Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Nodal disease burden of oral cancer in British Columbia and a novel approach for risk assessment Liu, Kelly Yi Ping 2016

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2016_may_liu_kellyyiping.pdf [ 6.56MB ]
Metadata
JSON: 24-1.0300166.json
JSON-LD: 24-1.0300166-ld.json
RDF/XML (Pretty): 24-1.0300166-rdf.xml
RDF/JSON: 24-1.0300166-rdf.json
Turtle: 24-1.0300166-turtle.txt
N-Triples: 24-1.0300166-rdf-ntriples.txt
Original Record: 24-1.0300166-source.json
Full Text
24-1.0300166-fulltext.txt
Citation
24-1.0300166.ris

Full Text

    NODAL DISEASE BURDEN OF ORAL CANCER IN BRITISH COLUMBIA AND A NOVEL APPROACH FOR RISK ASSESSMENT  by KELLY (YI PING) LIU B.Sc., Simon Fraser University, 2010    A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF  THE REQUIREMENT FOR THE DEGREE OF  MASTER OF SCIENCE  in  THE FACULTY OF GRADUATE AND POSTDOCTORAL STUDIES (Craniofacial Science)  THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver)  April 2016 © Kelly (Yi Ping) Liu, 2016 ii  ABSTRACT For patients of oral squamous cell carcinoma (OSCC), tumour spread to regional lymph nodes reduces survival by half. On the account of this widely demonstrated fact, prophylactic neck treatment has been advocated for clinically node negative (cN0 necks of which the risk of nodal disease is considerably high. However, there is a lack of sensitive and specific marker to determine such risk and benefits of prophylactic treatment await confirmation.  The first part of this thesis presents a population-based retrospective review on OSCC in British Columbia. The incidence of regional failure (RF) in early-stage, cN0 patients was 28%, with median time of only 10 months after local excision. This group of patients needed to be identified and treated at earliest time possible. Tumour depth of invasion (DOI) was significantly associated with RF (P=0.01). However, it has low accuracy in predicting nodal disease with AUC of 63%. Moreover, assessment of performance for 4mm cut-off of DOI showed 55% sensitivity and 68% specificity. Furthermore, we demonstrated that using DOI as an indicator of neck treatment resulted in 25% under-treated occult metastasis and 55% over-treated necks. Thus, we concluded that, at least for BC population, conventional histological attributes of tumour cannot predict RF and we need a new marker for risk assessment. The second part presents a pilot study exploring a novel approach of risk assessment by utilizing Quantitative Tissue Pathology (QTP) on tumour nests. We were able to quantitate and evaluate 120 features describing nuclear phenotypes of tumour cell nuclei and tissue architectures of tumour nests. Compared to node-negative (N0) group, cell nuclei of the node positive (N+) group had higher fractions of heterochromatin regions. Also, the combination of two features, which describe chromatin condensation, from the outermost two layers of tumour nests had performance of AUC 94%, sensitivity of 100% and specificity of 75% in discriminating N0 and N+ group. QTP may be a potential proxy for predicting the metastatic risk of OSCC.  Further investigation on potential biomarkers in risk assessment for nodal disease of early-stage OSCC patients is warranted to provide precision management to improve mortality and reduce morbidity.    iii  PREFACE This dissertation is an original intellectual product of the author, Kelly Yi Ping Liu. Contributions from the author included study design, and data collection, tabulation, graphical presentation, and statistical analysis for objectives in Chapters 2 and 3.  In Chapter 3, histology slides were reviewed and tumour nests were defined by oral pathologist, Dr. Catherine F. Poh. Feulgen-thionin staining was performed and image processing software was developed by the team lead by Dr. Martial Guillaud of the department of Integrative Oncology at the BC Cancer Research Centre.   iv  TABLE OF CONTENTS ABSTRACT ............................................................................................................................. ii PREFACE .............................................................................................................................. iii TABLE OF CONTENTS ......................................................................................................... iv LIST OF TABLES ................................................................................................................. vii LIST OF FIGURES ............................................................................................................... viii LIST OF ABBREVIATIONS ................................................................................................... x ACKNOWLEDGEMENT ....................................................................................................... xii DEDICATION ....................................................................................................................... xiii 1 INTRODUCTION ................................................................................................................ 1 1.1 Disease Burden of Oral Squamous Cell Carcinoma ................................................... 2 1.1.1 High Mortality and Morbidity ............................................................................ 2 1.1.2 Anatomical Sites and Risk Factors .................................................................. 2 1.1.3 TNM Staging .................................................................................................... 3 1.1.4 Treatment ........................................................................................................ 4 1.2 Nodal Disease of OSCC Patients ................................................................................ 4 1.2.1 Cervical Lymph Node Metastasis and Management ....................................... 4 1.2.2 Prophylactic Neck Treatment - the Dilemma ................................................... 5 1.2.3 Investigation of Nodal Disease - Pre-surgery .................................................. 6 1.2.4 Prognostic Factors of Cervical Lymph Node Metastasis - Post-surgery ......... 8 1.3 Quantitative Tissue Pathology (QTP) ........................................................................ 14 1.3.1 Background ................................................................................................... 14 1.3.2 Quantifiable Phenotypes ............................................................................... 15 1.3.3 Applications of QTP in Cancer Research ...................................................... 17 1.3.4 Study of Tumour Nests Using QTP ............................................................... 19 1.4 Organization of Thesis ............................................................................................... 19 2 DISEASE BURDEN OF NODAL METASTASIS IN ORAL CANCER PATIENTS, A PROVINCIAL WIDE EXPERIENCE ................................................................................ 21 2.1 Guiding Questions and Objectives ............................................................................ 22 2.2 Methodologies ........................................................................................................... 22 2.2.1 Patient Cohort ................................................................................................ 22 2.2.2 Date Source ................................................................................................... 23 v  2.2.3 Statistical Analysis ......................................................................................... 25 2.3 Results ....................................................................................................................... 27 2.3.1 All Primary OSCC (N=847) ............................................................................ 29 2.3.2 cN0, Neck Treatment and Survival (N=469) .................................................. 43 2.3.3 Regional Failure (cN0, LE-only, N=322) ........................................................ 48 2.4 Discussion ................................................................................................................. 56 2.4.1 Disease Burden ............................................................................................. 56 2.4.2 Effectiveness of Neck Treatment ................................................................... 56 2.4.3 To Dissect or Not to Dissect Necks ............................................................... 58 2.4.4 Study Limitations ........................................................................................... 59 2.4.5 Next Steps ..................................................................................................... 59 3 NOVEL COMPUTATIONAL IMAGE ANALYSIS TO IDENTIFYTHE RISK OF NODAL METASTASIS IN ORAL CANCERS ............................................................................... 61 3.1 Hypothesis and Objectives ........................................................................................ 62 3.2 Methodologies ........................................................................................................... 62 3.2.1 Patient and Surgical Specimen ..................................................................... 62 3.2.2 Feulgen-Thionin Staining ............................................................................... 63 3.2.3 Image Analysis .............................................................................................. 64 3.2.4 QTP Features Extraction - Nuclear Phenotypes ........................................... 65 3.2.5 QTP Features Extraction - Tissue Architecture ............................................. 70 3.2.6 Region of Interest (ROI) for QTP Analysis .................................................... 71 3.2.7 Statistical Analysis ......................................................................................... 72 3.3 Results ....................................................................................................................... 72 3.3.1 Patient Set for QTP Analysis ......................................................................... 72 3.3.2 Tumour Nests, Tumour Nest Layers, and Cell Nuclei for N0 and N+ Cases 73 3.3.3 QTP Features Overview ................................................................................ 74 3.3.4 Tumour Nests - Association of QTP with Nodal Status (N0 vs. N+) .............. 75 3.3.5 Tumour Nest Layers - Association of QTP with Nodal Status (N0 vs. N+) .... 79 3.3.6 Discriminative Ability of QTP Features on Nodal Status ............................... 88 3.4 Discussion ................................................................................................................. 92 3.4.1 Discussion of the Results .............................................................................. 92 3.4.2 Study Limitations ........................................................................................... 95 3.4.3 Next Steps ..................................................................................................... 96 4 CONCLUSIONS AND FUTURE DIRECTIONS ............................................................... 97 BIBLIOGRAPHY ................................................................................................................ 100 Appendix A. International Classification of Disease-09 Version: 2015 ....................... 110 Appendix B. Patient Demographics and Consultation Report ..................................... 111 vi  Appendix C. Surgical Pathology Report Form .............................................................. 112 Appendix D. Radiotherapy or Chemotherapy Report ................................................... 113 Appendix E. Quantitative Tissue Pathology (QTP) Phenotypes .................................. 114 Appendix F. Plots of Mean of QTP Features for Tumour Nests, by Nodal Status ...... 116 Appendix G. Plots of Mean of QTP Features for Tumour Nest Layers#1, #2, and #3, by Nodal Status .............................................................................................. 118 Appendix H. Plots of Mean of QTP Features for Layer#1-2, by Nodal Status ............. 121 Appendix I. Plots of Mean of QTP Features for Layer#2-3, by Nodal Status .............. 