the three groups. This was not discovered in any of the cases for either fluorescence or DLIPS and this agai n could be due to the high noise levels in the original measurements or that a one dimensional projection of this data is insufficient to optimally describe the differences between the three groups as well as similarities within groups. If the goal of th e technique is to discriminate pre malignant pathology from anything else then only the transformations that demonstrate this ability should be considered as options to explore further for the details of classification performance. These are hig hlighted in yellow in Figures 4 9 and 4 10 as post hoc analysis showed that under these particular transformations both the benign pathology and the healthy oral mucosa appeared different from the dysplastic tissue Measurement Classification Figures 4 11 through 4 1 3 correspond to the confusion matrices from transformat ions T1, T2, and T4 in Figure 4 8 for the fluorescence measurements. Figure 4 14 is the confusion matrix fo r transformation T4 in Figure 4 8 for the DLIPS measurements. These are the highlight ed transf orma tions in Figures 4 9 and 4 10 respectively. Despite not meeting the post hoc analysis criteria for utility as a detection basis when making DLIPS measurements, the confusion matrix for transformation T2 is shown as well because of the high classificati on accuracy (Figure 4 15) T he best detection performance s across all three classes (dysplasia, benign pathology, healthy oral mucosa) were in bases where the dysplastic tiss ue was isolated as one of the two grouping for constructing the change of basis (h ighlighted in Figur e 4 9 ) While each of these bases for detection showed excellent performance in finding dysplasia (100% Precision and Recall) the basis optimized around comparing benign

PAGE 132

132 pathology and dysplastic tissue had the highest overall acc uracy a t 83%. This implies that for this model, dysplastic tissue has a very distinct signature because these numbers do not reflect the performance predicted from commerci ally available systems 3, 13 Another possibility is that these algorithms that consider the benign pathology as well as healthy oral tissue in contrast to dysplastic tissue are uniquely suited to diagnostic tissue fluorescenc e. For classifications using the DLIPS measurements, the performance in terms of finding dysplasia was overall worse than using fluorescence alone but not by a lot. Under transformation T4, the Precision was 80% and the Recall was 89%. Overall the Accura cy was only 60% as the DLIPS technique tended to confuse benign pathology and healthy oral mucosa. Under transformation T2, the Precision was 100% and the Recall was 91% for finding dysplasia and the overall accuracy was 87%. While this overall accuracy is higher than the number reported for using fluorescence alone, the metrics for finding dysplasia are marginally worse. I t remains to be seen if these numbers accurately represent the performance of using the DLIPS technique to detect oral dysplasia or mere ly point out the increased sensitivity of this technique to measurement noise. Discussion Despite the shortcoming of this study in providing support for the DLIPS technique in the detection of oral cancer, a more fundamental question with regard to optica l detection of oral cancer was addressed: What is the correct detection problem to target devices towards in order to improve diagnostic accuracy and reduce the false positive rate.

PAGE 133

133 The extremely low values for the fluorescence spectra and the DLIPS spec tra (F igure s 4 4 an d 4 5 ) are likely related to a lack of fluorescence light returning through the fiber probe to the detector. As shown previously in Chapter 1 the DLIPS signal must overcome an order of magnitude more measurement noise than the individua l fluorescence measurements so in this case since the fluorescence levels were so low, the DLIPS signals were almost completely buried in noise. The large error bars and 25 point moving average associated with F igure s 4 5 support this notion. As a result o f this noise, pathology classification us ing the DLIPS data (Figures 4 14 and 4 15 ) was uniformly worse than when using fluorescence alone. Despite this, the results presented here indicate that by solving this signal to noise problem for the DLIPS system, improved detection performance should be possible. Figure s 4 11 to 4 13 show that dysplasia can be detected with 100% accuracy and no false positives when using fluorescence detection While this is a great result in terms of the overall goal for oral ca ncer detection devices, these results are not consistent with what is typically reported for clinical measurements. 3, 13 Future work a pplying optimal detection theory to the improvement of oral cancer diagnostic devices should address this, by either relaxing the criteria for consider tissue pre malignant, thereby modeling even earlier stages of disease, considering multiple benign patho logy, or by simply increasing the number of measurements/ test subjects. The overall best performance for finding dyspl asia was reported in Figure 4 11 for the classifier constructed by emphasizing the differences between benign pathology and dysplasia. It is possible that detecting benign pathology over dysplastic pathology is an inherently harder detection problem than discovering healthy tissue over dysplasia and

PAGE 134

134 by optimizing on that basis, the need to consider contrast with healthy tissue is obviate d. Typically, systems are trained or inherently designed to differentiate between healthy tissue and dysplasia. Take for example the Velscope system which uses loss of fluorescence as its biomarker of oral cancer and pre malignancies. 17, 20 Since the system relies upon comparative visual inspection with the remainder of the (presumably healthy) oral mucosa to detect this LOF, this device inadvertently operates in the detection regime of dysplasia versus benign tissue; although there is no optimization of the detection scheme in this case, it is simply the detection criteria being employed for th is device. So there is little consideration given to ability to uniquely differentiate benign pathology from dysplastic tissue and as such these devices commonly detect the LOF from leukoplakias and other lesions as cancerous when they are not. 13, 16, 17 But is it f easible to train and test devices for the basis comparing benign pathology and dysplastic pathology? Pre clinically, this is not problematic (as demonstrated here) but clinically, it could present a hurdle, as a gold standard for the evaluation of benign p athology (ie: biopsy) is not always available. That is because clinical testing protocols calling for biopsy of oral tissue that would not otherwise require biopsy is considered unethical. 3 Therefore other experimental means must be sought to realize the potential for enhanced detection capabilities by designing instruments to focus o n the harder, but more clinically relevant problem of sorting benign lesions from dysplastic lesions. This could include continuing to develop more accurate animal model of human oral pathology or attempting to better understand the feature based sources o f contrast within each type of oral tissue and design instruments around those

PAGE 135

135 detection requirements, visualizing the differences between benign pathology and dysplastic lesions. Even though the ultimate priority for tuning these instruments is detect ion of cancerous and pre malignant lesions above all other classes (presumably lumped together), here we demonstrate that by rethinking the methodology for training and testing these systems; instead optimizing for contrast between dysplastic and benign le sions, improvements in detection accuracy can be realized. By using this basis for contra st it is further expected that the rate of false positives due to benign lesions could be reduced and the specificity of these devices for finding oral pre malignancie s could be enhanced but it is unclear from these results as no false dysplasias were detected for transformations with th e best performance (Figures 4 11 to 4 13 ).

PAGE 136

136 Figure 4 1 Fiber probe used for DLIPS and fluorescence measu rements in the hamster cheek pouch

PAGE 137

137 Figure 4 2 Pictures of everted cheek pouches displaying the model p athologies used in this study. A) H ealthy oral mucosa. B) DMBA treated buccal tissue displaying a clear thickening and loss of elasticity, char acteristic of the pre malignant transformation. C) DMBA treated buccal tissue displaying flat regions of leukoplakia as well as a small velvety eryt hplakia lesion. D) SLS treated oral mucosa displaying a smoother, thicker layer of epithelial tissue and a l oss of elasticity. A B C D

PAGE 138

138 Figure 4 3. H&E stained histology sections of cheek pouches. A) C ontrol hamster (no treatment or carrier alone). B) SLS treated hamster. C ) DMB A tr eated hamster at week 8 or 9. A ) shows healthy, well organized section of superficial keratin, e pithelium, and lamina propia. B ) shows obvious signs of irritation. C ) is specifically a section through a verrucopapillary mass, a clear sign of dysplastic transformation 2 5 0 u m ( a ) 2 5 0 u m ( b ) C B A 250 um 250 um 250 um

PAGE 139

139 Figure 4 4 Average fluorescence spectra from 10 SLS treated benign irritation cheek pouches (blue), 10 carrier only treated control cheek pouches after 8 or 9 weeks of application (green), and 10 DMBA treated dysplastic cheek pouches a fter 8 or 9 weeks of application (red)

PAGE 140

140 Figure 4 5 Average DLIPS spectra from 10 SLS treated benign irritation cheek pouches (blue), 10 carrier only treated control cheek pouches after 8 or 9 weeks of application (green), and 10 DMBA treated dysplastic cheek pouches after 8 or 9 weeks of application (red)

PAGE 141

141 Figure 4 6 Percent of variance explained by each of the first ten principal components, p values for the ANOVA between the three experimental groups (in each principal component), and post hoc anal ysis to determine which group means are significantly different for the laser induced fluorescence data

PAGE 142

142 Figure 4 7 Percent of variance explained by each of the first ten principal components, p values for the ANOVA between the three experimental grou ps (in each principal component), and post hoc analysis to determine which group means are significantly different for the DLIPS data

PAGE 143

143 Figure 4 8 Data groups used to derive the six FLD transformations as well as the corresponding SNR between the H 0 an d H 1 data group pre and post transformation

PAGE 144

144 Figure 4 9 ANOVA and Tukey HSD post hoc analysis of the transformed fluorescence measurements

PAGE 145

145 Figure 4 10 ANOVA and Tukey HSD post hoc analysis of the transformed DLIPS measurements

PAGE 146

146 Figure 4 11 Confusion matrix for the fluorescence data under transformation T1

PAGE 147

147 Figure 4 12 Confusion matrix for the fluorescence data under transformation T2

PAGE 148

148 Figure 4 13 Confusion matrix for the fluorescence data under transformation T4

PAGE 149

149 Figure 4 14 Confusion matrix for the DLIPS data under transformation T4

PAGE 150

150 Figure 4 15 Confusion matrix for the DLIPS data under transformation T2

PAGE 151

151 CHAPTER 5 CONCLUSION The work in this dissertation focused on the development of DLIPS fluorescence biosensing, a new spectral technique, into a clinical detection technique, with the long malignancies earlier and with enhanced accuracy. Here we presented a proof of concept study (Chapter 2) in mice that established a key fundamental fact: DLIPS can be used to discriminate pathological tissue from healthy tissue in vivo. Further, when comparing this technique to fluorescence spectroscopy alone, it was found to perform at least as well as this technique; one which is seeing success in clinical trials for diagnosing disease using advanced techniques, but has seen mixed results in the currently available commercial systems. In an effort to understand the utility and opportunity for maximizing DL IPS clinical benefit, a fluorescence photochemical Monte Carlo model of the DLIPS technique was constructed (Chapter 3) Here it was discovered that DLIPS in fact highlights unique regional and fluorescent properties of disease that wer e previously inacces sible using fluorescence spectroscopy alone. This build s support for th e case that DLIPS biosensing could stand alone as a diagnostic technique or can complement detection systems for oral pre malignancies by providing an orthogonal dimension of informatio n. Finally, in a pre clinical assessment of this technique as a tool for detecting oral cancer (Chapter 4) while DLIP S failed to demonstrate superior detection capabilities due to extremely low signal levels, the investigation of fluorescence detection i n this multi class framework demonstrated that by using detection theory, the discrimination can potentially be enhanced. This result implies that by designing systems for solving

PAGE 152

152 the detection problem of discriminating benign pathology from oral pre malig nancies, the most common problems associated with currently marketed diagnostic aids can be mitigated. Specifically, this work demonstrates that this particular design and testing strategy can reduce the high false positive rate associated with screening without trading sensitivity to oral dysplasia. Overall, this dissertation provides the experimental and theoretical links between DLIPS as an abstract technique, diagnostic tissue spectroscopy, and the unmet need in oral pre malignancy detection. This work represents an important step forward for DLIPS biosensing techniques, extending the capabilities of this technique from in vtiro applications to in vivo diagnostics. Results from both modeling and experiments confirm that DLIPS accesses fundamentally diff erent biomolecular information that fluorescence spectroscopy alone. These results represent the foundation for the extension of this biosensing technique to other diseases as well as the coupling of laser induced perturbation with other sensing modalities (Raman, magnetic resonance, other optical spectroscopies). Further, for in vitro applications, the results from the Monte Carlo models in Chapter 4 imply a dual mechanism for inherent contrast associated with DLIPS: optical clearing of superficial interfe rence and direct coupling to cleaved biomolecules. This combination mechanism is a critical advantage of the DLIPS technique and in the Monte Carlo models accounted for the interrogation of FAD and keratin respectively (biomolecules inaccessible in the flu orescence spectra alone). While there is not an immediate direct impact for DLIPS on oral cancer detection due to the long regulatory path to clinical testing, these results confirm two important general facts about effective optical tools for non invasiv e screening and diagnosis. First