122    vii  LIST OF TABLES Table 1-1. Reported Association of Tumour Depth of Invasion with Nodal Disease ............ 10	Table 1-2. Reported Associations of Degree of Differentiation with Nodal Disease ............ 11	Table 1-3. Reported Association of Lymphovascular Invasion and Nodal Disease ............. 12	Table 1-4. Reported Association of Perineural Invasion and Nodal Disease ....................... 13	Table 2-1. Demographics and Clinical Characteristics at Initial Presentation (N=847) ........ 29	Table 2-2. Initial Treatment among OSCC Patients, by Clinical N and T Stage (N=847) .... 30	Table 2-3. Disease Outcomes among OSCC Patients (N=847) .......................................... 31	Table 2-4. Disease Outcomes, by Clinical N Stage and Treatment (N=847) ....................... 34	Table 2-5. Univariate and Multivariate Cox Proportional Hazards Analysis in 5-year OS (OSCC, N=847) ................................................................................................... 36	Table 2-6. Kaplan-Meier Estimated OS Rates, by Variables (OSCC, N=847) ..................... 37	Table 2-7. Univariate and Multivariate Cox Proportional Hazards Analysis in 5-year DSS (OSCC, N=847) ................................................................................................... 40	Table 2-8. Kaplan-Meier Estimated DSS Rates, by Variables (OSCC, N=847) ................... 41	Table 2-9. Demographics and Clinico-pathological Characteristics, by Neck Treatment Group (cN0, N=447) ............................................................................................ 44	Table 2-10. Nodal Disease Outcome among by Neck Treatment Groups (cN0, N=447) ..... 45	Table 2-11. Demographics and Clinico-pathological Characteristics (cN0 LE-only, N=322) 49	Table 2-12. Univariate and Multivariate Logistic Regression Analysis for Regional Failure (cN0 LE-only, N=322) .......................................................................................... 53	Table 2-13. Sensitivity and Specificity of Tumour Attributes for RF ..................................... 55	Table 3-1. Demographics and Clinico-pathological Characteristics of QTP Set (N=15) ...... 73	Table 3-2. ROIs and Segmented Cell Nuclei ....................................................................... 74	Table 3-3. QTP Features Measured ..................................................................................... 75	Table 3-4. QTP Features of Significant Difference Between N0 and N+ Tumour Nests ...... 76	Table 3-5. Number of nth Layer in 45 Tumour Nests ........................................................... 79	Table 3-6. QTP Features Different Between N0 and N+, Layer#1, or Layer#2, or Layer#3 80	Table 3-7. QTP Features Different between N0 and N+ Layers#1-2 or Layers#2-3 ............ 86	  viii  LIST OF FIGURES Figure 1-1. Examples of Tissue Architecture Derived from QTP Analysis ........................... 16	Figure 1-2. QTP Analysis and Cell Sociology ...................................................................... 17	Figure 2-1. Head and Neck Cancer OSCC Patients Identified from the BCCA Registry (2001-2007) ....................................................................................................... 28	Figure 2-2. Disease Outcome of Second Cancer (SC) or Distant Metastasis (DM) among OSCC Patients (N=847) .................................................................................... 32	Figure 2-3. Kaplan-Meier Survival Curve and Rates for OS;(OSCC,N=847) ....................... 35	Figure 2-4. Kaplan-Meier Survival Curves of 5-year OS (N=847) ........................................ 38	Figure 2-5. Kaplan-Meier Survival Curve and Rates on 5-year DSS (N=847) ..................... 39	Figure 2-6. Kaplan-Meier Survival Curves of 5-year DSS (OSCC, N=847) ......................... 42	Figure 2-7. Kaplan-Meier Survival Curves of 5-Year OS; by Neck Treatment at Surgery (cN0, N=447) ..................................................................................................... 46	Figure 2-8. Kaplan-Meier Survival Curves of 5-year DSS; by Neck Treatment at Surgery (cN0, N=447) ..................................................................................................... 47	Figure 2-9. Cumulative Incidence of Regional Failure (cN0 LE-only, N=322) ...................... 48	Figure 2-10. Cumulative Incidence of Regional Failure of cN0- -pN+ (N=89) ...................... 50	Figure 2-11. Kaplan-Meier Survival Curve and Rates for OS, by Regional Failure (cN0 LE-only, N=322) ...................................................................................................... 51	Figure 2-12. Kaplan-Meier Survival Curve and Rates for DSS, by Regional Failure (cN0 LE-only, N=322) ...................................................................................................... 52	Figure 2-13. Receiver Operating Characteristic (ROC) Curve for Tumour DOI for the Diagnosis of Nodal Status ................................................................................. 54	Figure 2-14. Over-treating or Under-treating cN0 Patients using Depth of Invasion ............ 55	Figure 3-1. QTP Analysis Workflow Schematic .................................................................... 63	Figure 3-2. Example of a grey-level image of a nucleus with its corresponding mask Nuclei (left) with a Mask Object (right). ........................................................................ 64	Figure 3-3. Examples of Excluded Cell Types and ’Bad’ Nuclei after Segmentation. .......... 65	Figure 3-4. Fractal Dimension Measurement ....................................................................... 66	Figure 3-5.  Chromatin Discrete Regions in a Cell Nuclei .................................................... 68	Figure 3-6. Examples of Nuclear Feature ............................................................................ 69	Figure 3-7. Examples of Tissue Architecture Features of Tumour Nests ............................. 70	Figure 3-8. Layered Tumour Nest ........................................................................................ 72	Figure 3-9. Box-and-whisker Plots of Distribution of Tumour Nest Area, Layers per Nest, and Cell Nuclei per Nest for N0 and N+ Cases ................................................. 74	ix  Figure 3-10. Box-and-whisker Plots of Distributions of Nuclear Phenotypes between N0 (gray) and N+ (red) Tumour Nests ................................................................... 77	Figure 3-11. Box-and-whisker Plots of Distributions of Tissue Architecture Between N0 (gray) and N+ (red) Tumour Nests ................................................................... 78	Figure 3-12. Box-and-whisker Plots of Distributions of QTP Features among Layer#1’s, Layer#2’s and Layer#3’s, by N0 (gray) and N+ (red) ....................................... 85	Figure 3-13. Box-and-whisker Plots of Distributions of QTP Features among Layer#1-2, by N0 (gray) and N+ (red) ..................................................................................... 87	Figure 3-14. Box-and-whisker Plots of Distributions of QTP features among Layers#2-3 by N0 (gray) and N+ (red) ..................................................................................... 88	Figure 3-15. Receiver Operating characteristic (ROC) Curve and Scatter Plot for QTP Features for Discrimination of N0 vs. N+, among Tumour Nests ..................... 89	Figure 3-16. Receiver Operating Characteristic (ROC) Curve Analysis for QTP Features for Discrimination of N0 vs. N+, among Layers#1s, or Layers#2, or Layers#3 ..... 90	Figure 3-17. Receiver Operating Characteristic (ROC) Curve for QTP Features for Discrimination of N0 vs. N+, among Layer#1-2 or Layer#2-3 .......................... 91	Figure 3-18. Scatter Plot and Box-whisker Plot of NC Ratio in N0 and N+ Tumour Nests . 94	x  LIST OF ABBREVIATIONS +/- With or without Stage I/II Early stage Stage III/IV Late or advanced stage ANOVA Analysis of variance AUC Area under the curve BC British Columbia BCCA British Columbia Cancer Agency Chemo Chemotherapy CI Confidence interval CIS Carcinoma in situ cN0 Clinically node negative cN+ Clinically node positive D1 Mild dysplasia D2 Moderate dysplasia D3 Severe dysplasia DOI Depth of invasion DM Distant metastasis DSS Disease-specific survival END Elective neck dissection FOM Floor of mouth H&E Hematoxylin and eosin HPV Human papillomavirus HR Hazard ration LE Local excision LR Local recurrence ND Neck dissection NPV Negative predictive value OD Optical density OR Odds ratio OS Overall survival OSCC Oral squamous cell carcinoma pN+ Pathology-proven node positive pN0 Pathology-proven node negative PPV Positive predictive value QTP Quantitative tissue pathology RF Regional failure xi  ROC Receiver operating characteristic ROI Region of interest RR Regional recurrence; recurrence of nodal disease RT Radiotherapy SC Second cancer SLNB Sentinel lymph node biopsy T1 Tumour stage 1; tumour size less than or equal to 2 cm T2 Tumour stage 2; tumour size greater than 2 cm and less than or equal to 4 cm T3 Tumour stage 3; tumour size greater than 4 cm T4 Tumour stage 4; tumour invades adjacent structures TNM Tumour, lymph node, distant metastasis WHO World Health Organization VC Verrucous carcinoma  xii  ACKNOWLEDGEMENT I would like to thank Catherine for her years of guidance, warm friendship and kindness, and most of all for her unwavering support that has helped me through the most challenging times. I thank Martial for introducing and sharing his deep knowledge of computational biology for his generosity in allowing me to use the excellent resources of his lab. I am grateful to Jonn for giving me uttermost constructive advice and for playing a pivotal role in helping me shape the clinical research goals. Sarah has been an incredible mentor and I am indebted to her for the constant support, encouragement and reassurance through my years as a grad student. I thank Monica for her friendship and patience in guiding through statistical challenges in this research. Jagoda, Anita, and Alan have provided me with critical help and facilitated my research, and I am grateful to them for this.  The grad school experience would not have been as enjoyable without the company of a bunch of wholesome wonderful friends who are too many to list here. I am especially thankful to life-long friends, Lisa and Allan for living with my absence and ignorance yet provided me with undivided attention at the hardest times. I am thankful to Eddy for his sharp and honest encouragement, and generously shared his experience as a successful scientist. I am incredibly thankful and honoured to be able to call everyone from the Poh lab my friends. The long working hours would be much more intolerable without them. I want to thank Eric for being the best brother-in-law one could ask for. Lastly and most importantly, my family is the pillar of strength all the way through. I cannot thank them enough.     xiii  DEDICATION         For Wayne, Shirley, and Jennifer    1         CHAPTER 1 1 INTRODUCTION   2  1.1 Disease Burden of Oral Squamous Cell Carcinoma 1.1.1 High Mortality and Morbidity Oral squamous cell carcinoma (OSCC) is solid malignant neoplasm on the surface squamous epithelium in the mouth. It accounts for 90% of all oral cancer types. Worldwide, annual incidence and deaths is estimated to be 325,000 and 157,000, respectively.1-3 It is a disease with well accepted concept of progressively transformation from premalignant squamous epithelial neoplasias (mild dysplasia, moderate dysplasia, severe dysplasia, or carcinoma in situ)4-6  to invasive cancerous lesions. According to several epidemiological studies, the proportion of transformation to OSCC is about 30% of all premalignant cases and may take up to 10 years.4,7 Thus, one would expect that, with early detection and easy accessibility for treatment, OSCC is “highly preventable” and “highly curable”. Quite on the contrary, global cancer statistics continue to associate OSCC with poor prognosis, with five-year survival rate going steady at about 65%, and rank it the eighth most common cancer (about 5.5%) of all malignancies.3,8 This is largely due to unawareness of its existence and ignorance of early clinical indications that leads to late diagnosis, and extending the time for accumulation of genetic and epigenetic changes resulting invasive cancer.9 Negative impacts on quality of life, severe disfigurement from treatment, and incidence of recurrences or distant metastasis have all been associated with poor survival in OSCC patients.  