PAGE 153

153 is that there are some fundamental flaws with currently technologies in terms of what information is integrated into the returned signal from the tissue. Specifically, the results in Chapter 3 indicate that fluorescence spe ctra from tissue heavily emphasize information from the stroma and basal epithelium, obscuring information from the superficial epithelial layers. In this same study, it was shown that DLIPS in fact targets these superficial structures, where other studies have shown that some of the earliest morphological and biochemical changes occur leadin g to oral malignancies 29, 90, 92 This result is disruptive to the market for oral cancer adjuvants as none of these instruments currently provide depth selectivity or t arget biomolecule selectivity, but this selectivity could be a critical feature for a successful diagnostic aid. Second, in Chapter 4 this work demonstrated that devices trained to discriminate benign pathology from dysplasia provide overall better diagno stic performance in animal models. This is again disruptive as the current design archetype design to discriminate known dysplastic lesions against healthy oral mucosa (eg: loss of fluorescence). What this design paradigm does not consider is the specificity of the observed contrast for cancer versus other pathologies ; this supposed to be a key factor driving down detection performance for current commercial devices. Implementation of this strategy is not easy and will require new and more rigorous approaches to clinical validation and testing, lengthening time to market and development costs. At the same time, the opportunity to drive up 5 year survival rates for oral cancer is immeasurably valuable. There are several important nex t steps for the development of in vivo DLIPS based screening and diagnostic technologies. Primarily these studies revolve around

PAGE 154

154 continuing to build the theoretical basis for this technique as well as confirming the results presented in this dissertation. One important next step is to better optimize this technique for in vivo analysis Based on the results in Chapters 3 and 4 it is clear that while DLIPS should minimize inter patient variability, it has a very low tolerance for measurement noise. This can likely be mitigated somewhat by averaging multiple measurements before and after perturbation or rethinking how to focus the excitation beam and collection optics to decrease the fluorescence interaction volume. If the volume of fluorescence interactions at 355 nm is around 100 times larger than the footprint of photochemical interactions, then the majority of that remitted fluorescence is cluttering assessment of the photochemical changes induced by the perturbation beam. If the goal is a more localized a nalysis of the perturbation region then optical strategies for targeting that tissue specifically should be considered. Or potentially coupling this technique with a more superficial spectral analysis technique, such as resonance Raman with a shorter wavel ength UV excitation laser to better restrict the penetration depth and focus more on the superficial sig nals. In short, there is a good deal of optimization work to be done on this technique, but t he results provided here give a firm starting point for inc remental t echnical improvements. Second, is establishing the safety profile for this technique, a key requirement to move forward into clinical testing. From the perspective of the Food and Drug Administration (FDA) as well as any Institutional Review Boar d, there are three apparent safety hazards associated with the current system that must be addressed. These are safety concerns regarding eye safety (with regards to the coherent light sources used in this device) for patients as well and healthcare profes sionals, the safety

PAGE 155

155 of applying sub ablative pulses of deep UV light to tissue in vivo, and protection from Fluorine gas exposure. In all of these cases there are pre existing, FDA approved clinical systems that can be used for patterning the design as wel l as provide supporting evidence for the safety profile of this device Finally, optimization of the optical and mechanical design for portability and integration with existing minimally invasive endoscopic systems could significantly broaden the applicati ons of this technique to detecting other diseases such as cervical and colorectal cancer or atherosclerosis. The long term outlook for this technique as an in vivo biosensing modality is promising, if nascent, based upon the results presented here. This wo rk enables the next steps forward for DLIPS as a non invasive detection technique: translation into clinical studies of oral pathology, and extension to other target tissues and pathologies. Along the way, a clearer picture of some technical issues will be necessary to troubleshoot and optimize the diagnostic perfo rmance for this challenging biosensing setting. A s the knowledge base for DLIPS technology continues to grow, other applications are dis covered and the application of DLIPS to superficial cancer detection matures, the full promise of this new spe ctral technique presented here will be realized

PAGE 156

156 APPE NDIX MATLAB CODE FOR MONTE CARLO SIMULATIONS %dlips_mc2.m %Raymond Kozikowski %Aug 1, 2012 %monte carlo code to describe dlips phenomena in tissue % Assume that photons dont split, each photon is a discrete packet until it %dies in the simulation %USE ANGLES IN DEGREES!!!!! function out=dlips_mc2(model, no_pho, beam, mask_mu, mask_f) tic res=.5; %pixels per um, assume pixels are square width=10e3; %width of the model (in um), to create an ~infinite width wrt beam width (1mm) randn( 'state' ,sum(100*clock)); %reinitilize randn each time the program runs rand( 'state' ,sum(100*clock)); %define excitation and perturbation parameters lambx=355; %excit ation in nm lambp=193; %perturbation in nm %describe the probe optical properties NA=.22; %taken from the fiber properties nARF=1.666; %assuming interface with a crystal quartz window %nfluor=1.56; %for fiber applications nfluor=1; %for free beam laser applications if strcmp(model, 'test' )==0 && strcmp(model, 'test2' )==0 %define tissue fluorophores (given as relative emission from tissue layers %which correspond to dominant fluorophores NADH=fluoro_fit([355 385 405 425 455 475 495 525 545 565],[0 .1 .29 .49 .69 .51 .41 .28 .2 .13],lambx); FAD=fluoro_fit([475 495 505 525 535 545 565],[0 .025 .075 .12 .15 .12 .1],lambx); Collagen=fluoro_fit([355 385 395 410 425 455 475 495 525 545 565],[0 .19 .35 .47 .35 .22 .14 .1 .06 .04 .03],lambx); Keratin=fluor o_fit([355 385 405 425 455 475 495 525 545 565],[0 .07 .1 .11 .1 .08 .05 .03 .02 .01],lambx); nNADH=sum(.5*(NADH{2}(2:end)+NADH{2}(1:end 1))./(NADH{1}(2:end) NADH{1}(1:end 1))); nFAD=sum(.5*(FAD{2}(2:end)+FAD{2}(1:end 1))./(FAD{1}(2:end) FAD{1}(1:end 1))); nCollagen=sum(.5*(Collagen{2}(2:end)+Collagen{2}(1:end 1))./(Collagen{1}(2:end) Collagen{1}(1:end 1))); nKeratin=sum(.5*(Keratin{2}(2:end)+Keratin{2}(1:end 1))./(Keratin{1}(2:end) Keratin{1}(1:end 1))); NADH2=(NADH{1}(2) NADH{1}(1))*NADH{2}/nNADH; FAD2=(N ADH{1}(2) NADH{1}(1))*FAD{2}/nFAD; Collagen2=(NADH{1}(2) NADH{1}(1))*Collagen{2}/nCollagen; Keratin2=(NADH{1}(2) NADH{1}(1))*Keratin{2}/nKeratin; %load and convert your hemoglobin data (mu_a, cm 1)

PAGE 157

157 hb_mat=load( 'C: \ Users \ Owner \ Desktop \ dissertation \ ref s for MC \ hb_ext_spectra.txt' ); hb_mat(1:end,2:3)=hb_mat(1:end,2:3)*2.303*150/64500; %change me for healthy versus disease if strcmp(model, 'cheek' )==1 blood_frac=.002; else blood_frac=3*.002; end O2_frac=.8; %skin_bkgd_mua=0.244+85.3*exp( (hb_m at(1:end,1) 154)/66.2); %Jacques approx for background skin_bkgd_mua=7.84e8.*hb_mat(1:end,1).^( 3.255); %Saidi approx for bkgd (maybe used in Pavlova?) wave_bkgd=193:1:1000; %skin_bkgd_mua2=0.244+85.3*exp( (wave_bkgd 154)/66.2); skin_ bkgd_mua2=7.84e8.*wave_bkgd.^( 3.255); net_hb=O2_frac*hb_mat(1:end,2)+(1 O2_frac)*hb_mat(1:end,3); net_stroma=blood_frac*(net_hb)+(1 blood_frac)*skin_bkgd_mua; hb_mat(53,1) st_corr=(1 blood_frac)*skin_bkgd_mua(53)/net_stroma(53); skin_bkgd_mua(53)/(net_hb (53)+skin_bkgd_mua(53)); if strcmp(model, 'ArF, cheek' )==0 figure(2) semilogy(hb_mat(1:end,1),hb_mat(1:end,2), r' ,hb_mat(1:end,1),hb_mat(1:end,3), b' ,wave_bkgd,skin_bkgd_mua2, g' ,hb_mat(1:end,1),net_stroma, k' ,hb_mat(1:end,1),net_hb, -k' ) legend( 'HbO_2' 'Hb' 'tissue bkgd' 'net \ mu_a' 'net Hb' ) xlabel( 'wavelength (nm)' ) ylabel( \ mu_a (cm^{ 1})' ) end if strcmp(model, 'ArF, cheek' )==0 %prepare figures of fluorescence response here figure(1) plot(NADH{1},.95*.74*NADH2, ,FA D{1},.4*.22*FAD2, ,Collagen{1},.9*Collagen2*st_corr, ,Keratin{1},.12*Keratin2, ) ylabel( 'Normalized Fluorescence' ) xlabel( 'Emission Wavelength (nm)' ) legend( 'NADH' 'FAD' 'Collagen' 'Keratin' 'Location' 'NorthWest' ) %title('Fluorescent efficiency (355nm ex) of the dominant fluorophores of the buccal epithelium') end end %initialize your tissue model if strcmp(model, 'cheekd' ) == 1 mods=[111/40; .3/.12; .4/.22; 95/66; 277/130; 256/320; .3/.88; .7/.9]; %layer1 thick, layer1 fluor, la y2 flour, layer3 ms, layer3 thick, layer4 ms, layer4 fluor, layer5 fluor d=[mods(1)*80 80 mods(5)*120 125 10e3]; no_layers=length(d); n=[1.5 1.34 1.38 1.4 1.42]; %may need a correction for wavelength (ArF) g=[.97 .97 .97 .88 .88];

PAGE 158

158 qy=[.3 .4*.4 .8 .3*st_corr .7*st_corr]; %qy's for keratin, FAD, NADH, and collagen at 355nm excitation f_layer={Keratin{2}; FAD{2}; NADH{2}; Collagen{2}; Collagen{2}}; draw={make_draw(Keratin{2},NADH{1}),make_draw(FAD{2},NADH{1}),make_draw( NADH{ 2},NADH{1}),make_draw(Collagen{2},NADH{1}),make_draw(Collagen{2},NADH{1})}; ms=cell(no_layers,1); %scattering coeff for each layer mus_wave1=[350, 355:20:555 810]; mus_wave2=[350, 350:50:650, 355:20:555 810]; mus_wave3=[35 0, 350:50:650, 355:20:555, 488, 700, 450, 650, 633, 635, 488]; mus_wave4=[350, 350:50:650]; ms{1}=mus_spec(mus_wave1,[204 200 190 180 170 160 155 150 145 135 130 125 88],lambx); ms{2}=mus_spec(mus_wave2,[66 50 45 40 35 30 27 25 65 61 57 54 52 5 0 49 48 47 46 45 33],lambx); ms{3}=mus_spec(mus_wave4,[66 50 45 40 35 30 27 25],lambx); ms{4}=mus_spec(mus_wave3,[320 315 280 250 235 225 200 180 315 300 280 270 260 250 240 225 220 210 205 270 150 248.5 172 187 244 239],lambx); ms{5}=ms{4}; ms{3}{2}=mods(4)*ms{3}{2}; ms{4}{2}=mods(6)*ms{4}{2}; figure(3) plot(ms{1}{1},ms{1}{2}, r' ,ms{2}{1},ms{2}{2}, b' ,ms{3}{1},ms{3}{2}, g' ,ms{4}{1},ms{4}{2}, k' ) legend( 'super. epith.' 'inter. epith.' 'basal epith.' 'stroma' ) hold on plot(mus_wave1,[204 200 190 180 170 160 155 150 145 135 130 125 88], 'or' ) plot(mus_wave2,[66 50 45 40 35 30 27 25 65 61 57 54 52 50 49 48 47 46 45 33], 'ob' ) plot(mus_wave4,[66 50 45 40 35 30 27 25], 'og' ) plot(mus_wave3,[320 315 280 250 235 225 200 180 315 300 280 270 260 250 240 225 220 210 205 270 150 248.5 172 187 244 239], 'ok' ) xlabel( 'wavelength (nm)' ) ylabel( \ mu_s (cm^{ 1})' ) hold off ma=cell(no_layers,1); ma{1}=mua_epi([355:40:555 350:50:650],[4 3 2.5 2 1.7 1. 5 4 3.2 2.5 2 1.8 1.7 1.6],lambx); ma{2}=ma{1}; ma{3}=ma{1}; ma{5}={hb_mat(1:end,1); net_stroma+4}; %ma{4}={hb_mat(1:end,1); skin_bkgd_mua+4}; ma{4}=ma{5}; figure(4) plot(ma{1}{1},ma{1}{2}, r' ,ma{5}{1},ma{5}{2}, b' ,ma{4}{1 },ma{4}{2}, g' ) legend( 'epithelium' 'deep stroma' 'superf. stroma' ) hold on plot([355:40:555 350:50:650],[4 3 2.5 2 1.7 1.5 4 3.2 2.5 2 1.8 1.7 1.6], 'or' ) xlabel( 'wavelength (nm)' ) ylabel( \ mu_a (cm^{ 1})' ) axis([350 700 0 20])

PAGE 159

159 hold off d_tot=sum(d); pho_mat_ex=zeros(width*res,round(d_tot*res),2); %use to track absorbed weight, NEED TO ROUND THIS... fluor_gen=zeros(width*res,round(d_tot*res)); %pho_mat_ARF=zeros(width*res,d_tot*res,5); %use to track photoche mical events %could break these up by layers (or even further) to add resolution %size(pho_mat_ex) offset=find(hb_mat(1:end,1)==354); %alignment for mu_a wrt fluorescence and mu_s lamb_pho_model=1; %assumes everything starts as excitati on elseif strcmp(model, 'cheek' ) == 1 d=[80 80 120 125 10e3]; no_layers=length(d); n=[1.5 1.34 1.38 1.4 1.42]; %may need a correction for wavelength (ArF) g=[.97 .97 .97 .88 .88]; qy=[.12 .22*.4 .74 .88*st_corr .9*st_corr]; %qy's for keratin, FAD, NADH, and collagen at 355nm excitation f_layer={Keratin{2}; FAD{2}; NADH{2}; Collagen{2}; Collagen{2}}; draw={make_draw(Keratin{2},NADH{1}),make_draw(FAD{2},NADH{1}),make_draw(NADH{ 2},NADH{1}),make_draw(Collagen{2},NADH {1}),make_draw(Collagen{2},NADH{1})}; ms=cell(no_layers,1); %scattering coeff for each layer mus_wave1=[350, 355:20:555 810]; mus_wave2=[350, 350:50:650, 355:20:555 810]; mus_wave3=[350, 350:50:650, 355:20:555, 488, 700, 450, 6 50, 633, 635, 488]; mus_wave4=[350, 350:50:650]; ms{1}=mus_spec(mus_wave1,[204 200 190 180 170 160 155 150 145 135 130 125 88],lambx); ms{2}=mus_spec(mus_wave2,[66 50 45 40 35 30 27 25 65 61 57 54 52 50 49 48 47 46 45 33],lambx); ms{3}=mus_ spec(mus_wave4,[66 50 45 40 35 30 27 25],lambx); ms{4}=mus_spec(mus_wave3,[320 315 280 250 235 225 200 180 315 300 280 270 260 250 240 225 220 210 205 270 150 248.5 172 187 244 239],lambx); ms{5}=ms{4}; figure(3) plot(ms{1}{1},ms{1}{2}, r' ,ms{2}{1},ms{2}{2}, b' ,ms{3}{1},ms{3}{2}, g' ,ms{4}{1},ms{4}{2}, k' ) legend( 'super. epith.' 'inter. epith.' 'basal epith.' 'stroma' ) hold on plot(mus_wave1,[204 200 190 180 170 160 155 150 145 135 130 125 88], 'or' ) plot(mus_wave2,[66 50 4 5 40 35 30 27 25 65 61 57 54 52 50 49 48 47 46 45 33], 'ob' ) plot(mus_wave4,[66 50 45 40 35 30 27 25], 'og' ) plot(mus_wave3,[320 315 280 250 235 225 200 180 315 300 280 270 260 250 240 225 220 210 205 270 150 248.5 172 187 244 239], 'ok' ) xlabel( wavelength (nm)' ) ylabel( \ mu_s (cm^{ 1})' ) hold off

PAGE 160

160 ma=cell(no_layers,1); ma{1}=mua_epi([355:40:555 350:50:650],[4 3 2.5 2 1.7 1.5 4 3.2 2.5 2 1.8 1.7 1.6],lambx); ma{2}=ma{1}; ma{3}=ma{1}; ma{5}={hb_mat(1:end,1); net_ stroma+4}; %ma{4}={hb_mat(1:end,1); skin_bkgd_mua+4}; ma{4}=ma{5}; figure(4) plot(ma{1}{1},ma{1}{2}, r' ,ma{5}{1},ma{5}{2}, b' ,ma{4}{1},ma{4}{2}, g' ) legend( 'epithelium' 'deep stroma' 'superf. stroma' ) hold on plot([355:40:55 5 350:50:650],[4 3 2.5 2 1.7 1.5 4 3.2 2.5 2 1.8 1.7 1.6], 'or' ) xlabel( 'wavelength (nm)' ) ylabel( \ mu_a (cm^{ 1})' ) axis([350 700 0 20]) hold off d_tot=sum(d); pho_mat_ex=zeros(width*res,round(d_tot*res),2); %use to track absor bed weight, NEED TO ROUND THIS... fluor_gen=zeros(width*res,round(d_tot*res)); %pho_mat_ARF=zeros(width*res,d_tot*res,5); %use to track photochemical events %could break these up by layers (or even further) to add resolution %size(pho_mat_e x) offset=find(hb_mat(1:end,1)==354); %alignment for mu_a wrt fluorescence and mu_s lamb_pho_model=1; %assumes everything starts as excitation elseif strcmp(model, 'cheekb' ) == 1 mods=[2; 120/100; 100/130; .17/.22; .15/.74; .15/.88; .15/ .9]; %th1 th2 th3 f2 f3 f4 f5 d=[mods(1)*80 mods(2)*80 mods(3)*120 125 10e3]; no_layers=length(d); n=[1.5 1.34 1.38 1.4 1.42]; %may need a correction for wavelength (ArF) g=[.97 .97 .97 .88 .88]; qy=[.12 .17*.4 .15 .15*st_corr .15 *st_corr]; %qy's for keratin, FAD, NADH, and collagen at 355nm excitation f_layer={Keratin{2}; FAD{2}; NADH{2}; NADH{2}; Collagen{2}}; draw={make_draw(Keratin{2},NADH{1}),make_draw(FAD{2},NADH{1}),make_draw(NADH{ 2},NADH{1}),make_draw(Collage n{2},NADH{1}),make_draw(Collagen{2},NADH{1})}; ms=cell(no_layers,1); %scattering coeff for each layer mus_wave1=[350, 355:20:555 810]; mus_wave2=[350, 350:50:650, 355:20:555 810]; mus_wave3=[350, 350:50:650, 355:20:555, 488, 70 0, 450, 650, 633, 635, 488]; mus_wave4=[350, 350:50:650]; ms{1}=mus_spec(mus_wave1,[204 200 190 180 170 160 155 150 145 135 130 125 88],lambx);

PAGE 161

161 ms{2}=mus_spec(mus_wave2,[66 50 45 40 35 30 27 25 65 61 57 54 52 50 49 48 47 46 45 33],lambx); m s{3}=mus_spec(mus_wave4,[66 50 45 40 35 30 27 25],lambx); ms{4}=mus_spec(mus_wave3,[320 315 280 250 235 225 200 180 315 300 280 270 260 250 240 225 220 210 205 270 150 248.5 172 187 244 239],lambx); ms{5}=ms{4}; figure(3) plot(ms{1}{1},ms{1 }{2}, r' ,ms{2}{1},ms{2}{2}, b' ,ms{3}{1},ms{3}{2}, g' ,ms{4}{1},ms{4}{2}, k' ) legend( 'super. epith.' 'inter. epith.' 'basal epith.' 'stroma' ) hold on plot(mus_wave1,[204 200 190 180 170 160 155 150 145 135 130 125 88], 'or' ) plot(mus_wave2 ,[66 50 45 40 35 30 27 25 65 61 57 54 52 50 49 48 47 46 45 33], 'ob' ) plot(mus_wave4,[66 50 45 40 35 30 27 25], 'og' ) plot(mus_wave3,[320 315 280 250 235 225 200 180 315 300 280 270 260 250 240 225 220 210 205 270 150 248.5 172 187 244 239], 'ok' ) xlabel( 'wavelength (nm)' ) ylabel( \ mu_s (cm^{ 1})' ) hold off ma=cell(no_layers,1); ma{1}=mua_epi([355:40:555 350:50:650],[4 3 2.5 2 1.7 1.5 4 3.2 2.5 2 1.8 1.7 1.6],lambx); ma{2}=ma{1}; ma{3}=ma{1}; ma{5}={hb_mat(1:end ,1); net_stroma+4}; %ma{4}={hb_mat(1:end,1); skin_bkgd_mua+4}; ma{4}=ma{5}; figure(4) plot(ma{1}{1},ma{1}{2}, r' ,ma{5}{1},ma{5}{2}, b' ,ma{4}{1},ma{4}{2}, g' ) legend( 'epithelium' deep stroma' 'superf. stroma' ) hold on plot( [355:40:555 350:50:650],[4 3 2.5 2 1.7 1.5 4 3.2 2.5 2 1.8 1.7 1.6], 'or' ) xlabel( 'wavelength (nm)' ) ylabel( \ mu_a (cm^{ 1})' ) axis([350 700 0 20]) hold off d_tot=sum(d); pho_mat_ex=zeros(width*res,round(d_tot*res),2); %use to t rack absorbed weight, NEED TO ROUND THIS... fluor_gen=zeros(width*res,round(d_tot*res)); %pho_mat_ARF=zeros(width*res,d_tot*res,5); %use to track photochemical events %could break these up by layers (or even further) to add resolution %size (pho_mat_ex) offset=find(hb_mat(1:end,1)==354); %alignment for mu_a wrt fluorescence and mu_s lamb_pho_model=1; %assumes everything starts as excitation elseif strcmp(model, 'test' ) ==1 d=[10e3]; width=100e3; res=.1;

PAGE 162

162 d_tot=sum(d ); no_layers=length(d); n=[1]; %may need a correction for wavelength (ArF) nfluor=1; g=[.8]; ma=cell(no_layers,1); ms=cell(no_layers,1); ma{1}={[355 355];[.1 .1]}; ms{1}={[355 355];[10 10]}; offset=0; lamb_pho_ model=2; %assumes everything starts as excitation pho_mat_ex=zeros(width*res,round(d_tot*res)); pho_mat_ARF=pho_mat_ex; elseif strcmp(model, 'test2' ) ==1 d=[10e3]; width=100e3; res=.1; d_tot=sum(d); no_layers=length(d); n=[1.37]; %may need a correction for wavelength (ArF) nfluor=1; g=[.8]; ma=cell(no_layers,1); ms=cell(no_layers,1); ma{1}={[355 355];[.1 .1]}; ms{1}={[355 355];[10 10]}; offset=0; lamb_pho_model=2; %assumes everyth ing starts as excitation pho_mat_ex=zeros(width*res,round(d_tot*res)); pho_mat_ARF=pho_mat_ex; elseif strcmp(model, 'ArF, cheek' )==1 %setup for cheek d=[80 80 120 125 10e3]; d_tot=sum(d); no_layers=length(d); n=[1.5 1.34 1.38 1. 4 1.42]; %may need a correction for wavelength (ArF) g=[.97 .97 .97 .88 .88]; %may be able to estimate a correction or better values nfluor=nARF; mus_wave1=[350, 355:20:555 810]; mus_wave2=[350, 350:50:650, 355:20:555 810]; mus_wave3=[350, 350:50:650, 355:20:555, 488, 700, 450, 650, 633, 635, 488]; mus_wave4=[350, 350:50:650]; ma=cell(no_layers,1); ms=cell(no_layers,1); ms193_1=ms193(6000,mus_wave1,[204 200 190 180 170 160 155 150 145 135 130 12 5 88]); ms193_2=ms193(8000,mus_wave2,[66 50 45 40 35 30 27 25 65 61 57 54 52 50 49 48 47 46 45 33]); ms193_3=ms193(8000,mus_wave4,[66 50 45 40 35 30 27 25]); ms193_4=ms193(1000,mus_wave3,[320 315 280 250 235 225 200 180 315 300 280 270 260 250 240 225 220 210 205 270 150 248.5 172 187 244 239]);

PAGE 163

1 63 ma{1}={[193 193];[6000 6000]}; %except theres no melanin unlike in the SC... ms{1}={[193 193];[ms193_1{1} ms193_1{1}]}; ma{2}={[193 193];[8000 8000]}; ms{2}={[193 193];[ms193_2{1} ms1 93_2{1}]}; ma{3}={[193 193];[8000 8000]}; ms{3}={[193 193];[ms193_3{1} ms193_3{1}]}; ma{4}={[193 193];[1000 1000]}; ms{4}={[193 193];[ms193_4{1} ms193_4{1}]}; ma{5}={[193 193];[1000 1000]}; ms{5}={[193 193];[ms193_4{1} ms193_4{1}]}; offset=0; lamb_pho_model=2; %assumes everything starts as excitation pho_mat_ex=zeros(width*res,round(d_tot*res)); arf_gen=zeros(width*res,round(d_tot*res),2); %pho_mat_ARF=zeros(width*res,d_tot*res,5); %use to track photochemical events photochem_pho=zeros(no_pho,8); %first 5 are flags for pc events, x, y, layer else 'ERROR invalid model name' return end ma{1}{1}(1+offset) ma{1}{2}(1+offset) ma{2}{2}(1+offset) ma{3}{2}(1+offset) ma{4}{2}(1+offset) ma{5}{2}(1+offset) % % ms{1}{1}(1) % ms{1}{2}(1) % ms{2}{2}(1) % ms{3}{2}(1) % ms{4}{2}(1) % ms{5}{2}(1) %initialize your disease model %initialize collection vectors reflect=zeros(no_pho,3); %weight, position, and angle respectively return_check=zeros(no_pho,1); %che ck vector for photons that didnt violate any rules or die new_pho_check=zeros(no_pho,1); %check for fluorescent photons that made it out fluor_pho=zeros(no_pho,5); %weight, wavelength, position, angle, (layer generated to come...) %fluor_gen=pho_mat_ex; %track where fluorescence is generated %pho_mat_ex=zeros(width*res,d_tot*res); layer_vec=ones(sum(d)*res,1); %ALSO NEED TO ROUND THIS d2=cumsum(d); d3=[0 d2(1:(end 1))]; for i=1:(no_layers 1) layer_vec(((res*d2(i))+1):end)=1+layer_vec(((res*d2(i) )+1):end); %COULD ALSO BE PROBLEMATIC BC OF ROUNDING end