1.1.2 Anatomical Sites and Risk Factors Diagnosis of OSCC is always confirmed by biopsy on the abnormal lesion in the oral cavity. There are six intraoral anatomical sites where OSCC can originate: the tongue, floor of mouth, soft palate, hard palate, gingiva, or buccal mucosa (interior cheeks). Historically, tongue (especially the lateral and ventral aspects) is associated with the highest risk and the most common location of OSCC. Floor of mouth has similar risk as its close proximity with tongue and can be invaded by tumours from anterior, middle, or posterior parts of the tongue.5,10-13 Soft palate is the fleshy soft palate in the back of the mouth and is associated with intermediate risk of developing cancer.5,11,12,14 Primary malignant origin from hard palate, gingival, or buccal mucosa are much rare compared to other anatomical sites; thus considered as low-risk sites in North American literatures.12,15 3  Alterations that eventually result in OSCC are products of either genetic predispositions or somatic mutations commonly caused by repetitive exposure to carcinogenic agents, such as tobacco products (cigarettes, betel quid, betel nut, smokeless or chewing tobacco), alcohol, or human papillomavirus (HPV) infection.10,16-20 These alterations result in a display of clinical changes, including discolouration and ulceration, which are warning signs for further investigation.5,6  1.1.3 TNM Staging Tumour staging allows clinicians to estimate the extent of the disease and prognosis of patients, and plan appropriate treatment. Such estimation is always performed at the time when a patient is first diagnosed with OSCC and before treatment; however, it can also happen any time of the disease. Staging of OSCC uses the TNM (tumour size, lymph nodes affected, metastases) classification system of the American Joint Committee on Cancer (AJCC).5,21 The T classification indicates the greatest tumour dimension and is sub-classified into T0 (no cancerous tumour), T1 (≤ 2 cm), T2 (> 2cm and ≤ 4 cm), T3 (> 4 cm), and T4 (invasion to adjacent structures such as bone, deep muscle, maxillary sinus, or skin / skull base or carotid artery). The N classification indicates the presence of tumour in regional lymph nodes; and is subdivided into N0 (none involved), N1 (≤ 3 cm in a single ipsilateral lymph node), N2a/b/c (> 3 to ≤ 6 cm in a single ipsilateral lymph node / ≤ 6 cm in multiple ipsilateral lymph nodes / ≤ 6cm in bilateral lymph nodes/ ant contralateral lymph node), N3 (>6cm in any lymph node). The M classification indicates spread of cancerous squamous cells from oral cavity to distant organs, most often lungs or brain, and can be subdivided into M0 or M1.5 Depending on status of each T, N, M, OSCC patients are staged into Stage I, II, III, and IV. Stage I (T1N0M0) or II (T2N0M0) is commonly known as “early” stage, while stage III (T3N0M0 or T1-3, N1, M0) or IV (T4, N0-3, M0 or T1-4, N1-3, M1) is referred to “late” or “advanced” stage. Staging is a routine procedure that must be performed for every patient in order to estimate the presence of locally advanced diseases, the involvement of cervical lymph nodes, and the prognosis of patients, and plan for appropriate treatment. When the pathology information confirmed the tumour size and regional lymph node or organ involved, the TNM staging will be updated from clinical staging, using clinical and/or 4  imaging technique, to pathological staging derived via information from surgical specimen, a more definitive staging and outcome assessment. 1.1.4 Treatment When planning the treatment for OSCC patients, clinicians have to consider in respect to both the primary site and potential cervical lymph node involvement. Types of treatment include 1) surgery, local excision (LE) of primary intraoral tumours with additional margin of 'normal looking' mucosa for safe-keeping; 2) radiation (RT), damaging the tumour by inducing irreversible DNA damage in tumour cells; and 3) chemotherapy (Chemo), often in combination with RT as an adjuvant modality, repressing tumour growth with chemicals or sensitize the effect of RT.22 As primary tumour in the oral cavity is relatively accessible, surgery is the mainstay for early-stage OSCC whereas multiple-modality therapy is the primary choice for advanced disease. Radiation, of curative or adjuvant intent, would also be in the favour when a tumour is hard to reach or unresectable, or when resection would lead to severe cosmetic disfigurement or functional deficit. Some early-stage patients may also be managed with RT post initial surgery if there is any presence of residual disease or “positive margin”.23-26 Unfortunately, despite refinement in treatment strategy over the years, there has been limited improvement in survival over the past five decades.27-30 Amongst the known prognostic factors, lymphatic metastasis from the primary to regional cervical (neck) lymph nodes (hereinafter referred to as "nodal disease") has been considered to be the most notorious one.27,31-38 1.2 Nodal Disease of OSCC Patients 1.2.1 Cervical Lymph Node Metastasis and Management Due to close proximity to the primary tumour in the oral cavity, lack of tight junctions between lymphatic endothelial cells, and less resistance from the capillary vessels, the lymphatic network in the neck region is a major and the most common site for metastasis.39,40 Approximately 45% of OSCC patients will be diagnosed with nodal disease;39 and occult neck metastasis is seen in 20% to 45% of clinically negative nodal disease (cN0) patients.41-44 For patients with cervical nodal disease, survival is significantly reduced. The impact of any extent of regional nodal disease is so prominent that each patient’s neck must be assessed carefully. For OSCC patients who present evident nodal 5  disease at initial presentation, surgery involves therapeutic neck dissection (ND), removing all 5 levels of the neck involved with or without disease grossly. Therapeutic ND is classified into four main types, namely, radical, modified, selective, or extended radical, with each targets different groups of nodes.45-50 Radical ND removes Levels 1 to 5, sternocleidomastoid muscle, accessory nerve and blood vessels. Modified radical removes levels 1 to 5, and additional structures if involved by tumour.  On the other hand, modified ND removes only the lymph nodes, and is preferred when metastasis is seen in levels I, II, and III with confined node(s) and no extracapsular spread.49 ND comes with intrinsic morbidity and functional deficits49 which are rarely captured in survival analysis. In several cross-sectional and survey studies, the extent of ND was positively correlated with experience in neck stiffness, constriction, pain, shoulder drop, and arm abduction.51-55 1.2.2 Prophylactic Neck Treatment - the Dilemma When OSCC patients presents with no clinical, radiologic, or pathological evidence of node metastasis, a decision must be made for either prophylactic ND or "watchful waiting" (i.e. only perform ND until metastasized). However, the major setback with “watchful waiting” approach is delaying treatment for cN0 patients who have a risk of harboring occult metastasis. Moreover, by the time tumour reaches the lymph nodes, the disease is usually too aggressive to treat and patients face high risk in local-regional recurrences or distant metastasis.31,56-58 In the absence of cN0 disease, the tumour is considered localized. However, it is known that a portion of cN0 patients will develop nodal disease during follow-up. For such cases, it is always the goal of the clinicians to remove nodes before spread happens. Thus, they may choose to practice elective neck dissection (END) for not only diagnostic purposes, but also for preventative measure of tumour spread by removing specific nodal groups at the earliest possible stage.42,59-64 However, the decision of performing END, especially for early-stage (T1/T2/cN0) OSCC patients, has long been controversial for two main reasons. First and foremost, there is no validated screening method(s) that accurately detects subclinical metastatic tumour or assesses patients' risk in development of nodal disease. With the risk about 1 in 4 and the introduction of severe morbidities from ND, it has always been difficult for clinicians to justify END as part of the treatment. 6  The second weight added to the dilemma of END is that, across various studies, there is no consistent and definitive evidence of improvement in survival in patients who received END compared to those who were managed under surveillance. Several retrospective studies reported the benefits of END in reducing incidence of regional recurrence in cN0 patients; thus, increasing survival rate. However, there have only been five randomized controlled trials investigating the benefits of END. A Cochrane meta-analysis of prospective randomized controlled trials included four studies with a total of 283 patients, and reported that END remarkably improved regional control rate and reduced risk of disease-specific death by approximately 40%.60,65-68 However, the survival improvement seen in these studies were not reproducible by other prospective studies, which reported no difference in survival benefits between “wait-and-treat” or END. These studies also raised the issues of the side effects from neck dissections, such as increased costs and complications such as damages of functional nerves and muscles.63,69,70 The significant impact on survival has stirred lots of interests from investigators to find potential predictive markers. 1.2.3 Investigation of Nodal Disease - Pre-surgery The impact of node metastasis on survival is so prominent that all OSCC patients must be clinically staged before assigning type of treatment. For better planning, patients may receive various imaging modalities or sentinel node biopsy to evaluate the nodal status.  Imaging Modalities Routine radiological imaging techniques, including computed tomography (CT), magnetic resonance imaging (MRI) or ultrasound (US) have been investigated in their sensitivity of determining the presence of abnormal cervical lymph nodes in cN0 necks.71-76 However, they have shown to be limited for assessing small tumour deposits of cN0 necks. Wooglar et al.77 reported that only 21% of pathology-proven positive (pN+) nodes out of 152 cN0 necks were detected by pre-operative MRI. Among the diseased lymph nodes, 56% were found to be 10 mm or less axial diameter, suggesting that microscopic tumour deposits evade the radar of routine imagine techniques. Merritt et al.78 reported similar sensitivity and specificity between palpation and CT. Interestingly, some small sample-size 7  studies indicate the sensitivity and specificity of manual palpation of nodal disease surpass CT or US.79-82 Limitations of these techniques come from the lack of threshold for separating benign or reactive from subclinical malignant nodes.83 For example, in CT or MRI scans, clinicians look for enlarged nodes with at least 1 cm in size and nodal morphology changes such as ill-defined border, shape, or intensity change. For nodes with small tumour deposits (<1cm) these imaging methods would be unreliable or less sensitive. On the other hand, the specificity of these imaging weakens when enlargement of nodes are caused by inflammation or infection. Sentinel Lymph Node Biopsy (SLNB) More recent adapted neck staging method for OSCC is SLNB. This method utilizes the tendency of tumour cell moves from the primary to the neck nodes in a predictable pattern.40 Tumour metastasizes from the oral cavity down to the neck in a pattern that mirrors the drainage pathway of the lymphatic system.84,85 Cervical lymph nodes are grouped into five levels (I, II, III, IV, and V) on each side of the neck based on the order of flow and proximity among them.46 Briefly, the boundary of the levels starts with Level I (IA/IB) from the inferior border of the mandible and is subdivided into submental and submandibular lymph nodes. Subsequent Levels II (IIA/IIB), III, and IV consist nodal groups along the jugular chain (upper, middle, and lower, respectively). Finally, Level