PAGE 164

164 %initialize sim param cutoff=1e 4; %cutoff for playing roulette x=0; y=0; angle=0; W=1; lamb_pho=lamb_pho_model; %assumes everything starts as excitation f_layer_hold= 1; %lauch a photon for i=1:no_pho %i %from where and at what angle? if isempty(beam) %pencil beam, everything starts in the middle nfluor=1; x=width/2; y=0; angle=0; %angle=launch_angle(NA,n(1)); elseif strcmp(beam 'diffuse' ) %point source at the surface nfluor=1; y=0.1; x=width/2; angle=(2*round(rand) 1)*180*rand; else %use the model that was loaded and draw from that distribution angle=launch_angle(NA,n(1 )); %angle=(2*round(rand) 1)*89*rand; %angle=0; y=0; beamw=1000; if strcmp(model, 'ArF, cheek' )==1 x=(2*round(rand) 1)*.5*beamw*sqrt(rand)+.5*width; else x=(2*round(rand) 1)*.5*beamw*sq rt(rand)+.5*width; end end %with what weight and wavelength? W=1; lamb_pho=lamb_pho_model; f_layer_hold= 1; %did it make it into the top layer? start_check=fresnel(nfluor,n(1),angle); if start_check{2,1} == 1 && strcmp(beam, 'diffuse' )==0 %then your photon was rejected, dont check for a sub surface pt source reflect(i,1)=W; reflect(i,2)=x; reflect(i,3)=angle; %'rejection' else %photon will propagate while W > 0 %is my weight large enough, or do I play roulette?

PAGE 165

165 if abs(angle)>180 input( 'ERROR angle out of range, Enter to continue' ) end if W < cutoff %'play roulette' m=10; %or should this be indep of cutoff? roul=rand; if roul < (1/m) %photon survives w weight mW W=m*W; else %photon is terminated W=0; break end end %find my layer layer=layer_vec(round(res*y)+1); %lets make a move to my next scattering event depx2=round(res*x)+1; depy2=round(res*y)+1; if depy2<=size(mask_mu,2) && isempty(mask_mu)==0 && y>0 && (strcmp(model, 'cheek' )==1 || strcmp(model, 'cheekd' )==1 || strcmp(model, 'cheekb' )==1) s= log(rand)./(ms{layer}{2}(lamb_pho)+mask_mu(depx2,depy2)*ma{layer}{2}(lamb_pho +offset)) ; else s= log(rand)./(ms{layer}{2}(lamb_pho)+ma{layer}{2}(lamb_pho+offset)); end x_new=x+s*1e4*sind(angle); %convert from cm to um y_new=y+s*1e4*cosd(angle); %did i go out of bounds ? if (x_new*res+1)>=(width*res) || (x_new*res+1)<=1 || (y_new*res+1)>=(d_tot*res) W=0; %'photon out of bounds' break end %did I try and escape? if y _new <= 0 escape_check=fresnel(n(1),nfluor,angle); if escape_check{1,1}==1 %photon escaped, but was bent return_check(i)=1; fluor_pho(i,1)=W; fluor_ pho(i,2)=lamb_pho; fluor_pho(i,3)=(x+x_new)/2; fluor_pho(i,4)=escape_check{2,1}; fluor_pho(i,5)=f_layer_hold; if lamb_pho>1 && (strcmp(model, 'cheek' )==1 || strcmp(model, 'cheekd )==1 || strcmp(model, 'cheekb' )==1) fluor_pho(i,6)=ma{f_layer_hold}{2}(lamb_pho+offset); end W=0; if lamb_pho > 1

PAGE 166

166 new_pho_check(i,1)=1; end %'photon heads out' break else %photon reflected back in so update the position y_new=2*0 s*1e4*cosd(angle) y; %'reflection at top' angle=escape_check{2,2}; end %did i change layers or reflect off the boundary? elseif ((y_new>d2(layer) && yd3(layer))) && no_layers>1 new_layer=layer_ve c(round(res*y_new)+1); %into which layer? ref_check=fresnel(n(layer),n(new_layer),angle); if ref_check{1,1}==1 %photon migrated, but was bent if layer > new_layer % then y_layer < y && y_new < y_layer %assume refraction is small deflection and just %update angle dy=abs(s*1e4*cosd(angle)); %suggested dy dy_layer=abs(y d3(layer)); %move to the boundary frac=dy_layer/dy; s2=frac*s*1e4; x_layer=s2*sin(angle)+x; angle=ref_check{1,2}; x_new=x_layer+(s*1e4 s2)*sin(angle); y_new=d3(layer) abs((s*1e4 s2)*cos(angle)); end if layer < new_layer % then y_layer > y && y_new > y_layer dy=abs(s*1e4*cosd(angle)); %suggested dy dy_layer=abs(y d 2(layer)); %move to the boundary frac=dy_layer/dy; s2=frac*s*1e4; x_layer=s2*sin(angle)+x; angle=ref_check{1,2}; x_new=x_layer+(s*1e4 s2) *sin(angle); y_new=d2(layer)+abs((s*1e4 s2)*cos(angle)); end layer=new_layer; else %photon reflected back in so update the position if layer > n ew_layer y_new=2*d3(layer) s*1e4*cosd(angle) y; angle=ref_check{2,2}; %'bounced off of a layer coming back, angle should point in, abs<90' end i f layer < new_layer y_new=2*d2(layer) s*1e4*cosd(angle) y; angle=ref_check{2,2}; %'bounced off of a layer going in, angle should point out, abs>90' end end end %need one more check for exiting

PAGE 167

167 if y_new<=0 %made it out by bouncing off of another boundary return_check(i)=1; fluor_pho(i,1)=W; fluor_pho(i,2 )=lamb_pho; fluor_pho(i,3)=(x+x_new)/2; fluor_pho(i,4)=angle; fluor_pho(i,5)=f_layer_hold; if lamb_pho>1 && (strcmp(model, 'cheek' )==1 || strcmp(model, 'cheekd' )==1 || strcmp(model, 'cheekb' )==1) fluor_pho(i,6)=ma{f_layer_hold}{2}(lamb_pho+offset); end W=0; if lamb_pho > 1 new_pho_check(i,1)=1; end %'photon heads out' break end %need one last check to see if it exited the tissue via the %sides if (x_new*res+1)>=(width*res) || (x_new*res+1)<=1 || (y_new*res+1)>=(d_tot*res) W=0; %'photon out of bounds' break end %lets update our weight and deposit some weight in the new %position depx=round(res*x_new)+1; depy=round(res*y_new)+1; if depy<=size(mask_mu,2) && isempty(mask_mu)==0 dW=W.*ma{layer}{2}(lamb_pho)*mask_mu(depx,depy)./(ms{layer}{2}(lamb_pho)+mask _mu(depx,depy)*ma{layer}{2}(lamb_pho+offset)); else dW=W.*ma{layer}{2}(lamb_pho)./(ms{layer}{2}(lamb_pho)+ma{la yer}{2}(lamb_pho+o ffset)); end if (strcmp(model, 'cheek' )==1 || strcmp(model, 'cheekd' )==1 || strcmp(model, 'cheekb' )==1) if depy<=size(mask_mu,2) && isempty(mask_mu)==0 pho_mat_ex(depx,depy,1)=pho_m at_ex(depx,depy,1)+(dW./(mask_mu(depx,depy)*ma{l ayer}{2}(lamb_pho+offset))); else pho_mat_ex(depx,depy,1)=pho_mat_ex(depx,depy,1)+(dW./(ma{layer}{2}(lamb_pho+o ffset))); end pho_mat_ex(depx ,depy,2)=pho_mat_ex(depx,depy,2)+dW; else pho_mat_ex(depx,depy)=pho_mat_ex(depx,depy)+dW; end W=W dW;

PAGE 168

168 %otherwise lets update my position and angle (via scattering) and get ready to %do it again nu=.5/g(layer).*(1+g(layer).^2 (1 g(layer).^2).^2.*(1 g(layer)+2*g(layer)*rand).^( 2)); dangle=(2*round(rand) 1).*acosd(nu); angle=angle+dangle; if strcmp(model, 'ArF, cheek' )==1 %bc rayleigh sca ttering is isotropic angle=180*rand*(2*round(rand) 1); end %did i fluoresce? %make sure to load new vectors of optical properties for me if lamb_pho==1 %still excitation so now test to see if fl uorescence happens if depy<=size(mask_mu,2) && isempty(mask_mu)==0 fluor_check=pfluor(mask_f(depx,depy)*f_layer{layer}, mask_mu(depx,depy)*ma{layer}{2}(lamb_pho+offset), NADH{1},draw{layer},mask_f(depx,depy)*qy(layer),ms {layer}{2}(lamb_pho)); else fluor_check=pfluor(f_layer{layer}, ma{layer}{2}(lamb_pho+offset), NADH{1},draw{layer},qy(layer),ms{layer}{2}(lamb_pho)); end if fluor_check{1,1}==1 %then fluoresc ence happend lamb_pho=fluor_check{2,2}; angle=180*rand*(2*round(rand) 1); fluor_gen(depx,depy)=fluor_gen(depx,depy)+W; f_layer_hold=layer; %input('Fluorescence!, click to continue') end end % if strcmp(model, 'ArF, cheek')==1%check to see if a photochemical event happend % ArF_check=pArF(layer, W, ma{layer}{2}(2), ms{layer}{2}(2)); % if ArF_che ck{1,1}==1 % photochem_pho(i,1)=ArF_check{1}; % photochem_pho(i,2)=ArF_check{2}; % photochem_pho(i,3)=ArF_check{3}; % photochem_pho(i,4)=ArF_check{4}; % photochem_pho(i,5)=ArF_ check{5}; % photochem_pho(i,6)=x_new; % photochem_pho(i,7)=y_new; % photochem_pho(i,8)=layer; % arf_gen(depx,depy,1)=arf_gen(depx,depy,1)+ArF_check{4}; % arf_gen(depx,depy,2)=a rf_gen(depx,depy,2)+ArF_check{5}; % W=0; % break % end % end if abs(angle)>180 %preserves the +/ 180 convention angle= (360 abs(angle))*angle/abs(angle); %input('angle error found, was it resolved?') end x=x_new; y=y_new;

PAGE 169

169 end ; end end %volume corrections to absorbed photon %weights x_ax=linspace( width*1e 4/2,width*1e 4/2,wid th*res); %tissue width from beam center in cm x_ax2=linspace( width*1e 4/2,width*1e 4/2,width*res+1); y_ax=linspace(0,d_tot,round(d_tot*res)); %tissue depth in um dz=y_ax(2)*1e 4; %depth of a voxel in cm delta_x_ax=abs(pi*x_ax2(2:end).^2 pi*x_ax2(1:(end 1)).^ 2); %absorbed photon fraction matrix (cm 3) dfluence=repmat(delta_x_ax,length(y_ax),1); if strcmp(model, 'ArF, cheek' )==1 out2=cell(5,6); out2{1,1}=sum(pho_mat_ex(:))/no_pho; out2{1,3}=out2{1,1}*5.5e3/(5.5e3+35*3.5e4/534)*.023; out2{1,4}= out2{1,1}*(3.5e4*35/534)/(5.5e3+35*3.5e4/534)*(.104); out2{1,2}=out2{1,3}+out2{1,4}; out2{1,5}=out2{1,4}; out2{1,6}=out2{1,4}+35/534*out2{1,3}; % out2{1,2}=sum(photochem_pho(1:end,1))/no_pho; % out2{1,3}=sum(photochem_pho(1:end,2))/no_p ho; % out2{1,4}=sum(photochem_pho(1:end,3))/no_pho; % out2{1,5}=sum(photochem_pho(1:end,4))/no_pho; % out2{1,6}=sum(photochem_pho(1:end,5))/no_pho; % for i=2:5 % for j=1:6 % if j==1 % if i==5 % out{i,j}=sum(sum(pho_mat_ex(1:end,(res*d3(i 1)+1):(res*d2(i)))))/no_pho; % else % out{i,j}=sum(sum(pho_mat_ex(1:end,(res*d3(i 1)+1):(res*d2(i 1)))))/no_pho; % end % else % holder=photochem_pho(1:end,8).*photochem_pho(1:end,(j 1)); % if i==5 % ind=find(holder>=4); % else % ind=find(holder==(i 1)); % end % out2{i, j}=length(ind)/no_pho; % end % end % end % specular=sum(reflect(1:end,1)) % weight_out=sum(fluor_pho(1:end,1)) % photochem=sum(photochem_pho(1:end,1)) % peptides_broken=sum(photochem_pho(1:end,2))

PAGE 170

170 % aromatics_aas _broken=sum(photochem_pho(1:end,3)) % abs355_broken=sum(photochem_pho(1:end,4)) % fluor355_broken=sum(photochem_pho(1:end,5)) % higher_layers=length(find(photochem_pho(1:end,8)>1)) if no_pho~=0 pho_mat_ex2=pho_mat_ex./(ma{1}{2}(1 )*dz*no_pho.*dfluence'); %fluor_gen2=fluor_gen./(dz*no_pho.*dfluence'); figure(4) %fluence and fluence rate and transport used interchangably %divide by dfluence which corrects for the radial cross sectional area pcolor(1e4* x_ax(2000:3000),y_ax(1:res*d2(1)),log10(pho_mat_ex2(2000:3000,1:re s*d2(1))')) xlabel( 'distance from center of the source ( \ mum)' ) ylabel( 'depth in tissue ( \ mum)' ) %title('ArF log(Fluence) (cm^{ 2}) in the Superficial Epithelium') shading interp ; colorbar; g=get(colorbar, 'YTick' ); set(colorbar, 'YTickLabel' ,10.^g) figure(5) mid_flu=(pho_mat_ex2(round(res*width*.5),1:res*d2(1))+pho_mat_ex2(round(res*w idth*.5)+1,1:res*d2(1))+pho_ma t_ex2(round(res*width*.5) 1,1:res*d2(1))+pho_mat_ex2(round(res*width*.5)+2,1:res*d2(1))+pho_mat_ex2(rou nd(res*width*.5) 2,1:res*d2(1)))/5; tot_flu=sum(pho_mat_ex2); size(tot_flu) size(y_ax(1:res*d2(1))) beer_flu=tot_flu(1)*e xp( 1e 4*(ma{1}{2}(1)+ms{1}{2}(1)).*y_ax(1:res*d2(1))); [Y,I]=max(mid_flu); cross_flu=pho_mat_ex2(1:end,I(1)); plot(y_ax(1:res*d2(1)),tot_flu(1:res*d2(1)), r' ,y_ax(1:res*d2(1)),beer_flu, -k' ); xlabel( 'tissue depth ( \ mum)' ) ylabel( 'Total Fluence Rate, (cm^{ 2})' ) figure(6) semilogy(x_ax,cross_flu, 'r ) xlabel( 'radial distance from source' ) ylabel( 'Fluence Rate, (cm^{ 2})' ) figure(7) semilogy(y_ax(1:res*d2(1)),tot_flu(1: res*d2(1)), r' ,y_ax(1:res*d2(1)),beer_flu, -k' ) xlabel( 'tissue depth ( \ mum)' ) ylabel( 'Fluence Rate, (cm^{ 2})' ) %now adjust the mua and fluorescence response for super epithelium mua_old=mua_epi([355:40:555 350:50:650],[4 3 2.5 2 1.7 1.5 4 3.2 2.5 2 1.8 1.7 1.6],lambx); fluoro_old=Keratin; %mua_bkgd=0.244+85.3*exp( (Keratin{1} 154)/66.2); %figure(8)

PAGE 171

171 %semilogy(mua_old{1},mua_old{2},fluoro_old{1},fluoro_old{2},Keratin{1},mua_bk gd) %find a line of best fit to the experimental data using beer's law temp_mask1=pho_mat_ex(1:end,1:res*d2(1))/no_pho*(3.5e4*35/534)/(5.5e3+35*3.5e 4/534)*(.104); temp_mask2=pho_mat_ex(1:end,1:res*d2(1))/no_pho*5.5e3/(5.5e3+35*3.5e4/534)*.0 23*35/534 +pho_mat_ex(1:end,1:res*d2(1))/no_pho*(3.5e4*35/534)/(5.5e3+35*3.5e4 /534)*(.104); % tm1=temp_mask1(:,1:10); % tm2=temp_mask2(:,1:10); % cut1=find(sum(tm1,2)>0); % cut2=find(sum(tm2,2)>0); % for i=1:length(cut1) % t=find(temp_mask1(cut1(i),:)>0); % if length(t)>1 % t2=log(temp_mask1(cut1(i),t)); % t3= 1e 4*(ma{1}{2}(1)+ms{1}{2}(1))*t/res; % p=polyfit(t3,t2,1); % fit1=exp(p(2)).*exp( 1e 4*(ma{1}{2}(1)+ms{1}{2}(1))*((1:res*d2(1)) 1)/res); % temp_mask1(cut1(i),:)=fit1; % else % fit1=temp_mask1(cut1(i),t).*exp( 1e 4*(ma{1}{2}(1)+ms{1}{2}(1))*((1:res*d2(1)) 1)/res); % temp_mask1( cut1(i),:)=fit1; % end % end % for i=1:length(cut2) % t=find(temp_mask2(cut2(i),:)>0); % if length(t)>1 % t2=log(temp_mask2(cut2(i),t)); % t3= 1e 4*(ma{1}{2}(1)+ms{1}{2}(1) )*t/res; % p=polyfit(t3,t2,1); % fit1=exp(p(2)).*exp( 1e 4*(ma{1}{2}(1)+ms{1}{2}(1))*((1:res*d2(1)) 1)/res); % temp_mask2(cut2(i),:)=fit1; % else % fit1=temp_mask2(cut2(i),t).*exp( 1e 4*(ma{1}{2}(1)+ms{1}{2}(1))*((1:res*d2(1)) 1)/res); % temp_mask2(cut2(i),:)=fit1; % end % end dz2=y_ax(2)*1e 6; %depth of a voxel in m delta_x_ax2=abs(pi*1e 2*x_ax2(2:end).^2 pi*1e 2*x_ax2(1:(end 1)).^ 2); %radial area of a voxel in m^2 dfluence2=repmat(delta_x_ax2,length(y_ax(1:res*d2(1))),1); k_mask1=6.022e23.*dz2.*(35/534).*ones(size(pho_mat_ex2(1:end,1:res*d2(1)))).* dfluence2'; %amount of molecules k_mask2=k_mask1; pulse =2.4e17; %number of photons for an actual experiment post_drop1=(k_mask1 pulse*temp_mask1)./k_mask1; %multiplier for mu_a [row,col]=find(post_drop1<=0); post_drop1(row,col)=0;

PAGE 172

172 post_drop2=(k_mask2 pulse*temp_mask2)./k_mask2; %mu ltiplier for fluorescence [row,col]=find(post_drop2<=0); post_drop2(row,col)=0; nf=7; figure(8) % mid_filt=filter2(ones(nf)/(nf.^2),post_drop2,'valid'); % post_drop2(4:end 3,4:end 3)=mid_filt(1:end,1:end); % mid_filt=filter2(ones(nf)/(nf.^2),post_drop1,'valid'); % post_drop1(4:end 3,4:end 3)=mid_filt(1:end,1:end); pcolor(post_drop1') shading interp colorbar out=cell(3,1); out{1}=post_drop1'; out{2}=post_drop2'; out{3}=out2; end elseif strcmp(model, 'test' )==1 || strcmp(model, 'test2' )==1 specular=sum(reflect(1:end,1)); weight_out=sum(fluor_pho(1:end,1)); drx=0:.01:(1e 4*width/2); %in cm delta_drx=pi*drx(2:en d).^2 pi*drx(1:(end 1)).^2; f2=1e 4*fluor_pho(1:end,3) (.5*width*1e 4); %convert to cm and radial drW=zeros((length(drx) 1),1); for step=1:(length(drx) 1) ind2=find(abs(f2)>=drx(step) & abs(f2)1); fr2=find(fluor_pho(1:end,2)==1); out2{1}=sum(reflect(1:end,1))/no_pho; %to tal reflected off of surface, a bit of an outlier out2{2}=sum(fluor_pho(1:end,1))/no_pho; %total diffuse reflectance out2{3}=sum(fluor_pho(fr2,1))/no_pho; %total 355 out

PAGE 173

173 out2{4}=sum(fluor_pho(fr,1))/no_pho; %total diffuse fluorescence reflectan ce out3=cell(8,1); ref=sum(reflect(1:end,1)); rem=sum(fluor_pho(1:end,1)); out3{1}=ref/(ref+rem); %total reflected off of surface, a bit of an outlier out3{2}=rem/(ref+rem); %total diffuse reflectance out3{3}=sum(fluor_pho(fr2,1))/( ref+rem); %total 355 out out3{4}=sum(fluor_pho(fr,1))/(ref+rem); %total diffuse fluorescence reflectance for i=2:5 %layer if i==5 fr=find(fluor_pho(1:end,2)>1); dummy=fluor_pho(fr,[1 5]); fr2=find(dummy(1: end,2)>=4); out2{8}=sum(dummy(fr2,1))/no_pho; out3{8}=sum(dummy(fr2,1))/(ref+rem); out1{1,i}=sum(sum(pho_mat_ex(1:end,(round(res*d2(end 2))+1):round(res*d2(end)),2)))/no_pho; out1{2,i}=sum(sum(fluor_gen(1:end ,(round(res*d2(end 2))+1):round(res*d2(end)))))/no_pho; out1{3,i}=out1{2,i}/(out1{1,i}+out1{2,i}); else fr=find(fluor_pho(1:end,2)>1); dummy=fluor_pho(fr,[1 5]); fr2=find(dummy(1:end,2)==(i 1)); out2{i+3}=sum(dummy(fr2,1))/no_pho; %sort by layer of origin for remitted fluorescence out3{i+3}=sum(dummy(fr2,1))/(ref+rem); %sort by layer of origin for remitted fluorescence out1{1,i}=sum(sum(pho_mat_ex(1:end,(round(res*d3(i 1))+1):round(res*d2(i 1)),2)))/no_pho; out1{2,i}=sum(sum(fluor_gen(1:end,(round(res*d3(i 1))+1):round(res*d2(i 1)))))/no_pho; out1{3,i}=out1{2,i}/(out1{1,i}+out1{2,i}); end end % pho_mat_ex2=pho_mat_ex(1:end,1:end,1 )./(dz*no_pho.*dfluence');%need division by mua!!! (handled in the loop bc of fluorescence) % fluor_gen2=fluor_gen./(dz*no_pho.*dfluence'); % figure(5) % bar([out2{1:8}]) % %title('sources of detectable photons') % set(gca,'XTickLabel', {'Surface Reflected', 'Total Remittance', 'Excitation', 'Fluorescence','Super Ep Fluor','Inter Ep Fluor','Basal Ep Fluor','Stroma Fluor'}) % ylabel('fraction of photons launched') figure(6) Y=[out1{1,1:end};out1{2,1:end};out1{3,1:end}]; bar (Y) %title('per layer fluorescence metrics') set(gca, 'XTickLabel' ,{ 'photons absorbed' 'fluorescence generated' 'layer efficieny (gen/(abs+gen))' }) ylabel( 'fraction of photons launched' ) legend( 'total' 'super ep' 'inter ep' 'basal ep' 'str oma' ) % figure(7)