V (VA/VB) denotes spinal accessory lymph nodes running deep in the posterior region of the neck between the base of skull and clavicle. There are three main pathways tumour cells adapt in OSCC: 1) the main lymphatic pathway denotes drainage to Level IA/IB, subsequently directed to Level IIA, III and IV; 2) the posterior accessory pathway denotes drainage from the root of the tongue to Level IIB, then to VA and VB; and 3) the anterior lymphatic pathway involves initiation from anterior oral cavity (tip of the tongue, floor of mouth, lower lip, anterior buccal mucosa and gingival areas) to Level IA then to Level III or Level IV.86,87 Based on the sites where primary tumour originates SLNB targets the lymph nodes that are most likely be as their first 'pit stop' of metastasis. Ross et al.88 prospectively investigated the agreement between radioactive nodes and H&E examination, subsequent ND of upstaged patients, and clinical outcome. Of the 43 cN0 patients included in the 8  study, 16 patients had subclinical nodal disease of whom 15 (94%) were upstaged by SLNB, with 13 confirmed by subsequent therapeutic ND. In other studies, SLNB method has been shown with high positive and negative predictive value for nodal disease with looking directly at the first lymph node(s) of where the tumour cells spread to.89-93 The clinical adaptability of SLNB for head and neck cancer patients is being evaluated in a few clinical trials. Preliminary results from a multicenter trial by Alkureish et al.94 demonstrated a 91% sensitivity and 95% negative predictive values in 93% successfully harvested sentinel nodes, suggesting negating END and avoidable complications and costs. Although the result is encouraging, the cost and time required for this technique and the reproducibility of these high sensitivity values is still limited as it requires experienced surgeons and pathologists.89,95 Most importantly, the method is only reported in detecting micrometastatic tumour cells originating from the floor of mouth and the tongue. 1.2.4 Prognostic Factors of Cervical Lymph Node Metastasis - Post-surgery As clinical examinations and imaging studies are not very accurate, interests are shifted to post-operative prediction of the risk of node metastasis in cN0 necks. One has to remember that these factors are measured post-formalin fixation and on the excised tumourous tissue.   Depth of Invasion Association of tumour depth of invasion (DOI) with increased risk of metastasis has been reported as the most significant prognostic factor among other histological characteristics. Indeed, the more aggressive the tumour is, the deeper it invades and more likely to penetrates into the lymphatic system.  Many of investigators reported the association of DOI in early-stage N0 patients and node metastasis, in hope to identify the optimal cut-off depth for prediction of nodal involvement. However, findings from these different groups were inconsistent with DOI ranges from 3.0 mm to 10.0 mm. Table 1-1 summarizes some of the literatures on DOI cut-offs concerning nodal metastasis.  Although these studies did demonstrate a significant relationship between larger DOI and node metastasis, they are criticized by lack of consensus in reported cut-off values of DOI due to interchangeably used definition of ‘tumour thickness’ or ‘tumour DOI’, and variation in how measurements were made.96 According to the definition from the World 9  Health Organization (WHO), ‘tumour thickness’ was used and defined as “the deepest tumour invasion to the presumed original surface level, ignoring exophytic growth or assessing the original surface level in ulcerated tumours.”5 However, some authors adopted the technique developed by Breslow et al.97 and measured vertically starting from the tumour surface or the base of the ulcer base. Others used measurement technique cited by Moore et al.98 on reconstructing an imaginary “normal mucosal line” as the starting point in cases of exophytic and or verrucous carcinoma. However, it is not always easy to distinguish the exact technique that was used and most authors did not clearly explain how DOI was measured. In addition, the credibility of optimal DOI cut-off reported by these studies suffers from small sample size, poor sampling, and variation in tumour sections or characteristics of study cohort. In addition, most investigations focus on T1 / T2 or tongue cancers which might underestimate the threshold by neglecting cases with larger tumour size. Despite the heterogeneity in study design and lack of transparency in how tumour DOI is defined and measured, most commonly suggested and adopted cut-off of the DOI is 4 mm for indication of END.  10  Table 1-1. Reported Association of Tumour Depth of Invasion with Nodal Disease Author Study Type Study Period No. Pt. TNM Site Initial Treatment Adaptation of Measurementb N+ cut-off (mm) Almangush, A. et al99 P ‘79-'09 233 T1-2,N0 T LE Depth(A) ≥4 Asakage, T. et al100 R ‘80-'95 44 T1-2, cN0 T LE Depth(A) >4 Byers et al101 P ‘90-94 91 T1-4, N0-3 T LE+ ND Depth(B) ≥4 Chen, YW, et al102 P ’00-‘03 94 T1-4, N0-2 T LE +/- ND Depth(A) >3 Fakih et al60 RTC ‘85-88 51 T1-2, cN0 T LE +/- END ns >4 Fukano, H. et al103 R ‘80-91 34 T1-4, N0-3 T LE+ ND ns >5 Hayashi et al104 R ‘97-'00 20 T1-2, cN0 T LE +/- END Ultrasound >5 Hosal, AS. et al105 R ’70-‘92 60 T1-4,N0-3 T LE+ ND Depth(A) >9 ICOR in Head and Neck Cancer106 R ’90-‘11 3,149 T1-4,N1-3 OCa LE+ ND Depth(A) and Depth(B) T1, >5;  T2-4, >10 Jones, KR. et al107 R ‘81-‘90 49 T1-2,cN0 T; FOM; SP: HP: BM; lip LE +/- END Thickness(A) >5 Jung, J. et al108 R ’02-‘05 50 T1-2,N0-N2 T LE +/- ND Depth(A) >9 Kim, HC. et al109 R ‘73-'90 90 T1-2, cN0 OCa LE or RT Thickness(A) >6 Kligerman, J. et al67 RTC ‘87-'92 67 T1-2, cN0 T; FOM LE +/- END Thickness(A) >4 Kurokawa, H. et al110 R ‘85-'96 50 T1-2, cN0 T LE ns >4 Lim, SC. et al111 R ‘92-'00 56 T1-2, cN0 T LE Thickness(A) >4 Mohit-Tabatabai et al112 R ‘63-'82 84 T1-2, cN0 FOM LE +/- RT Thickness(B) >1.5 Nakagawa, T. et al113 R ‘71-'98 616 T1-2, cN0 T Brachytherapy +/- RT ns >4 Nathanson et al114 R ns 58 T1cN0 T LE ns >10 O-charoenrat, P. et al115 P ‘81-‘98 50 T1-2,cN0 T LE Thickness(A) >5 Sparano, A. et al116 R ’95-01 45 T1-2,N0 T LE + END Depth(A) >4 Spiro et al117 R ‘77-'81 92 T1-4, cN0 T; FOM LE +/- END Thickness(A) >2 Steinhart and Kleinsasser. et al118 R ‘80-'92 48 T1-4, cN0-3 FOM LE ns >5 Urist, MM. et al119 R ‘59-'85 89 T1-4, cN0-3 BM LE Thickness(B) >6 Yuen, APW. et al120 P ‘87-'98 72 T1-2,N0 T; FOM LE +/- END Thickness(A) >3 aOC, oral cavity, anatomical sub-site not specified. bThickness(A): From the surface of the tumour to the deepest point of invasion, excluding keratin, parakeltin, and inflammatory exudate; Thickness(B): From the surface of the tumour to the deepest point of invasion, regardless of exophytic growth or ulceration; Depth(A): In cases of exophytic or ulcerated tumour, from reconstructed normal mucosal line to the deepest point of invasion; Depth(B): From the base of the tumour to the deepest point of invasion. ICOR, International Consortium for Outcome Research; P, prospective; R, retrospective; RCT, randomized controlled trial; No. Pt, number of patients; TNM, tumour, lymph nodes, metastasis; cN0, clinically node-negative; T, tongue; OC, oral cavity; FOM, floor of mouth; SP, soft palate; HP, hard palate; BM, buccal mucosa; LE, local excision; NE, neck dissection; END; elective neck dissection; RT, radiotherapy; ns, not specified; N+, node positive 11  Degree of Differentiation The degree of squamous differentiation refers to the resemblance of tumour cells to normal epithelium according to the degree of keratinization, polarity and cellular atypia. For OSCC, the degree of differentiation is graded into well, moderately, or poorly differentiated, at the worst graded area of the tumour.5 Although the degree of differentiation indicates the severity and advancement of the disease, it suffers from inter-observer variability and sampling bias and has been reported as an unreliable predictor for nodal metastasis.121,122 Table 1-2 lists literatures that reported weak relationship between tumour differentiation and nodal metastasis. Table 1-2. Reported Associations of Degree of Differentiation with Nodal Disease Author Study Type (period) No.Pt. TNM Site Initial Treatment Gradea N+, No. Pt Association with N+ (p-value) Jones, KR. et al107 R (’81-’90) 49 T1-2, cN0 T; FOM; SP: HP: BM; lip LE I vs. II vs.III 11 0.23 Asakage, T. et al100 R (’80-’95) 44 T1-2, cN0 T LE I vs. II vs.III 21 ns Okamoto, M. et al.123 R (’65-’84) 98 T1-4, N0-N3 T ns I vs. II vs.III 32 0.14 Sparano, A, et al.116 R (’95-’01) 45 T1-2, cN0 T LE+ END I-II vs.III 13 0.10 aGrade I, well-differentiated; Grade II, moderately differentiated; Grade III, poorly-differentiated. R, retrospective; No. Pt, number of patients; TNM, tumour, lymph node, metastasis; T, tongue; FOM, floor of mouth; OC, oral cavity; SP, soft palate; HP, hard palate; BM, buccal mucosa; LE, local excision; END; elective neck dissection; ns, not specified; N+, node-positive Pattern of Invasion The development of OSCC is known to be a multi-step process during which cellular changes are graded by the amount of proliferation of immature squamous cells and loss of normal cellular organization. The increasing gradation of premalignant lesions include mild, moderate, severe, and carcinoma in situ. Finally, the breakthrough of basement membrane into the stroma marks the presence of OSCC. There are several patterns of invasion (POI) observed in OSCC and they have been associated with poor prognosis and nodal metastasis. Tumours can invade in either cohesive sheets or aggregates of cells (i.e. tumour nests), or non-cohesive groups or single cells dispersed throughout the stroma.4,124 In the literature, worst prognosis was associated with non-cohesive groups with the presence of small groups of cells or single cancerous cells. Odell et al. studied the survival 12  in 47 tongue OSCCs in regards to clinicopathological characteristics including POI. Higher rates of local recurrences (P=0.02) and metastasis (P<0.001) were seen in tumours with non-cohesive pattern. Angadi et al.125 reported on 75 OSCC cases of which tumours with small groups or cords of infiltrative cells at invasive front were associated with nodal metastasis (P=0.03). In another study by Almangush et al.99, worst POI was small tumour islands or ‘satellite’ tumours and associated with poor prognosis (HR, 4.47; 95% CI, 1.6-12.5; P=0.004). Finally, Crissman et al.126 suggested that OSCC invading as large cohesive aggregates had a significantly better prognosis than those invading as thin, irregular cords or as individual cells. Nakanishi Y et al.127 later reported that a wealth of information from tumour nests configuration can be extracted and used to elucidate the aggressiveness of the tumour. From the 159 esophageal SCCs, patients with asteroid-shaped tumour nests with a speculated margin or scattered small tumour nests had significantly poorer survival compared to those with sheet-like tumour nests with round margin. This type of invasion pattern was also associated with tumour depth of invasion and lymph node metastasis. Lymphovascular Invasion  Lymphovascular invasion is the presence or absence tumour cells deposits or attached to the wall of the vessel, and has been reported as weak predictor of nodal metastasis in most studies (Table 1-3). Table 1-3. Reported Association of Lymphovascular Invasion and Nodal Disease Author Study Type (period) No. Pt. TNM Site Initial Treatment N+,  No. Pt. Association with N+  (p-value) Chen, YW. et al102 P (’00-’03) 94 T1-T4, N0-N2 T  LE +/- ND N+ (43) 0.01 Rahima, B. et al.128 R (ns) 101 T1-4, N0-3 T; FOM; BM; ORO LE +/- ND; RT; Chemo; or Combined N+ (14) 0.74 Hosal, AS. et al.105 R (’70-’92) 60 T1-4, N0-3 T LE +/- ND N+ (35) ns Brown, B. et al.13 R (’74-’84) 87 T1-4, N0-3 T; FOM LE +/- ND N+ (14) 0.006 R, retrospective; No. Pt, number of patients; TNM, tumour, lymph node, metastasis; T, tongue; FOM, floor of mouth; OC, oral cavity; SP, soft palate; HP, hard palate; BM, buccal mucosa; ORO, oropharynx; LE, local excision; ND, neck dissection; RT, radiotherapy; Chemo, chemotherapy; ns, not specified; N+, node-positive   13  Perineural Invasion Perineural invasion (PNI) is defined as presence or absence of invasion of the perineural space or