PAGE 174

174 % bar([out3{1:8}]) % title('sources of detectable photons') % set(gca,'XTickLabel',{'Surface Reflected', 'Total', 'Excitation', 'Fluorescence','Super Ep Fluor','Inter Ep Fluor','Basal Ep Fluor','Stroma Fluor'}) % ylabe l('fraction of detectable photons') figure(8) X=[out2{4:8}]./[out1{2,1:end}]; bar([out1{2,1:end};out2{4:8}]) legend( 'total' 'super ep' 'inter ep' 'basal ep' 'stroma' ) set(gca, 'XTickLabel' ,{ 'fluorescence generated' 'fluorescence remitte d' }) ylabel( 'fraction of photons launched' ) %now pull spectra from the total fluor remittance and the %layer by layer data (might need to widen dlambda the reduce noise %NADH{1} are my relavent wavelengths out4=cell(6,2); out4{6,1}= NADH{1}; out4{6,2}=NADH{1}; spect_tot=zeros(length(NADH{1}),1); spect1=zeros(length(NADH{1}),1); spect2=zeros(length(NADH{1}),1); spect3=zeros(length(NADH{1}),1); spect4=zeros(length(NADH{1}),1); spect_totb=zeros(length(NADH{1}) ,1); spect1b=zeros(length(NADH{1}),1); spect2b=zeros(length(NADH{1}),1); spect3b=zeros(length(NADH{1}),1); spect4b=zeros(length(NADH{1}),1); for i=2:length(NADH{1}) fr=find(fluor_pho(1:end,2)==i); %sorts by wavelengths sp ect_tot(i)=sum(fluor_pho(fr,1)); dummy=fluor_pho(fr,[1 5 3]); fr1=find(dummy(1:end,2)==1); fr2=find(dummy(1:end,2)==2); fr3=find(dummy(1:end,2)==3); fr4=find(dummy(1:end,2)>=4); spect1(i)=sum(dummy(fr1,1)); spect2(i)=sum(dummy(fr2,1)); spect3(i)=sum(dummy(fr3,1)); spect4(i)=sum(dummy(fr4,1)); frb=find(abs(dummy(1:end,3))<=5000); dummy2=dummy(frb,[1 2]); spect_totb(i)=sum(dummy2(:,1)); fr1b=find(dummy2(1:e nd,2)==1); fr2b=find(dummy2(1:end,2)==2); fr3b=find(dummy2(1:end,2)==3); fr4b=find(dummy2(1:end,2)>=4); spect1b(i)=sum(dummy2(fr1b,1)); spect2b(i)=sum(dummy2(fr2b,1)); spect3b(i)=sum(dummy2(fr3b,1)); spect4b(i)=sum(dummy2(fr4b,1)); end nf=7; out4{1,1}=filter(ones(1,nf)/nf,1,spect_tot/no_pho); out4{2,1}=filter(ones(1,nf)/nf,1,spect1/no_pho); out4{3,1}=filter(ones(1,nf)/nf,1,spect2/no_pho); out4{4,1}=filter(ones(1,nf)/nf,1,spect3/ no_pho);

PAGE 175

175 out4{5,1}=filter(ones(1,nf)/nf,1,spect4/no_pho); out4{1,2}=filter(ones(1,nf)/nf,1,spect_totb/no_pho); out4{2,2}=filter(ones(1,nf)/nf,1,spect1b/no_pho); out4{3,2}=filter(ones(1,nf)/nf,1,spect2b/no_pho); out4{4,2}=filter(ones(1,n f)/nf,1,spect3b/no_pho); out4{5,2}=filter(ones(1,nf)/nf,1,spect4b/no_pho); out=out4; figure(9) plot(NADH{1},out4{1},NADH{1},out4{2},NADH{1},out4{3},NADH{1},out4{4},NADH{1}, out4{5}) xlabel( 'wavelength (nm)' ) ylabel( 'Relative Flu orescence Intensity (au)' ) legend( 'Total' 'Super Ep' 'Inter Ep' 'Basal Ep' 'Stroma' ) %ylabel('fractional photon count (wrt total injected)'); axis tight length(find(fluor_pho(:,5)==1)) length(find(fluor_pho(:,5)==2)) length(find(flu or_pho(:,5)==3)) length(find(fluor_pho(:,5)>=4)) %then fit to each layers data and sum the fits to see how it looks for %the total spectrum % figure(10)%fluence, in terms of photon weight currently % x_ax=linspace( width*1e 4/2,wid th*1e 4/2,width*res); % y_ax=linspace(0,d_tot,d_tot*res); % top=4*res*d2(4); % %pcolor(1e4*x_ax,y_ax(1:4*res*d2(4)),log10(medfilt2(pho_mat_ex2(1:end,1:4*res *d2(4))',[3,3]))) % %pcolor(1e4*x_ax,y_ax(1:top),log10(pho_mat_ex2(1:end,1:top)')) % pcolor(1e4*x_ax,y_ax(1:round(top)),log10(filter2(ones(3)/9,pho_mat_ex2(1:end, 1:round(top))'))) % %colormap gray % shading interp; % colorbar; % g=get(colorbar,'YTick'); % set(colorbar,'YTickLabel',10.^g) % xlabel('distance from center of the source ( \ mum)') % ylabel('depth in tissue ( \ mum)') % %title('Total photon log(Fluence) (cm^{ 2}) in the epithelium and upper stroma') % hold on % plot(1e4*[x_ax(1) x_ax(end)],[d2(1) d2(1)],'k -',1e4*[x_ax(1) x_ax(end)],[d2(3) d2(3)],'k -',1e4*[x_ax(1) x_ax(end)],[d2(2) d2(2)],'k -',1e4*[x_ax(1) x_ax(end)],[d2(4) d2(4)],'k -',[ 500 500],[0 y_ax(round(top))],'k -',[500 500],[0 y_ax(round(top))],'k -') % hold off end end %this is the end of the main function

PAGE 176

176 function out =fresnel(n1,n2,theta) %returns P(refraction) and P(reflection) for propagation into 2 from 1 out=cell(2); %first make sure the angle is usable (ie: + and less than 90) if abs(theta)==90 || abs(theta)==270 out{1,1}=1; out{2 ,1}=0; out{1,2}=theta; elseif n1==n2 %check for actual boundary out{1,1}=1; out{2,1}=0; out{1,2}=theta; %transmission at the same directional angle %out{2,2}=sqrt( 1); else %condition theta to conform w stds for fresnel eq and snells law if abs(theta)>90 thetab=180 abs(theta); elseif abs(theta)==180 thetab=0; else thetab=abs(theta); end if (n1 > n2) && ~isreal(a sind(n1*sind(thetab)/n2)) %check for TIR out{1,1}=0; out{2,1}=1; %out{1,2}=sqrt( 1); if theta==0 out{2,2}=180; elseif theta==180 out{2,2}=0; else out{2,2}=theta*(180 abs(theta))/abs(theta); end else %actually det R and T here theta2b=asind(n1*sind(thetab)/n2); R1=((n1*cosd(thetab) n2*cosd(theta2b))./(n1*cosd(thetab)+n2*cosd(theta2b))).^2; R2=((n2*cosd(thetab) n1*cosd(theta2b))./(n2*cosd(thetab)+n1*cosd(theta2b))).^2; R=(R1+R2)/2; T=1 R; r=rand; if r<=R %reflection out{1,1}=0; out{2,1}=1; else %transmission out{1,1}=1; out{2,1}=0; end out{2,2}=theta*(180 abs(theta))/abs(theta); if theta > 0 && theta < 90 out{1,2}=theta2b; elseif theta > 90 && thet a < 180

PAGE 177

177 out{1,2}=180 theta2b; elseif theta > 90 && theta < 0 out{1,2}= theta2b; elseif theta > 180 && theta < 90 out{1,2}= (180 theta2b); elseif abs(theta)==0 out{2,2}=180; out{1,2}=0; elseif abs(theta)==180 out{2,2}=0; out{1,2}=180; end %now consider how to re condiiton the angle to conform w the %direction std s of the MC code (should be done above end end end function out=fluoro_fit(wave, em, ex) out=cell(2,1); if wave(1) ~= ex xgrid=(wave(1):2:580) wave(1); wave=wave wave(1)+1; modelFun = @(p,x) p(3).*(x./p(1)).^(p (2) 1).*exp( (x./p(1)).^p(2)); startingVals=[10 2 1]; coefEsts=nlinfit(wave,em,modelFun,startingVals); temp_em=modelFun(coefEsts,xgrid); buffer=length(ex:2:580) length(xgrid); out{1}=ex:2:580; out{2}=[zeros(b uffer,1); temp_em']'; else wave=wave ex+1; xgrid=(ex:2:580) ex; modelFun = @(p,x) p(3).*(x./p(1)).^(p(2) 1).*exp( (x./p(1)).^p(2)); startingVals=[10 2 1]; coefEsts=nlinfit(wave,em,modelFun,startingVals); temp_em=modelFun(coefEsts,xgrid); out{1}=ex:2:580; out{2}=temp_em; end end function out=mus_spec(wave, mus, ex) out=cell(3,1); xgrid=ex:2:580; %modelFun = @(p,x) p(1).*x.^( p(2))+p(3).*x.^( 4); %startingVals=[73 .2 2e1 2]; modelFun = @(p,x) p(1).*x.^( .22)+p(2).*x.^( 4); startingVals=[75 2e12]; coefEsts=nlinfit(wave,mus,modelFun,startingVals); out{2}=modelFun(coefEsts,xgrid); out{1}=xgrid; out{3}=[coefEsts(1) .22 coefEsts(2) 4]; end function o ut=mua_epi(wave, mua, ex)

PAGE 178

178 out=cell(3,1); xgrid=250:2:1000; modelFun = @(p,x) p(1)+p(2).*exp( (x p(3))./p(4)); startingVals=[.244 85.3 154 66.2]; coefEsts=nlinfit(wave,mua,modelFun,startingVals); out{1}=xgrid; out{2}=modelFun(coe fEsts,xgrid); out{3}=coefEsts; end function out=launch_angle(NA, n) r= 1+2*rand; out=r.*asind(NA./n); end function out=pfluor(fluor, mua, wave, draw, qy, ms) out=cell(2,2); dy=fluor(2:end) fluor(1:(end 1)); dx=wave(2:end) wave(1 :(end 1)); auc=sum(dy.*dx)/2; %p=auc./mua;%probability of fluorescence (or should I leave mu_a total out of it?), account for other absorbers in this case p=qy*mua/(mua+ms); r=rand; if r<=p out{1,1}=1; out{2,1}=0; wave_new=draw(round((length(draw) 1)*rand+1)); out{1,2}=wave_new; out{2,2}=find(wave==wave_new); else out{1,1}=0; out{2,1}=1; end end function out=make_draw(fluor, wave) %makes a distribution of wavelengths for em ission that can be drawn with "rand" I=find(fluor>realmin); [Y,I2]=min(fluor(I)); fluor=round(fluor./Y(1)); draw=zeros(sum(fluor),1); top=cumsum(fluor); bottom=[1,(top(1:(end 1))+1)]; length(top) length(wave) for n=1:len gth(top) if top(n)==0 %skip that wavelength else draw(bottom(n):top(n))=wave(n); end end out=draw; end

PAGE 179

179 function out=ms193(mua,wave,mus) out=cell(2,1); %implied reduced scattering coeff at 193nm, assuming penetration depth %of all tissues approx 1um %mu_eff=10e3;%cm 1 prolly only holds in the mus > mua range %mus_reduced=mu_eff^2/(3*mua) mua p=polyfit(wave.^ 4,mus,1); out{1}=p(1).*(193^ 4)+p(2); out{2}=p; end function out =pArF(layer, W, ma, ms) %returns flags for any photochem being done and if the optical %properties (mua(355), emission from 355ex) have changed and how for %each out=cell(5,1); %flags for photochem, 193pep, 193aro, 355abs, 355fluor out{1 }=0; out{2}=0; out{3}=0; out{4}=0; out{5}=0; p_pchem=W*ma/(ma+ms); r=rand; if r<=p_pchem %some photochem was done out{1,1}=1; %now figure out what happened r2=rand; if layer==1 %keratin p_aro_k=35/534; if r2<=p_aro_k out{3,1}=1; %193aro out{4,1}=1; %355abs out{5,1}=1; %355fluor else out{2,1}=1; %193pep r3=rand; if r 3<=p_aro_k out{5,1}=1; %355fluor end end elseif layer==2 || layer==3 %fad & nadh but cytosolic proteins absorb at 193 before those fluorophores p_aro_cyto=.1/1; if r2<=p_aro_cyto out{4,1}=1; %355abs out{3,1}=1; %193aro else out{2,1}=1; %193pep end elseif layer==4 || layer==4 %collagen (20%) and other proteins (5%) and 75% water p_coll=20/(20+5) ; if r2<=p_coll p_crosslink=.2; %??? Need a reference for this r3=rand; if r3<=p_crosslink

PAGE 180

180 out{4,1}=1; %355abs out{5,1}=1; %355fluor else p_aro_coll=.1/1; %??? need a ref for this r3=rand; if r3<=p_aro_coll out{4,1}=1; %355abs out{3,1}=1; %193aro else out{2,1}=1; %193pep end end else p_aro_cyto=.1/1; r3=rand; if r3<=p_aro_cyto out{4,1}=1; %355abs out{3,1}=1; %193aro else out{2,1}=1; %193pep end end end else out{1,1}=0; out{2,1}=0; out{3,1}=0; out{4,1}=0; out{5,1}=0; end end

PAGE 181

181 LIST OF REFERENCE S 1. M. C. Skala, G. M. Palmer, C. Zhu, Q. Liu, K. M. Vrotsos, C. L. Marshek Stone, A. Gendron Fitzpatrick, and N. Ramanujam, "Investigation of fiber optic probe designs for optical spectroscopic diagnosis of epithelial pre cancers," Lasers in Surge ry and Medicine 34 25 38 (2004). 2. S. E. Waggoner, "Cervical Cancer," Lancet 361 2217 2225 (2003). 3. M. W. Lingen, J. R. Kalmar, T. Karrison, and P. M. Speight, "Critical evaluation of diagnostic aids for the detection of oral cancer," Oral Oncology 44 10 22 (2008). 4. S. N. Rogers, J. S. Brown, J. A. Woolgar, D. Lowe, P. Magennis, R. J. Shaw, D. Sutton, D. Errington, and D. Vaughan, "Survival following primary surgery for oral cancer," Oral Oncology 45 201 211 (2009). 5. W. Saman, "Global epidemiolog y of oral and oropharyngeal cancer," Oral Oncology 45 309 316 (2009). 6. R. Sankaranarayanan, K. Dinshaw, B. M. Nene, K. Ramadas, P. O. Esmy, K. Jayant, T. Somanathan, and S. Shastri, "Cervical and oral cancer screening in India," Journal of Medical Scree ning 13 35 38 (2006). 7. C. A. Squier, and M. J. Kremer, "Biology of Oral Mucosa and Esophagus," J Natl Cancer Inst Monogr 29 7 15 (2001). 8. T. A. Winning, and G. C. Townsend, "Oral mucosa embryology and histology," Clinics in Dermatology 18 (2000). 9 I. Pavlova, C. R. Weber, R. A. Schwarz, M. D. Williams, A. M. Gillenwater, and R. Richards Kortum, "Fluorescence Spectroscopy of Oral Tissue: Monte Carlo Modeling with Site Specific Tissue Properties," Journal of Biomedical Optics 14 (2009). 10. D. R. Eyre, M. A. Paz, and P. M. Gallop, "Cross linking in collagen and elastin," Ann Rev Biochem 53 717 748 (1984). 11. L. J. Kleinsmith, Priniciples of Cancer Biology (Benjamin Cummings, San Francisco, 2006). 12. "WHO Collaborating Centre for Oral Precancerou s Lesions. Definition of leukoplakia and related lesions: an aid to studies on oral precancer," Oral Surg Oral Med Oral Pathol 46 518 539 (1978).

PAGE 182

182 13. L. L. Patton, J. B. Epstein, and A. R. Kerr, "Adjunctive Techniques for Oral Cancer Examination and Lesio n Diagnosis: A Systemic Review of the Literature," Journal of the American Dental Association 139 896 905 (2008). 14. V. Thomas, S. R. Devi, V. Jeyaseelan, and L. Jeyseelan, "Mucosal Disorders with Oral Epithelial Dysplasia risk Development of a simple s creening tool for general health care setting," Oral Oncology in press (2012). 15. P. Wilder Smith, J. Holtzman, J. Epstein, and A. Le, "Optical diagnostics in the oral cavity: an overview," Oral Diseases 16 717 728 (2010). 16. K. Huff, P. C. Stark, and L. W. Solomon, "Sensitivity of direct tissue fluorescence visualization in screening for oral premalignant lesions in general practice," General Dentistry, 34 38 (2009). 17. K. H. Awan, P. R. Morgan, and S. Warnakulasuriya, "Evaluation of an autofluoresce nce based imaging system (VELscope) in the detection of oral potentially malignant disorderes and benign keratoses," Oral oncology 47 274 277 (2011). 18. C. Scully, J. V. Bagan, C. Hopper, and J. B. Epstein, "Oral cancer: current and future diagnostic tec hniques," Am J Dent 21 199 209 (2008). 19. M. A. Jaber, "Oral epithelial dysplasia in non users of tobacco and alcohol: An analysis of clinicopathological characteristics and treatment outcome," J Oral Sci 52 13 21 (2010). 20. E. L. Truelove, D. Dean, S. Maltby, M. Griffith, K. Huggins, M. Griffith, and S. Taylor, "Narrow band (light) imaging of oral mucosa in routine dental patients. Part I: Assessment of value in detection of mucosal changes," General Dentistry, 281 291 (2011). 21. A. J. Welch, and M. J C. v. Gemert, Optical Thermal Response of Laser Irradiated Tissue (Springer, New York, 2011). 22. L. V. Wang, and H. I. Wu, Biomedical Optics (Wiley, 2007). 23. W. F. Cheong, S. A. Prahl, and A. J. Welch, "A Review of the Optical Properties of Biological Tissues," IEEE Journal of Quantum Electronics 26 2166 2185 (1990). 24. R. Steiner, "Laser Tissue Interactions," in Laser and IPL Technology in Dermatology and Aesthetic Medicine C. Raulin, and S. Karsai, eds. (Springer Verlag, Berlin, 2011), pp. 23 35.

PAGE 183

183 25. R. Richards Kortum, and E. Sevick Muraca, "Quantitative Optical Spectrscopy for Tissue Diagnosis," Annual Review of Physical Chemistry 47 555 606 (1996). 26. R. Alfano, D. Tata, J. Cordero, P. Tomashefsky, F. Longo, and M. Alfano, "Laser induced fluor escence spectroscopy from native cancerous and normal tissue," Quantum Electronics, IEEE Journal of 20 1507 1511 (1984). 27. A. Policard, "Etude sur les aspects offerts par des tumours experimentales examinees a la lumiere de Wood," CR Soc. Biol. 91 742 743 (1924). 28. S. K. Chang, N. Marin, M. Follen, and R. Richards Kortum, "Model based analysis of clinical fluorescnce spectrscopy for in vivo detection of cervical intraepithelial dysplasia," Journal of Biomedical Optics 11 1 12 (2006). 29. I. Pavlova, C. R. Weber, R. A. Schwarz, M. Williams, A. El Naggar, A. Gillenwater, and R. Richards Kortum, "Monte Carlo model to describe depth selective fluorescence spectra of epithelial tissue: applications for diagnosis of oral precancer," Journal of Biomedical Op tics 13 (2008). 30. C. R. Weber, R. A. Schwarz, E. N. Atkinson, D. D. Cox, C. MacAulay, M. Follen, and R. Richards Kortum, "Model based analysis of reflectance and fluorescence spectra for in vivo detection of cervical dysplasia and cancer," Journal of B iomedical Optics 13 1 10 (2008). 31. I. E. Kochevar, "Biological Effects of Excimer Laser Radiation," Proceedings of the IEEE 80 833 837 (1992). 32. G. H. Pettit, and M. N. Ediger, "Corneal tissue absorption coeffients for 193 and 213 nm ultraviolet rad iation," Applied Optics 35 3386 3391 (1996). 33. H. Gorner, "Photochemistry of DNA and related biomolecules: quantum yields and consequences of photoionization," Journal of Photochemistry and Photobiology 26 117 139 (1994). 34. M. Mulcahy, J. G. McInerne y, D. N. Nikogosyan, and H. Gorner, "193 nm Photolysis of Aromatic and Aliphatic Dipeptides in Aqueous Solution: Dependence of Decomposition Quantum Yield on the Amino Acid Sequence," Biol. Chem. 381 1259 1262 (2000). 35. D. N. Nikogosyan, and H. Gorner, "Laser induced photodecomposition of amino acids and peptides: extrapolation to corneal collagen," IEEE J. Selected Topics Quantum Electron. 5 1107 1115 (1999).

PAGE 184

184 36. S. E. Smith, L. M. Shanyfelt, K. D. Buchanan, and D. W. Hahn, "Differential laser induced perturbation spectroscopy using a deep ultraviolet excimer laser," Opt. Lett. 36 2116 2118 (2011). 37. S. E. Smith, K. D. Buchanan, and D. W. Hahn, "Excimer laser induced perturbation spectroscopy: an approach to biosensing," journal article 7674 767409 (2010). 38. R. T. Kozikowski, S. E. Smith, J. A. Lee, W. L. Castleman, B. S. Sorg, and D. W. Hahn, "Differential laser induced perturbation spectroscopy: demonstration of a new spectroscopic technique in a murine model of emerging skin pathology," Journal of Biomedical Optics 17 067002 (2012). 39. M. Brownstein, R. A. Hoffman, R. Levenson, T. E. Milner, M. L. Dowell, P. A. Williams, G. S. White, A. K. Gaigalas, and J. C. Hwang, "Biophotonic Tools in Cell and Tissue Diagnostics," Journal of Research of the National Institute of Standards and Technology 112 139 152 (2007). 40. J. R. Lakowicz, Principles of Fluorescence Spectroscopy (Springer, New York, 2006). 41. B. W. Chwirot, M. Kowalska, N. Sypniewska, Z. Michniewicz, and M. Gradziel, "Spectrally resolved fluorescence imaging of human colonic adenomas," Journal of Photochemistry and Photobiology B: Biology 50 174 183 (1999). 42. D. C. G. De Veld, M. J. H. Witjes, H. J. C. M. Sterenborg, and J. L. N. Roodenburg, "The status of in vivo autofluorescence spec troscopy and imaging for oral oncology," Oral oncology 41 117 131 (2005). 43. J. Haringsma, and G. N. J. Tytgat, "Fluorescence and autofluorescence," Bailliere's Clinical Gastroenterology 13 1 10 (1999). 44. M. A. Funovics, R. Weissleder, and U. Mahmood, "Catheter based in Vivo Imaging of Enzyme Activity and Gene Expression: Feasibility Study in Mice1," Radiology 231 659 666 (2004). 45. N. Ramanujam, "Fluorescence Spectroscopy of Neoplastic and Non neoplastic Tissues," Neoplasia 2 89 117 (2000). 46. L. Gortzak Uzan, W. Jimenez, S. Nofech Mozes, N. Ismiil, M. A. Khalifa, V. Dub, B. Rosen, J. Murphy, S. Laframboise, and A. Covens, "Sentinel lymph node biopsy vs. pelvic lymphadenectomy in early stage cervical cancer: Is it time to change the gold standard? ," Gynecologic Oncology 116 28 32 (2010).

PAGE 185

185 47. M. F. Mitchell, S. B. Cantor, N. Ramanujam, G. Tortolero Luna, and R. Richards Kortum, "Fluorescence Spectroscopy for Diagnosis of Squamous Intraepithelial Lesions of the Cervix," Obstetrics & Gynecology 93 4 62 470 (1999). 48. V. Ravery, S. Dominique, X. Panhard, M. Toublanc, L. Boccon Gibod, and L. Boccon Gibod, "The 20 Core Prostate Biopsy Protocol -A New Gold Standard?," The Journal of Urology 179 504 507 (2008). 49. C. Brookner, U. Utzinger, M. Follen, R. Richards Kortum, D. Cox, and E. N. Atkinson, "Effects of biographical variables on cervical fluorescence emission spectra," Journal of Biomedical Optics 8 479 483 (2003). 50. S. F. Martin, A. D. Wood, M. M. McRobbie, M. Mazilu, M. P. McDonald, I. D. W. S amuel, and C. S. Herrington, "Fluorescence spectroscopy of an in vitro model of human cervical precancer indentifies neoplastic phenotype," International Journal of Cancer 120 1964 1970 (2007). 51. A. J. Bailey, R. G. Paul, and L. Knott, "Mechanisms of ma turation and ageing of collagen," Mech Ageing Dev 106 1 56 (1998). 52. K. Calabro, A. Curtis, J. R. Galarneau, T. Krucker, and I. J. Bigio, "Gender variations in the optical properties of skin in murine animal models," Journal of Biomedical Optics 16 011 008 (2011). 53. N. Kollias, R. Gillies, M. Moran, I. E. Kochevar, and R. R. Anderson, "Endogenous Skin Fluorescence Includes Bands that may Serve as Quantitative Markers of Aging and Photoaging," 111 776 780 (1998). 54. A. C. Meyer, B. J. Meyer, J. F. Mo rrison, and W. J. Pepler, "Calcium, Collagen, Elastin and hexosamine levels in the arteries of whites and Bantu," S A Med J 39 1017 1020 (1965). 55. A. Moragas, M. Garcia Bonafe, M. Sans, N. Toran, P. Huguet, and C. Martin Plata, "Image analysis of dermal collagen changes during skin aging," Anal Quant Cytol Histol 20 493 499 (1998). 56. T. Rechberger, K. Postawski, J. A. Jakowicki, Z. Gunja Smith, and J. F. Woessner, "Role of fascial collagen in stress urinary incontinence," Am J Obst Gynecol 179 1511 1 514 (1998). 57. S. H. Tseng, P. Bargo, A. Durkin, and N. Kollias, "Chromophore concentrations, absorption andscattering properties of human skin in vivo," Opt. Express 17 14599 14617 (2009).