epineurim. Presence of PNI has been associated with aggressiveness, poor prognosis and loco-regional recurrences in OSCC patients.105,129 However, these studies included patients of advanced stage (III/IV) with N+ at time of surgery (Table 1-4). Thus, results can only suggest the aggressiveness of the tumour and cannot be applied to assess the usefulness of PNI in predicting risk of nodal metastasis in cN0 patients. Table 1-4. Reported Association of Perineural Invasion and Nodal Disease Author Study Type (period) No. Pt. TNM Site Initial Treatment N+, No. Pt Association with N+ (p-value) Yuen, APW.et al120 P (’87-‘98) 72 T1-2,  N0 T;FOM LE +/- END N+ (31) 0.23 Chen, YW. et al102 P (’00-’03) 94 T1-4,  N0-N2 T LE +/- END N+ (43) < 0.01 Flynn, CJ. et al130 R (’00-’06) 223 T1-4,  N1-3 HN LE +/- END; RT; Chemo; or Combined OS 0.01 Rahima, B. et al128 R (ns) 101 T1-4,  N0-N3 T; FOM; BM; PHAa LE +/- END; RT; Chemo; or Combined N+ (14) 0.03 Fagan, JJ. et al131 R (’81-’91) 142 T1-4, N0-3 OC*; Oroa LE + ND +/- RT or Chemo N+ (119) 0.03 aPHA, pharynx, includes oropharynx, larynx, nasopharynx, hypopharynx P, Prospective; R, Retrospective; NO. Pt. number of patient; TNM, tumour, lymph node, metastasis; T, Tongue; FOM, Floor of Mouth; HN, head and neck; BM, buccal mucosa; Oro, oropharynx; LE, local excision; w/wo, with or without; END, elective neck dissection; RT, radiotherapy; N+, node-positive; OS, overall survival regression HN, Head and Neck, Oral cavity; PHA, pharynx Anatomical Site of Primary Tumour The vast majority of OSCC is found on the tongue or the floor of mouth. However, in studies investigating prognostic values clinic-histological characteristics in either survival or nodal metastasis, the location of the tumour has not been shown as a risk factor.107,132 In summary, tumour DOI is the most frequently reported prognostic factor with significant relationship concerning nodal metastasis. Thus, many clinicians adapt the cut-off of 4 to 5 mm for END. However, to date, none of the suggested cut-offs are validated and brought to standard practice. This is largely due to their subjectivity and being impossible to confirm pre-or intra-operatively.   14  1.3 Quantitative Tissue Pathology (QTP) 1.3.1 Background Quantitative tissue pathology (QTP) involves computational image analysis of sections of biospecimen as a mean to obtain objective information concerning the diagnosis and prognosis of cancers. When the section of the specimen is stained using feulgen-thionin stain for nuclear DNA, nuclear phenotypes can be calculated. The process of image analysis begins with capturing and storing images of a stained section of the specimen. Once digitized, the in-house computer program can automatically segment regions of interest (ROIs) on the scanned image, process them and extract quantitative measurements. During carcinogenesis, tumour cells go through progressive changes in appearance due to accumulative genetic alterations, making them phenotypically different from normal cells. As a result, comparison of QTP measurements allows the study of phenotypic changes for difference disease phases and elucidates phenotypes associated with metastatic potential. The current ‘gold standard’ for diagnosing cancerous disease remains histopathology, the observation of any abnormal characteristics under microscope by the pathologists. In general, assessment of the degree of tumour invasiveness concerns with many pathological descriptors, such as, i) degree of differentiation, ii) DOI, iii) presence of tumour in perineural space or lymphovascular space, or iv) patterns of invasion (i.e., broad cohesive sheets of cells, strands of cells, or non-cohesive groups or single cells). While this information gives accurate diagnosis of current stage of disease in guiding management, describing these pathological features are by definition subjective with inherent issues of inter- and intra-observer variability. Furthermore, in cases where only a few highly invasive squamous cells are embedded in a sea of normal epithelial or inflammatory cells, spotting them require much experience and knowledge in deciding and correlating their clinical importance. With computer and technology aids, QTP acts as an assistive technology that enhances the reliability, reproducibility and capability in describing the pathological changes. There is a wealth of cancer research dedicated to applying image analysis technique to quantify microscopic features in order to understand the cancer pathology, diagnosis, and differential characteristics in ‘at-risk’ pre-malignant cells that are undergoing carcinogenic transformation. 15  1.3.2 Quantifiable Phenotypes As the name implies, QTP quantifies phenotypic attributes of a population of cell nuclei from an image of a tissue specimen, pixel by pixel. Tissues are stained with dye agents, typically Fuelgen-Thionin staining, in which the dye, Thionin acetate, absorbs light at wavelength around 600 nm. This absorbance (also called optical density, OD) is proportional to the amount of DNA. Knowing the OD of every pixel allows subsequent mathematical calculation on the entire image. The end results are numerical values that describe nuclear phenotypes and tissue architecture. Nuclear Phenotypes  Nuclear phenotypes describe nuclear structures, which, in cancer cells, are altered compared to normal cells. Cancer cell nuclei exhibit nuclear pleomorphism133 (variation of nuclear size and shape) and irregular chromatin pattern (irregular coarse clumps or vesicular, chromosomal relocation, or condensation state).134-137 These changes are products of genetic and epigenetic alterations causing modification in nuclear matrix, nuclear membrane abnormalities, and nuclear chromatin organization.138-140 In this thesis, we apply QTP analysis to measure features that describe nuclear morphology, optical density, and chromatin textures. Detailed descriptions of each category are described in Chapter 3, Section 3.2. Tissue Architecture  Tissue architecture (Figure 1-1) describes the organization of a population of cells in a defined region (i.e., a tumour nest, Figure 1-1A). Through architecture, we are able to describe A) local neighborhood, and B) global organization. Together, they imply local or global network of cell-cell interactions.  A. Local neighborhood refers to spatial relationship between a cell and its neighbors. Such information are obtained from drawing Voronoi diagrams, which partition a tumour nests into regions based ‘seeds’ (i.e. centre of the cell nuclei) and shortest distance to their neighboring seeds. The end result is an image of polygons which now have quantifiable characteristics such as number of sides, area, and shape (Figure 1-1B). Intimately interrelated to each Voronoi diagram is Delaunay triangulation (Figure 1-1C), which 16  divides a tumour nest into triangles, with the seeds (used in Voronoi diagram) as vertices and that no seed is inside the circumcircle of any triangle. B. Global organization can be thought of as looking at the entire city, instead of a few blocks. Global organization considers all cells in a tumour nest by connecting the centre of each cell where the total length of all the connecting lines is minimized. One of the most widely used descriptors is Euclidean minimum spanning tree (EMST) (Figure 1-1D) A Feulgeun-Thionin stained B Voronoi diagram C Delaunay triangulation D Euclidean minimum spanning tree     Figure 1-1. Examples of Tissue Architecture Derived from QTP Analysis From left to right: (Fig. 1-1A) example of Feulguen-Thionin stained tumour nests demarcated as shown with blue border; (Fig. 1-1B) Voronoi diagram with each cell nuclei correspond to red points which are partitioned by Voronoi polygons (blue). (Fig. 1-1C) Delaunay triangulation of all cell nuclei; and (Fig. 1-1D) Euclidean minimum spanning tree (EMST) connection of cell nuclei corresponds to cell density taken from 1C and 1D.   Combining information from nuclear phenotype and tumour nests architecture allows us to develop the concept of Cell Sociology, the spatial distribution of different groups of cells within a tissue (Figure 1-2).  Morphology ArchitectureCell SociologyNuclei (Cells) Tissue OrganizationStudy of spatial proximity and correlation between interacting cells+17  Figure 1-2. QTP Analysis and Cell Sociology  1.3.3 Applications of QTP in Cancer Research Cancer Progression Image analysis has been used in studying changes of nuclear features in pre-invasive phases and their correlation with risk of cancer progression, including those from oral cavity, cervix, lung, prostate, and colorectal regions. For OSCC Guillaud et al.141 presented a promising risk assessment tool for oral cancer progression by constructing and building a model to recognize nuclear phenotypes that are significantly discriminative between normal, mild, moderate, severe dysplasias, carcinoma in situ, and SCC tissues. Collectively, the significantly different phenotypes against the normals made up a nuclear phenotype score, which was reported to be correlated with clinical characteristics and molecular phenotypes such as loss of heterozygosity. For cervical cancers, the same group also applied image analysis to relate state of cell proliferation in epithelial layers against varying grades of cervical intra-epithelial neoplasia (CIN), HPV status, and smoking habits.142 They were able to quantify and demonstrate an increase cell proliferation with increase disease grade, HPV-positive cases and in normal smokers. In colorectal cancer, Mulder et al.143 studied the relationship between genetic mutations commonly seen in colorectal cancer (ras proto-oncogene, adenomatous polyposis coli (APC), mutated in colorectal carcinoma (MCC), deleted in colorectal carcinoma (DCC), and p53 proto-oncogene) and nuclear changes measured by image analysis. They found significant correlation between the irregularity of nuclear area and shape and progression of adenoma and, interestingly, the nuclear feature is independent from genetic mutations. From this study, the authors suggested that instead of genetic mutations, nuclear phenotypic changes are contributed by other factors in the extracellular matrix. Dong et al.144 demonstrated the ability of computed nuclear features to discriminate low-risk ductal hyperplasia from various degree of high-risk ductal carcinoma in situ with AUC from 0.90 to 0.97.   18  Cancer Survival In breast cancer, Tambasco et al.145 demonstrated circularity of nuclear shape and total perimeter had shown to be significantly associated with survival with an AUC to 0.73 for disease-specific survival (DSS) and 0.75 for overall survival (OS). Wang LW et al. 146 was able to quantified features of tumour nests on tissue microarrays of 202 invasive ductal carcinomas. The patients’ 5-year disease-free survival was found to be negatively correlated with the number of tumour nests, circularity, and perimeter of the tumour nests. The hazard ratios of these three parameters were higher than N stage and hormone receptor status. Beck et al.147 provided a robust prognostic model built from features measured in both breast cancer epithelium and stroma. Epithelial features associated with survival concerned with nuclear texture, intensity, chromatin localization and nuclear peripheral changes. The authors also discovered features in stroma variation that predicted prognosis. The association of nuclear features and their prognostic values has also been reported for bladder, esophageal, and colorectal cancers.148-150 However, there has not been any study on prognostic value in OSCC. Cancer Metastasis There have also been investigations on correlating QTP features and metastasis outcomes in aim to aid the decision in treatment. Zarella et al.151 studied 101 breast cancer primary tumours from which they built a metastasis score based on features regarding shape, architectural, and color. They demonstrated the discriminative ability for two thirds of the study cohort with sensitivity of 86% and specificity of 90% at thresholds >0.606 and <0.460.  Veltri et al.152 reported findings from constructing tissue microarrays of 182 matched pairs of tumour areas and their adjacent normal regions of prostate cancerous specimen. From the 60 calculated nuclear features, the authors proposed a model with AUC of 79.9% for predicting metastatic prostate cancers. This study also highlighted the hidden metastatic changes in the normal appearing nuclei surrounding the peripheral zone of cancerous area. In regarding lymph node metastasis in OSCC, a group in India analyzed 16 preoperative incisional biopsy tissues. They reported significant differences observed in node-positive and node-negative in terms of nuclear circularity and variation in nuclear area.153  Karino et al.154 analyzed H&E and QTP images of 88 preoperative OSCC biopsies 19  (46 N+ and 42 N0). They observed significantly larger nuclear area and perimeter in N+ cases.  1.3.4 Study of Tumour Nests Using QTP As introduced in Section 1.2.4, infiltrative tumours are thought as a work of group of cells (i.e. tumour nests) rather than actions of single cells. These aggregates of tumour cells appear as either ill-defined and irregular or well-defined circular or oval shape. Perhaps cells in these aggregates are at the highest activity of interaction with the tumour microenvironment to facilitate spreading. In current literature, there is little research on study tumour nests and nodal disease in OSCC. Together with the objectivity and reproducibility offered by QTP analysis, we aim to investigate pathological features of tumour nests of node-negative or node-positive OSCC.  1.4 Organization of Thesis  The theme of this thesis centralizes around the disease burden of nodal disease in OSCC patients. The introduction in Chapter 1 is followed by Chapters 2 and 3 presenting two studies, each with different objectives, research questions or hypothesis, methodology, results, and discussion. The concluding chapter, Chapter 4, contains an overall discussion, summary of research findings, and future directions. Chapter 2, “Disease Burden of Lymph Node Metastasis in Oral Cancer Patients, a Province-Wide Experience”, presents evidential data from a population-based cohort with retrospective review on disease history of OSCC patients in British Columbia (BC).To our knowledge, there has not been any BC-based research reviewing and investigating the determinants of nodal disease of OSCC patients. A major advantage of conducting such retrospective study is the wealth of collected data can be used to generate other research questions and to direct subsequent prospective investigations. From this study, we can start to understand the occurrence and incidence of nodal disease for BC population, as well as scientifically elucidate solutions to address the lack of biological understanding and clinical dilemma in existing prospective studies,155 and future clinical studies. Research activities in this study were approved by the University of British Columbia/BC Cancer Agency (BCCA) Research Ethics Board (REB# H11-00659).  20  In Chapter 3, “Novel Computational Image Analysis to Identify the Risk of Nodal Metastasis in Oral Cancers”, we report a pilot study on utilizing QTP analysis to measure microscopic phonotypical changes and explore their potential as biomarkers in predicting risk of nodal disease in OSCC.  The study was approved by the BC Cancer Agency Research Ethics Board (REB# H09-03090) and Vancouver Coastal Health Research Institute (VCHRI Study #V10-03090)   21         CHAPTER 2 2 DISEASE BURDEN OF NODAL METASTASIS IN ORAL CANCER PATIENTS, A PROVINCIAL WIDE EXPERIENCE  22  2.1 Guiding Questions and Objectives The research questions guiding this retrospective study are as follows. i. What was the incidence of primary OSCC in BC? What was the nodal status at initial presentation for these patients? ii. What were the characteristics of these patients, in demographics and risk factors, and the clinico-pathological attributes of the primary tumours? iii. How were these patients managed? iv. What are the outcomes of these patients, with or without nodal disease? v. Were there any identifiable factors associated with survival or nodal disease? Specific objectives are: i. Identify patients with primary OSCC from the BCCA Registry and to collect demographics, risk factors of the patients, clinic-pathological information of the tumour and cervical nodal status, treatment, and outcome through chart review.  ii. Determine factors impacting survival of all patients iii. Investigate neck management for early-stage clinically node-negative (cN0) patients and associated survival. iv. Identify potential predictors for the development of nodal disease of cN0 patients?   2.2 Methodologies  This is a population-based retrospective study using data extracted from cancer registry and chart review.  2.2.1 Patient Cohort In BC, cancer is a registered disease. Patients with suspicious oral lesions will receive a biopsy by dental or medical specialists. When an oral specimen is diagnosed as malignant, a copy of pathology report will be faxed to the BCCA Registry. From there, patient is referred and under the care of BCCA for consultation on management plans. Since such registration and referral pathway is centralized, we have the opportunity to retrospectively collect cases at a population level and to study the natural disease history. In order to collect enough sample size and ensure enough follow-up time for outcome 23  information, we identified patients with primary OSCC patients diagnosed between January 2nd, 2001 to December 24th, 2007.  The inclusion and exclusion criteria are detailed as follows.  Inclusion Criteria Patients included in the retrospective review were: • Diagnosed and registered with OSCC (histology code, 80703, 80713, 80513, 80103, 80203, 80763, 80003) between 2001 and 2007, and coded anatomical sites of OSCC include ICD-10 Code C02.0 to C06.9, including tongue, floor of mouth, soft palate, hard palate, gingiva, and buccal mucosa (see Appendix A).  Exclusion Criteria Patients excluded from the retrospective review were: • Diagnosed with OSCC at the base of tongue (ICD-10 code C01) or external or internal aspect of lip (ICD-10 code C00).  • Having history of OSCC prior to 2001.  • Having only pathology registration without any related clinical information. There were no exclusion criteria based on patient age, sex, or ethnicity.   2.2.2 Date Source Cancer Agency Information System (CAIS) The BCCA houses a centralized database called CAIS (Cancer Agency Information System). CAIS centralizes all records of reported cancer cases referred to the agency. CAIS stores information which can be categorized into two types: 1) variables with standardized terms and conventions that can be operationalized and 2) verbal and non-standardized descriptions that are often found in follow-up progress notes or correspondence among clinicians.  Date Abstraction For each identified OSCC patient, chart review was performed under supervision in order to ensure data accuracy. Relevant covariates were collected using electronic PDF 24  forms, which were designed to preserve completeness, accuracy, consistency, and objectivity of data coding (Appendix B to Appendix D). The Patient Demographics and Consultation Form (Appendix B) is used to collect demographics (date of birth, sex, and smoking history), tumour characteristics at initial presentation (anatomical site and laterality), and initial treatment type (surgery, radiotherapy, or chemotherapy) and treatment intent (curative or palliative). Results from neck palpitation and imaging tests (Computed Tomography (CT), X-ray, Magnetic Resonance Imaging (MRI) or Positron Emission Tomography (PET)) were also recorded in order to capture preoperative clinical nodal staging results. The form also collects patients survival status (living, death from disease, and death from other causes), and dates of last date known alive or date of death. The Pathology Report Form (Appendix C) is used for all patients to collect dates and findings from the diagnostic biopsy for primary tumour (incisional / excisional tissue biopsy or fine needle aspirate biopsy). It is also used for all patients who received surgical excision as initial treatment in order to collect pathological variables (tumour differentiation, tumour depth of invasion, and surgical margin status). If available, other core pathology information (POI, presence of perineural, lymphovascular, muscle, or bone invasion, presence of inflammation, tumour growth pattern, and invasion focality) are also captured. If performed, details on type of ND, levels of neck lymph node dissected, and pathological information (extracapsular spread and the largest size of diseased lymph nodes) are also recorded.  The Radiotherapy / Chemotherapy Report (Appendix D) is used for all patients who received radiation (RT) or chemotherapy (Chemo). Treatment intent (curative, adjuvant or palliative), RT start date / end date, treatment modifier (conventional, altered fractionation or intensity modulated), dosage (cGy) and fraction, and Chemo intent were collected.   The Pathology Form and Radiotherapy / Chemotherapy Report Form need to be completed for any surgical, RT, and Chemo treatments a patient has ever received after 25  the initial treatment for primary OSCC, regardless of which organ site. This allows us to capture any recurrences of disease, or development of second cancer, or metastasis. Once completed, the forms are saved and read into a data collection software (HP Teleform®, Hewlett-Packard Development Company, L.P, CA, USA) through which data can be output in both raw form and prescribed data codes into Microsoft Access Database. Each patient was assigned with a 4-digit unique non-personal patient identifier in order to link data collected from individual forms together in the database. In addition, accuracy of data was reinforced by revisiting patient’s chart whenever questionable entry rises. Examples of such are large number (e.g. 40 mm) of tumour depth of invasion, no treatment dates (e.g., palliative care due to unresectable tumour or under a patient’s wish), or early age of a patient (e.g., 25 years old).  2.2.3 Statistical Analysis Statistical analysis was performed in R (V3.23) with built-in functions and packages. Tests were performed two-tailed with P < 0.05 was considered significant.  Study Outcomes Study outcomes include disease outcomes and time to survival outcomes. Disease Outcomes a. Local Recurrence (LR): Defined as the reappearance of OSCC at the same anatomical region as the primary tumour in at least 3-month post-initial treatment. Time-to-LR is calculated from the time of initial treatment to first post-treatment pathological diagnosis of severe dysplasia (D3), carcinoma in situ (CIS), verrucous carcinoma (VC), or SCC on the same anatomical region of the oral cavity. b. Regional Recurrence (RR): Defined as the reappearance of squamous cancerous cells, either documented with imaging scans or fine needle biopsy pathology report, in the cervical lymph nodes in patients who have had prior history nodal disease. c. Regional Failure (RF): Defined as presence of squamous cancerous cells in the neck cervical lymph nodes in patients who have not had prior history of nodal disease. Time-26  to-RF is calculated from time of initial treatment to pathologically proven node positive (pN+).  d. Second Cancer (SC): Defined as development of new, unrelated cancer, excluding non-melanoma skin cancer, during survivorship after initial treatment of the primary tumour. Second cancers can happen in the oral cavity at different anatomical site or other organs or sites.  e. Distant Metastasis (DM): Defined as the presence of SCC outside of the oral cavity or neck followed the initial treatment of the primary tumour. The relationship to primary OSCC was clearly dictated in patients’ medical charts.  Survival Outcomes a. Overall Survival (OS): Defined as death due to any cause. Time-to-Death (year) is calculated from the date of initial treatment to the date of death due to any cause, or last known date alive b. Disease-specific Survival (DSS): Defined as death due to primary OSCC, either at local, regional or distant sites. Time-to-DSS (year) is calculated from the date of initial treatment to the date of death due to primary OSCC, or last known date alive.  