PAGE 186

186 58. D. Chakravarti, J. C. Pelling, E. L. Cavalieri, and E. G. Ro gan, "Relating aromatic hydrocarbon induced DNA adducts and c H ras mutations in mouse skin papillomas: The role of apurinic sites," Prc Natl Acad Sci 92 10422 10426 (1995). 59. L. Coghlan, U. Utzinger, R. Drezek, D. Heintzelmann, A. Zuluaga, C. Brookner, R. Richards Kortum, I. Gimenez Conti, and M. Follen, "Optimal fluorescence excitation wavelengths for detection of squamous intra epithelial neoplasia: results from an animal model," Opt. Express 7 436 446 (2000). 60. L. Coghlan, U. Utzinger, R. Richards Kortum, C. Brookner, A. Zuluaga, I. Gimenez Conti, and M. Follen, "Fluorescence Spectroscopy of Epithelial Tissue Throughout the Dysplasia Carcinoma Sequence in an Animal Model: Spectroscopic Changes Precede Morphologic Changes," Lasers in Surgery and Med icine 29 1 10 (2001). 61. O. H. Iverson, "A course of very small doses of DMBA, each of them allegedl with no promoting potency, acts with clear synergistic effect as a strong promoter of DMBA initiated mouse skin carcinogenesis. A comparison of the tumor igenic and carcinogenic effects of DMBA (7,12 dimethylbenz alpha anthracene) and TPA (12 O tetradecanoyl phorbol 13 acetate) used as initiators and promoters in classical two stage experimental protocols," APMIS Suppl 41 1 38 (1994). 62. X. Li, J. Eckard, R. Shah, C. Malluck, and K. Frenkel, "Interleukin 1alpha Up Regulation In Vivo by a Potent Carcinogen 7,12 Dimethylbenz(a)anthracene (DMBA) and Control of DMBA induced Inflammatory Responses," Cancer Research 62 417 423 (2002). 63. M. A. Nelson, B. W. Fu tscher, T. Kinsella, J. Wymer, and G. T. Bowden, "Detection of mutant Ha ras genes in chemically initiated mouse skin epidermis before the development of benign tumors," Prc Natl Acad Sci 92 6398 6402 (1992). 64. M. Quintanilla, K. Brown, M. Ramsden, and A. Balmain, "Carcinogen specific mutation and amplification of Ha ras during mouse skin carcinogenesis," Nature 322 78 80 (1986). 65. A. C. Rencher, Methods of Multivariate Analysis (Wiley, New York, 1995). 66. K. Thumanu, W. Tanthanuch, D. Ye, A. Sangmal ee, C. Lorthongpanich, R. Parnpai, and P. Heraud, "Spectroscopic signature of mouse embryonic stem cell derived hepatocytes using synchrotorn Fourier transform infrared microspectroscopy," Journal of Biomedical Optics 16 057005 (2011).

PAGE 187

187 67. J. L. Jesneck, L. W. Nolte, J. A. Baker, C. E. Floyd, and J. Y. Lo, "Optimized approach to decision fusion of heterogeneous data for breast cancer diagnosis," Med Phys 33 2945 2954 (2006). 68. H. L. VanTrees, Detection, Estimation, and Modulation Theory (Part I) (Wiley, New York, 1968). 69. Z. I. Botev, J. F. Grotowski, and D. P. Kroese, "Kernel density estimation via diffusion," Annals of Statistics 38 2916 2957 (2010). 70. C. Zhu, E. S. Burnside, G. A. Sisney, L. R. Salkowski, J. M. Harter, B. Yu, and N. Ramanujam, "F luorescence Spectroscopy: An Adjunct Diagnostic Tool to Image Guided Core Needle Biopsy of the Breast," IEEE Transactions on Biomedical Engineering 56 2518 2528 (2009). 71. Y. Liao, L. W. Nolte, and L. M. Collins, "Decision Fusion of Ground Penetrating Ra dar and Metal Detector Algorithms A Robust Approach," IEEE Transactions on Geoscience and Remote Sensing 45 398 409 (2007). 72. G. Bottiroli, A. C. Croce, D. Locatelli, R. Marchesini, E. Pignoli, S. Tomatis, C. Cuzzoni, S. Di Palma, M. Dalfante, and P. S pinelli, "Natural 16 48 60 (1995). 73. C. Eker, R. Rydell, K. Svanberg, and S. Andersson Engels, "Multivariate analysis of laryngeal fluore scence spectra recorded in vivo," Lasers in Surgery and Medicine 28 259 266 (2001). 74. C. Y. Wang, T. Tsai, H. C. Chen, S. C. Chang, C. T. Chen, and C. P. Chiang, "Autofluorescence spectroscopy for in vivo diagnosis of DMBA induced hamster buccal pouch p re cancers and cancers," J Oral Path Med 32 18 24 (2003). 75. M. F. Haque, M. Harris, S. Meghji, and A. W. Barrett, "Immunolocalization of cytokines and growth factors in oral submucous fibrosis," Cytokine 10 713 719 (1998). 76. P. Diagaradjane, M. A. Ya seen, J. Yu, M. S. Wong, and B. Anvari, "Autofluorescence Characterization for the Early Diagnosis of Neoplastic Changes in DMBA/TPA Induced Mouse Skin Carcinogenesis," Lasers in Surgery and Medicine 37 382 395 (2005). 77. I. Pathak, N. L. Davis, Y. N. Hs iang, N. F. Quenville, and B. Palcic, "Detection of squamous neoplasia by fluorescence imaging comparing porfimer sodium fluorescence to tissue autofluorescence in the hamster cheek pouch model," American journal of surgery 170 423 426 (1995).

PAGE 188

188 78. C. Y. W ang, C. T. Chen, C. P. Chiang, S. T. Young, S. N. Chow, and H. K. Chiang, "A Probability based Multivariate Statistical Algorithm for Autofluorescence Spectroscopic Identification of Oral Carcinogenesis," Photochemistry and Photobiology 69 471 477 (1999). 79. G. M. Palmer, C. Zhu, T. M. Breslin, F. Xu, K. W. Gilchrist, and N. Ramanujam, "Comparison of Multiexcitation Fluorescence and Diffuse Reflectance Spectroscopy for the Diagnosis of Breast Cancer," IEEE Transactions on Biomedical Engineering 50 1233 1 242 (2003). 80. N. Ramanujam, M. Follen Mitchell, A. Mahadevan Jansen, S. Thomsen, G. Staerkel, A. Malpica, T. Wright, N. Atkinson, and R. Richards Kortum, "Cervical precancer detection using multivariate statistical algorithm based on laser induced fluore scence spectra at multiple excitation wavelengths," Photochem Photobiol 64 720 735 (1996). 81. G. R. Lanckreit, T. DeBie, N. Cristianini, M. I. Jordan, and W. S. Noble, "A statistical framework for genomic data fusion," Bioinformatics 20 (2004). 82. G. R. Lanckreit, M. Deng, N. Cristianini, M. I. Jordan, and W. S. Noble, "Kernel based data fusion and its application to protein function prediction in yeast," in Proceedings of the Pacific Symposium on Biocomputing (World Scientific Press, Kohala Coast, H awaii, 2004), pp. 300 311. 83. P. Pavlidis, J. Weston, J. Cai, and W. S. Noble, "Learning gene functional classifications from multiple data types," J Comp Biol 9 401 411 (2002). 84. C. S. Herrington, and M. Wells, Premalignant and malignant squamous lesi ons of the cervix (Churchill Livingstone, Edinburgh, 2002). 85. C. Scully, and S. Porter, "ABC of oral health: swellings and red, white, and pigmented lesions," British Medical Journal 321 225 228 (2000). 86. W. G. Shafer, M. K. Hine, and B. M. Levy, Text book of Oral Pathology (W. B. Saunders, Philadelphia, 1983). 87. R. Drezek, K. Sokolov, U. Utzinger, I. Boiko, A. Malpica, and M. F. R. Richards Kortum, "Understanding the contributions of NADH and collagen to cervical tissue fluorescence spectra: Modeling measurements, and implications," Journal of Biomedical Optics 6 385 396 (2001). 88. Q. Liu, C. Zhu, and N. Ramanujam, "Experimental validation of Monte Carlo modeling of fluorescence in tissues in the UV visible spectrum," Journal of Biomedical Optics 8 223 236 (2003). 89. Y. Wu, and J. Y. Qu, "Combined depth and time resolved autofluorescence spectroscopy of epithelial tissue," Optics Letters 31 1833 1835 (2006).

PAGE 189

189 90. Y. Wu, P. Xi, J. Y. Qu, T. H. Cheung, and M. Y. Yu, "Depth resolved fluorescence spe ctroscopy of normal and dysplastic cervical tissue," Opt. Express 13 382 387 (2005). 91. D. Y. Churmakov, I. V. Meglinski, S. A. Piletsky, and D. A. Greenhalgh, "Analysis of skin tissue spatial fluorescence distribution by the Monte Carlo simulation," Jou rnal of Physics D 36 1722 1728 (2003). 92. I. Pavlova, M. Williams, A. El Naggar, R. Richards Kortum, and A. Gillenwater, "Understanding the Biological Basis of Autofluorescence Imaging for Oral Cancer Detection: High Resolution Fluorescence Microscopy in Viable Tissue," Clinical Cancer Research 14 2396 2404 (2008). 93. T. Collier, M. Follen, A. Malpica, and R. Richards Kortum, "Sources of scattering in cervical tissue: determination of the scattering coefficient by confocal microscopy," Applied Optics 44 2072 2081 (2005). 94. S. L. Jacques, "Skin Optics Summary," Oregon Medical Laser Center News, (1998). 95. I. Georgakoudi, B. C. Jacobson, M. G. Muller, E. E. Sheets, K. Badizadegan, D. L. Carr Locke, C. P. Crum, C. W. Boone, R. R. Dasari, J. V. Dam, and M. S. Feld, "NAD(P)H and Collagen as in Vivo Quantitative Fluorescent Biomarkers of Epithelial Precancerous Changes," Cancer Research 62 682 687 (2002). 96. P. Marefat, and G. Shklar, "Experimental production of lingual leukoplakia and carcinoma," Oral S urg 44 578 586 (1977). 97. M. C. Skala, G. M. Palmer, K. M. Vrotsos, A. Gendron Fitzpatrick, and N. Ramanujam, "Comparison of a physical model and principal component analysis for the diagnosis of epithelial neoplasias in vivo using diffuse reflectance sp ectrscopy," Optics Express 15 7863 7875 (2007). 98. K. Sokolov, J. Galvan, A. Myakov, A. Lacy, R. Lotan, and R. Richards Kortum, "Realistic three dimensional epithelial tissue phantoms for biomedical optics," Journal of Biomedical Optics 7 148 156 (2002) 99. R. Zhang, W. Verkruysse, B. Choi, J. A. Viator, B. Jung, L. O. Svaasand, G. Aguilar, and J. S. Nelson, "Determination of human skin optical properties from spectrophotometric measurements based on optimization by genetic algorithms," Journal of Biome dical Optics 10 024030 (2005). 100. J. H. Baert, R. J. Veys, K. Ampe, and J. A. D. Boever, "The effect of sodium lauryl sulphate and triclosan on hamster cheek pouch mucosa," Int J Exp Pathol 77 73 78 (1996).

PAGE 190

190 101. R. J. Veys, J. H. Baert, and J. A. D. Bo ever, "Histological changes in the hamster cheek pouch epithelium induced by topical application of sodium lauryl sulphate Int J Exp Pathol 75 203 209 (1994). 102. C. M. Bishop, Pattern Recognition and Machine Learning (Spring Science+Business Media, Si ngapore, 2006). 103. T. A. Schonhoff, and A. A. Giordano, Detection and Estimation Theory and its Applications (Pearson Education, New Jersey, 2006). 104. G. J. Hahn, and S. S. Shapiro, Statistical Models in Engineering (John Wiley and Sons, New York, 1967 ). 105. N. L. Johnson, "Systems of frequency curves generated by methods of translation," Biometrika 36 149 176 (1949). 106. K. Tang, R. Wang, and T. Chen, "Towards Maximizing the Area Under the ROC Curve for Multi Class Classification Problems," Proceedi ngs of the Twenty Fifth AAAI Conference on Artificial Intelligence, 483 488 (2011). 107. A. Santoro, G. Pannone, M. Contaldo, F. Sanguedolce, V. Esposito, R. Serpico, L. L. Muzio, S. Papagerakis, and P. Bufo, "A Troubling Diagnosis of Verrucous Squamous Ce ll Carcinoma ("the Bad Kind" of Keratosis) and the Need of Clinical and Pathological Correlations: A Review of the Literature with a Case Report," Journal of Skin Cancer 2011 1 4 (2011).

PAGE 191

191 BIOGRAPHICAL SKETCH Raymond received his Ph.D. in the fall of 2 012 from the University of Florida J. Crayton Pruitt Department of Biomedical Engineering. While there, his research spanned multiple disciplines ; fusing imaging and spectroscopy, signal processing, nanotechnology, and molecular medicine into a synergist ic focus on improving cancer diagnosis and treatment. He is the recent recipient of a CIMIT Prize for Primary Healthcare Finalist Award for his work developing a new laser based differential spectroscopy technique for diagnosing superficial cancers. Other work at Florida includes developing next generation liposomal and polymeric nanoparticle drug formulation designed to selectively deliver therapeutic payloads in the tumor interstitium as well as describing a new functional imaging modality, blood supply t ime, to study how vessel growth responds to tumor formation or treatment. Before coming to UF, he worked at Advanced Liquid Logic, designing and validating of microfluidic cartridges for miniaturized point of care diagnostics. Prior to that, he was employe d as a scientist at GlaxoSmithKline developing embedded sensing solutions for pharmaceutical manufacturing, drug discovery, an b