Frequencies and Descriptive Data Patient demographics, clinical, and tumor characteristics were summarized in descriptive absolute and relative frequencies, and means with standard deviation (SD). Categorical variables were compared by Chi square or Fisher’s exact test. Continuous variables were compared with Student’s t-test. Association of Variables to Study Outcomes Kaplan-Meier (KM) survival analysis was used to estimate the proportion of patients reaching outcomes over time with log-rank tests for comparing the differential survival rates between subgroups of demographic, clinical, or pathological variables. Univariate and multivariate Cox proportional-hazards regression analysis was used to estimate the effect of potential prognostic factors (age, gender, smoking history, lesion site, tumor size, 27  presence of nodal disease, or initial treatment) on risk of outcomes. Cumulative incidence rate of RF was performed after adjusting for death from any cause, a known competing risk factor.  Performance of Depth of Invasion in Discriminating Nodal Status Receiver operating characteristics (ROC) curve analysis was used to determine the ability of DOI in discriminating between outcomes of no RF or RF among cN0 patients. Area under the curve (AUC) is quantified over a range of continuous values of DOI. We defined the best cut-off point as the point along the curve where we reach maximum sensitivity and specificity.  2.3 Results From the BCCA registry, 3,054 patients were identified from the Head and Neck Tumour Group over 7 year period (2001-2007). Among these, 847 (28%) were diagnosed with OSCC, dictated within the anatomical regions of ICD-09code 02.0 to 06.9. To keep on course with the theme of nodal disease burden, the data are presented in the following fashion. We first describe the entire study cohort by nodal status at initial presentation of OSCC (N=847, Section 2.3.1). We then dedicate the remaining Chapter 2 on clinically node-negative (cN0) patients who received surgery as the intent-to-cure treatment, by focusing on neck management (N=469; Section 2.3.2); and by reporting regional failure in cN0 patients without any neck treatment (N= 322, Section 2.3.3) to observe the nature history of the nodal disease and to identify potential predictors. Figure 2-1 diagrams the branching of the OSCC population by nodal status at time of initial presentation (yellow shaded), to the initial treatment type for primary OSCC with any upstaging of nodal status (gray shaded), and to nodal status (orange shaded) through to the end of follow-up or death.   28   Figure 2-1. Head and Neck Cancer OSCC Patients Identified from the BCCA Registry (2001-2007) From top to bottom, all head and neck patients (light gray) by diagnosis and anatomical sites; OSCC patients with nodal status at initial presentation (yellow), initial treatment for primary OSCC tumour by curative intent, delivery of neck dissections, and nodal staging (blue); events of nodal disease development during clinical follow-up (pink). Solid connecting lines are periods from initial presentation to end of initial treatment and start of clinical follow-up. Dashed connecting lines are periods from start the clinical follow-up to last follow-up or death. OSCC, Oral Squamous Cell Carcinoma; cN0, clinically Node-negative at the time of initial presentation; cN+, clinically Node-positive at the time of initial presentation; RT, radiotherapy; Chemo, chemotherapy; END, elective neck dissection; pN0, pathologically proven Node-negative at the time of surgery; pN+, pathologically proven node-positive at the time of surgery or during follow-up; RF, Regional Failure, development of pN+ in cN0 patients at follow-up. OSCC(847)Clinically N0cN0 (609, 72%)Surgery(469, 76%)with END(100, 21%)with END and RT(25, 5%)with RT(22, 5%)Local Excision Only(322, 69%)RF pN+(89, 28%)No RF, N0(233, 72%)upstaged pN+(9, 9%)pN0(91, 91%)RF pN+(10, 10%)(Chemo) Radio(140, 23%)Clinically N+cN+ (238, 28%)Surgery(114, 47%) (Chem0) Radio (124, 51%)With ND(31, 27%)With ND and RT(63, 55%)With RT(5, 4%)Local Excision Only(15, 13%)RFpN+(4, 18%)upstaged pN+pN+ (6, 24%)pN0(19, 76%)RF pN+(2, 11%)Initial PresentationInitial TreatmentFollow-upRR pN+(2, 13%)RR pN+(4, 13%)RR pN+(17, 27%)RR pN+(13, 10%)RF pN+(17, 12%)Head and Neck(2001 – 2007)N = 3,068Pharynx(1,485 48%)Oral Cavity(1,116, 36 %)Major Salivary Gland(302, 10%)Lip(165, 5%)BCCA Tumour Group29  2.3.1 All Primary OSCC (N=847) Demographics and Tumour Characteristics In this cohort, the average age of OSCC patients were 65 years old, there were more males (58%), ever smokers (68%), were affected on high risk lesion sites (either at the tongue, 44%, or floor of mouth, 22%), and more early T-stage (73%). Among these, 28% presented with nodal diseases at the time of initial diagnosis (cN+).  Patient demographics, smoking habits, and clinical tumour characteristics at initial presentation are summarized in Table 2-1. Table 2-1. Demographics and Clinical Characteristics at Initial Presentation (N=847)   OSCC, N(%) cN0, N(%) cN+, N(%) P  Variables* 847 609 238 Age, years; mean±SD 65.1±14.1 65.1±14.3 65.0±13.5 0.89 Age Group    0.48 <45 74 (9) 55 (9) 19 (8)  45-65 352 (41) 245 (40) 107 (45)  >65 421 (50) 309 (51) 112 (47)  Gender    0.04 Male 493 (58) 341 (56) 152 (64)  Female 354 (42) 268 (44) 86 (36)  Smoking History    0.10 Never 205 (24) 154 (25) 51 (22)  Ever 578 (68) 397 (72) 181 (78)  Missing 64 (7) 58 (9) 6 (2)  Lesion Site Risk of Cancera    0.25 Low 229 (27) 165 (27) 64 (27)  Intermediate 68 (8) 43 (7) 25 (10)  High 550 (65) 401 (66) 149 (63)  Clinical T Stage    <0.001 T1 320 (38) 284 (47) 36 (15)  T2 298 (35) 216 (35) 82 (34)  T3 105 (12) 45 (7) 60 (25)  T4 124 (15) 64 (10) 60 (25)  Death due to oral cancer 257(30) 145 (24) 112 (47) <0.001 Time to death due to oral cancer or last follow-up, year; mean±SD 3.9±3.5 4.5±3.5 2.6±3.1 <0.001 aLesion Site Risk: Low, buccal mucosa, gingiva, or hard palate; Intermediate, soft palate or soft palate complex; High, tongue or floor of mouth OSCC, oral squamous cell carcinoma; cN0, clinically node-negative; cN+, clinically node-positive  30  Nodal Disease Burden The majority of the OSCC patients were initially presented without clinically evident nodal disease (cN0; 609, 72%). Comparing cN0 and cN+ at the time of initial disease presentation, there were no differences in age, smoking history, or lesion site risk; but slightly more males in cN+ group. Moreover, in the cN+ group, there were significantly more patients presented with larger T-stage (T3, N=60 or T4, N=60, 50%vs. 17%), and died due to oral cancer (47% vs.24%, P<0.001). The median time to death due to oral cancer or last follow-up was significantly shorter in the cN+ group (4.5±3.5 vs. 2.6±3.1 years, P<0.001) (Table 2-1).  Treatment Modality at Initial Disease Presentation Table 2-2 summarizes types of initial treatment received. Compared to cN0 patients, the majority of cN+ patients received curative radiotherapy, with or without concurrent chemotherapy (RT +/- Chemo) (52% vs. 23%, P<0.001). There were significant differences in treatment received among different T staging groups. For cN0 patients who received RT +/- Chemo, more were with advanced T3/T4 stage (55% vs. 16%T1/T2, P<0.001); vice versa, those who received surgery were mostly in early T1/T2 stage (84% vs. 45%). Similar trend was also seen among cN+ patients. Table 2-2. Initial Treatment among OSCC Patients, by Clinical N and T Stage (N=847)  OSCC (N=847) cN0 (N=609) cN+(N=238) Initial Treatment cN0 (609) cN+ (238) P T1/2 (500) T3/4 (109) P T1/2 (118) T3/4 (120) P Surgerya 469 (76) 114 (47) <0.001 420 (84) 49 (45) 0.001 66 (56) 48 (40) 0.01 RT +/- Chemob 140 (23) 124 (52)  80 (16) 60 (55)  52 (44) 72 (60)  aSurgery includes local excision only, local excision with neck dissection b RT+/- Chemo includes local excision and neck dissection with or without radiotherapy; radiotherapy with or without chemotherapy RT, radiotherapy; Chemo, chemotherapy; cN0, clinically node-negative; cN+, clinically node-positive Disease OutcomeTable 2-3 summarizes the frequency, mean and median time to each disease outcome as documented during the course of patient’s follow-up period. The mean follow-up time for this cohort was 3.9±3.5 years after initial treatment which allowed us to observe incidences outcomes. A total of 102 (12%) patients developed LR with median time of 1.2 years, whereas 137 (16%) developed RF of which 56 (7%) had RR within median 6 months. There were a small portion of patients (27, 3%) patients experienced both LR and RR. In regards to development of SC, 143 (17%) developed with 31  median time of 2.5 years. Of interest, the most common site of SC is oral cavity (65, 55%) followed by lung (13,11%) and oro-pharynx (8,7%) (Figure 2-2; SC, light gray bars). Among incidence of SCC metastasis to distant organs (28, 3%), lung is the most common site (19, 79%) followed by oro-pharynx (1, 4%) and prostate (1, 4%) (Figure 2-2; DM, dark gray bars).Strikingly, for the3% of DM, the median time was only 1.5 year following initial treatment (25%-75% percentile, 1.1-2.6 years). Table 2-3. Disease Outcomes among OSCC Patients (N=847) Disease Outcomes OSCC (N=847) Median (25th-75thpercentile) Mean±SD (year) LR 102 (12) 1.2 (0.5-3.7) 2.5±2.8 RF 137 (16) 0.8 (0.4-1.7) 1.5±2.0 RRa 56 (7) 0.6(0.5-1.2) 1.1±1.2 LR-RR 27 (3) 0.6 (0.4-3.1) 2.1±2.8 SC 120 (14) 2.5 (1.0-5.1) 3.3±2.7 DM 24(3) 1.5(1.1-2.6) 2.1±1.7 aRegional Recurrence was determined as relapse of nodal disease after previous therapeutic treatment OSCC, oral squamous cell carcinoma; LR, local recurrence; RF; regional failure; RR, regional recurrence; LR-RR, loco-regional recurrence; SC, second cancer; DM, distant metastasis    32    Figure 2-2. Disease Outcome of Second Cancer (SC) or Distant Metastasis (DM) among OSCC Patients (N=847) Light gray bars represent number of patients who developed second cancer at corresponding sites. Dark gray bars represent proportion of patients who development distant metastasis at corresponding organ site.  aOther new primary sites include: Blood (1), Bone (1), Cervix (1), Kidney (1), Liver (1), Melanoma (1), Muscle (1), Ovary (1), Pancreas (1); Other distant metastasis (DM) sites include: Bone (2);Liver (1). UADT, upper aero-digestive tract    55%11%7% 5% 4% 3% 3% 2% 2%8%79%4% 4% 13%-30-20-10010203040506070OralLungOro-pharynxProstateUADTBreastUrinary TractColonGastricOtherNumber of PatientsSecondary Cancer (SC) Distant Metastasis (DM)a Other33  Disease Outcome by Initial Nodal Status Leading from the observation of disease relapse, possibly due to failure to surgery or radiotherapy, we investigated the incidence of disease outcomes based on the initial nodal status (cN0 or cN+). The decision to separate patients into these two groups is due to aggressiveness of nodal stage and its influence in deciding treatment modality, which may impact differently on post-treatment disease status.  Table 2-4 summarizes the frequency and the median time to each disease outcome among cN0 and cN+, grouped by initial treatment received (“Surgery” - surgery only or “RT” - treatment with RT). Notable differences concerning disease outcomes were observed among cN0 patients. Between the two treatment group, there was no difference in number of patients who experienced LR; however, the median time to LR was shorter for patients who received RT (2.6 years vs. 0.7 years, P<0.01). Also, while there was higher incidence of RF in the Surgery group (26% vs. 16%, P<0.01) time to RF was similar compared to Radiotherapy group. We did not perform comparison analysis for other disease outcomes for cN+ patients due to low number of outcomes and would not yield meaningful results.   34  Table 2-4. Disease Outcomes, by Clinical N Stage and Treatment (N=847)  Disease outcomes, N (%) Clinical Nodal Stage (N) LR  RFd  RR  LR-RR  SC  DM   P  P  P  P  P  P cN0 (609) 70 (11) 0.78 137 (22) <0.01 20 (15) 0.37 18 (3) 0.59 89 (15) 0.84 10 (2) - Surgerya (422) 50 (12)  108 (26)  14 (13)  14 (3)  63 (15)  7 (2)  RTb (187) 20 (11)  29 (16)  6 (21)  4 (2)  26 (14)  3 (2)  Timec  <0.01  0.66  0.24  0.65  0.61  0.43 Surgerya 2.6 (1.2-2.8)  0.8 (0.4-0.9)  0.8 (0.6-0.9)  1.3 (1.2-1.4)  2.9 (1.3-3.0)  2.4 (1.4-2.7)  RTb 0.7 (0.4-0.7)  0.7 (0-0.67)  1.1 (0.5-1.7)  0.8 (0.6-0.9)  3 (1.0-3.0)  2.2 (1.2-2.4)               cN+ (238) 32 (13) - - - 36 (15)  9 (4)  30 (13)  14 (6)  Surgerya (46) 1 (2)    6 (13)  0  1 (2)  2 (4)  RTb (192) 31 (16)    30 (16)  9 (5)  29 (15)  12 (6)  Timec  - - -  -  -  -  - Surgerya -    0.5 (0.5-0.5)  -  -  -  RTb 0.5 (0.3-0.5)    0.6 (0.4-0.6)  0.5 (0.4-0.5)  1.1 (0.6-1.2)  1.4 (0.8-1.5)  aSurgery includes local excision only, local excision with neck dissection b RT+/- Chemo includes local excision and neck dissection with or without radiotherapy; radiotherapy with or without chemotherapy cMedian time in years with 25th to 75th percentile in brackets dRF (regional failure), by definition, does not apply to cN+ RT, radiotherapy; LR, local recurrence; RF, regional failure; RR, regional recurrence; LR-RR, Loco-regional recurrence; SC, second cancer; DM, distant metastasis; cN0, clinically node-negative; cN+, clinically node-positive. 35  Overall Survival and Disease Specific Survival In this section, we report the survival rates for the entire OSCC study population, within the first 5 years from time of initial treatment, using Kaplan-Meier estimation with Log-rank tests. Cox proportional hazards (Cox PH) regression analyses was used to identify patient demographics or tumour characteristics associated with increased risk of survival outcomes.  5-year Overall Survival (OS) Of the 847 OSCC patients, 447(53%) had diedwithin5 years after initial treatment. Most frequent causes of death, in descending order, were primary cancer for 246 (55%), second primary for 114 (25%), and other cause/unknown cause for 87 (20%) (Figure 2-3)  OSCC, N=847 Median Survival, years (95% CI) Estimated Survival Rates, % (95% CI) 2 years 3 years 5 years Death due to any cause 3.8 (3.1-4.6) 61.1 (57.9-64.6) 53.8 (50.4-57.3) 44.9 (41.5-48.6) Figure 2-3. Kaplan-Meier Survival Curve and Rates for OS;(OSCC,N=847) OSCC, oral squamous cell carcinoma; CI, confidence interval   36  Table 2-5 summarizes time-to-death analyses by Cox PH regression. In univariable analyses, older age groups (45-65 years and >65 years), larger T-stage (>T1), cN+ status, and late-TNM-stage disease were highly associated (P<0.001) with OS. Ever smoking history also showed significant association in elevated risk of death (P=0.04). In multivariate analysis, patients, in descending order of risk, of age>65 years, with T4 size tumour, cN+, or late TNM-stage disease were associated elevated risk of death of any cause. Table 2-5. Univariate and Multivariate Cox Proportional Hazards Analysis in 5-year OS (OSCC, N=847) Variables Univariate Multivariate HR (95% CI) P HR (95% CI) P Age Group     <45 1    45-65 1.6 (1.0-2.7) 0.05   >65 2.9 (1.8-4.6) <0.001 1.9 (1.5-2.3) <0.001 Gender     Male 1    Female 1.0 (0.8-1.2) 0.98   Smoking History     Never 1    Ever 1.3 (1.0-1.6) 0.04   Lesion Site Risk of Cancera     Low 1    Intermediate 0.8 (0.6-1.2) 0.4   High 0.8 (0.7-1.0) 0.05   Clinical T Stage     T1 1    T2 1.8 (1.4-2.3) <0.001   T3 2.9 (2.1-3.9) <0.001   T4 4.2 (3.2-5.6) <0.001 1.6 (1.2-2.1) <0.001 Clinical N Stage     cN0 1    cN+ 2.5 (2.1-3.1) <0.001 1.4 (1.0-1.9) 0.02 Clinical TNM Stage     I/II 1    III/IV 3.0 (2.5-3.6) <0.001 1.8 (1.3-2.5) <0.001 aLesion Site Risk: Low, buccal mucosa, gingiva, or hard palate; Intermediate, soft palate or soft palate complex; High, tongue or floor of mouth cN0, clinically Nodal negative; cN+, clinically Nodal positive; HR, hazard ratio; CI, confidence interval  37  In Kaplan-Meier analysis for 5-year OS, the median time was 3.8 years (95% CI, 3.1-4.6), and with survival rates 61% (57.9-64.6), 53.8% (50.4-57.3), and 44.9% (41.5-48.6) at 2, 3, and 5 years after initial treatment, respectively. Comparison of survival rates between sub-groups of patient in demographics or tumour characteristics are described in Table 2-6 with Kaplan-Meier curves displayed in Figure 2-4. In ascending order, patients who had T4 tumour, or were cN+, or were at late-TNM-stage (TNM III/IV), or were >65 years old, or were ever smokers had significantly worse 5-year OS compared to their counterparts with P<0.001 (Table 2-6) Table 2-6. Kaplan-Meier Estimated OS Rates, by Variables (OSCC, N=847)  Median, years (95% CI) Estimate Survival Rates, % (95% CI) P Age Group 5-year 2 year 3 year 5 year <0.001 <45 NR 77.2 (67.6-88.1) 73.6 (63.6-85.3) 70.0 (59.5-82.4)  45-65  NR 68.4 (63.2-73.6) 62.0 (57.0-67.4) 53.4 (48.2-59.1)  >65  2.2 (1.7-2.9) 52.5 (47.9-57.6) 43.9 (39.3-49.0) 34.2 (29.8-39.2)  Gender 5-year 2 year 3 year 5 year 0.98 Male 3.8 (2.9-NR) 61.7 (57.5-66.3) 53.6 (49.2-58.3) 45.3 (40.9-50.1)  Female 4.0 (2.9-NR) 61.0 (56.0-66.5) 54.8 (49.7-60.5) 45.3 (40.2-51.1)  Smoking  5-year 2 year 3 year 5 year 0.04 Never NR 64.3 (57.9-71.4) 60.0 (53.5-67.3) 53.0 (46.3-60.6)  Ever 3.14 (2.5-4.2) 59.2 (55.3-63.5) 50.8 (46.8-55.1) 42.4 (38.4-46.8)  Lesion Site Risk of Cancera 5-year 2 year 3 year 5 year 0.14 Low 3.1 (2.2-4.4) 58.0 (51.8-64.8) 50.1 (43.9-57.1) 40.0 (33.9-47.1)  Intermediate 3.5 (2.0-NR) 61.3 (50.7-74.2) 52.3 (41.6-65.8) 44.3 (33.8-58.1)  High 4.3 (3.5-NR) 63.0 (58.9-67.3) 56.1 (51.9-60.6) 47.7 (43.5-52.3)  Clinical T Stage  5-year 2 year 3 year 5 year <0.001 T1 NR 78.7 (74.2-83.5) 72.6 (67.7-77.9) 62.0 (56.5-67.9)  T2 3.5 (2.6-NR) 62.8 (57.5-68.7) 53.2 (47.7-59.3) 43.8 (38.3-50.0)  T3 1.1 (1.0-2.2) 41.7 (33.1-52.6) 35.7 (27.4-46.4) 31.4 (23.4-42.0)  T4 1.0 (0.7-1.4) 30.2 (23.0-39.6) 24.0 (17.4-33.1) 17.1 (11.3-25.7)  Clinical N Stage  5-year 2 year 3 year 5 year <0.001 cN0  NR 70.9 (67.3-74.4) 63.3 (59.4-67.4) 53.7 (49.7-58.1)  cN+ 1.2 (1.1-1.6) 36.9 (31.2-43.7) 30.2 (24.8-36.8) 23.2 (18.3-29.5)  Clinical TNM Staging 5-year 2 year 3 year 5 year <0.001 I/II  NR 77.6 (73.9-81.4) 69.6 (65.5-73.8) 59.9 (55.5-64.6)  III/IV  1.2 (1.1-1.6) 38.2 (34.4-44.8) 32.8 (28.1-38.2) 25.0 (20.7-30.2)  aLesion Site Risk: Low, buccal mucosa, gingiva, or hard palate; intermediate, soft palate or soft palate complex; High, tongue or floor of mouth cN0, clinically Node-negative; cN+, clinically Node-positive; CI, confidence interval; NR, not reached 38  A. Age Group B. Gender C. Smoking History D. Lesion Site Risk of Cancer     E. Clinical N Stage F. Clinical T Stage G. Clincial TNM Staging      Figure 2-4. Kaplan-Meier Survival Curves of 5-year OS (N=847) Survival curves are shown for sub-group of patients stratified by A. Age Group (<45, 45-65, >65); B. Gender (Male or Female); C. Smoking History (Never or Ever); D. Lesion Site Risk (Low, Intermediate, or High); E. Clinical N Stage (cN0 or cN+); F. Clinical T Stage (T1,T2,T3 or T4); and G. Clinical TNM Stage (I/II or III/IV). P-values are derived from log-rank test with P<0.05. 39  5-year Disease-Specific Survival (DSS) Of the 847 OSCC patients, 246(29%) reported deaths due to oral cancer in 5 years after initial treatment. Disease-specific survival (DSS) rates were 74.7% (95% CI, 71.6-77.9), 69.9% (95% CI, 66.5-73.3), and 66.5% (95% CI, 63.0-70.1) at 2, 3, and 5 years post-treatment, respectively (Figure 2-5).    OSCC, N=847 Median Survival, years (95% CI) Estimated Survival Rate, % (95% CI) 2 years 3 years 5 years Death due to oral cancer  NR 74.7 (71.6-77.9) 69.9 (66.5-73.3) 66.5 (63.0-70.1) Figure 2-5. Kaplan-Meier Survival Curve and Rates on 5-year DSS (N=847) OSCC, oral squamous cell carcinoma; CI, confidence interval; NR, not reached      40  In univariate Cox-PH analysis for time-to-death due to oral cancer (Table 2-7), similar factors, except for smoking history, were associated with elevated risk of oral cancer related death as those seen in death by any cause (Table 2-5). In multivariable analysis, T4 size tumour, cN+ status, and late-TNM-stage were independent factors associated with higher risk of death from oral cancer (P< 0.05).  Table 2-7. Univariate and Multivariate Cox Proportional Hazards Analysis in 5-year DSS (OSCC, N=847)  Univariate Multivariate HR (95% CI) P HR (95% CI) P Age Group     <45 1    45-65 1.1 (0.7-2.0) 0.64   >65 1.8 (1.1-3.1) 0.02   Gender     Male 1     Female 1.0 (0.8-1.3) 0.92   Smoking History     Never 1     Ever 1.1 (0.8-1.4) 0.71   Lesion Site Risk of Cancera     Low 1    Intermediate 0.6 (0.4-1.2) 0.14   High 1.0 (0.7-1.3) 0.85   Clinical T Stage     T1 1    T2 2.1 (1.4-3.0) <0.001   T3 3.8 (2.5-5.8) <0.001   T4 5.4 (3.7-7.9) <0.001 1.7 (1.2-2.4) 0.002 Clinical N Stage     cN0 1     cN+ 3.0 (2.3-3.9) <0.001 1.5 (1.0-2.1) 0.04 Clinical TNM Stage     I/II 1    III/IV 3.7 (2.8-4.8) <0.001 2.1 (1.4-3.2) <0.001 aLesion Site Risk: low, buccal mucosa, gingiva, or hard palate; Intermediate, soft palate or soft palate complex; High, tongue or floor of mouth cN0, clinically Nodal negative; cN+, clinically Nodal positive; HR, hazard ratio; CI, confidence interval  41  In Kaplan-Meier analysis comparing 5-year DSS curves within each variable, similar characteristic associated with lower DSS rates were observed as those with OS. Patients with T4 tumour, or were cN+, or were late stage (III/IV), or were >65 years old, or were ever smokers had the significantly worst survival rate within their counterparts (Table 2-8 and Figure 2-6). Table 2-8. Kaplan-Meier Estimated DSS Rates, by Variables (OSCC, N=847) Variables Median Time, Years (95% CI) Estimate Survival Rates, % (95% CI) P Age Group 5-year 2 year 3 year 5 year <0.001 <45 NR 81.1 (72.0-91.4) 77.5 (67.6-88.7) 75.6 (65.5-87.3)  45-65  NR 80.2 (76.0-84.7) 76.3 (71.7-81.2) 71.8 (66.8-77.1)  >65  NR 68.5 (63.8-73.5) 62.7 (57.7-68.1) 60.6 (55.5-66.3)  Gender 5-year 2 year 3 year 5 year 0.71 Male NR 75.2 (71.2-79.4) 70.3 (66.0-74.8) 67.4 (63.0-72.2)  Female NR 73.8 (69.1-78.9) 69.3 (64.2-4.7) 65.7 (60.4-71.5)  Smoking  5-year 2 year 3 year 5 year 0.71 Never NR 75.2 (69.1-81.8) 71.4 (65.0-78.4) 67.2 (60.5-74.7)  Ever NR 74.1 (70.3-78.0) 68.7 (64.7-72.9) 65.7 (61.5-70.2)  Lesion Site Risk of Cancera 5-year 2 year 3 year 5 year 0.32 Low NR 73.7 (67.8-80.1) 68.8 (62.5-75.7) 65.9 (59.3-73.2)  Intermediate NR 82.1 (72.9-92.3) 77.8 (67.6-89.4) 75.5 (64.9-87.8)  High NR 74.1 (70.2-78.1) 69.3 (65.2-73.6) 65.9 (61.7-70.5)  Clinical T Stage  5-year 2 year 3 year 5 year <0.001 T1 NR 87.5 (83.7-91.5) 84.0 (79.7-88.4) 82.6 (78.2-87.3)  T2 NR 76.1 (71.1-81.4) 70.4 (65.0-76.2) 65.3 (59.6-71.6)  T3 NR 59.9 (50.5-71.2) 52.9 (43.1-64.9)   T4 1.7 (1.2-4.1) 47.1 (38.0-58.2) 42.8 (33.7-54.3) 35.0 (25.6-47.7)  Clinical N Stage  5-year 2 year 3 year 5 year <0.001 cN0  NR 82.7 (79.6-86.0) 78.2 (74.7-81.9) 75.5 (71.8-79.4)  cN+ 2.3 (1.7-NR) 52.9 (46.3-60.4) 47.2 (40.5-55.0) 42.3 (35.5-50.4)  Clinical TNM Staging 5-year 2 year 3 year 5 year <0.001 I/II  NR 86.5 (83.3-89.7) 82.2 (78.6-85.9) 80.0 (76.2-83.9)  III/IV  3.1 (2.1- NR) 56.3 (50.9-62.4) 50.6 (44.9-56.9) 45.4 (39.5-52.1)  aLesion Site Risk: Low, buccal mucosa, gingiva, or hard palate; Intermediate, soft palate or soft palate complex; High, tongue or floor of mouth cN0, clinically Node-negative; cN+, clinically Node-positive; CI, confidence interval; NR, not reached    42  A. Age Group B. Gender C. Smoking History D. Lesion Site Risk of Cancer     E. Clinical N Stage F. Clinical T Stage G. Clincial TNM Staging      Figure 2-6. Kaplan-Meier Survival Curves of 5-year DSS (OSCC, N=847) Survival curves are shown for sub-group of patients stratified by A. Age Group (<45, 45-65, >65); B. Gender (Male or Female); C. Smoking History (Never or Ever); D. Lesion Site Risk (Low, Intermediate, or High); E. Clinical N Stage (cN0 or cN+); F. Clinical T Stage (T1,T2,T3 or T4); and G. Clinical TNM Stage (I/II or III/IV). Difference in curves was compared by log-rank test.  43  2.3.2 cN0, Neck Treatment and Survival (N=469) In this section, we focus on neck management towards cN0 patients at the time of initial surgical treatment. We first report the incidence of nodal disease occurred and its impact on survival. This is followed by findings on potential prognostic factors on regional failure (RF). Lastly, we investigated the predictive value of tumour depth of invasion (DOI) in identifying a subpopulation of high-risk cN0 patients, who are likely to benefit from preventative neck treatment. Of the entire study cohort, 469 cN0 patients received curative surgical treatment. Among these, 447 were managed with either “watchful waiting” approach or prophylactic neck treatment at time of local excision (LE) of primary tumour. Sixty-nine percent (N=322) had local excision only (LE-only) and 27% (125) had LE with prophylactic neck treatment, which consists of either END alone (100, 21%) or END +/- RT (25 (5%). Table 2-9 summarizes and compares demographics and clinico-pathological characteristics between the different neck treatment groups.  Compared to LE-only, the prophylactic neck treatment group had significantly higher proportion of ever-smoker, of low or intermediate risk lesion site having cancer, of larger tumor size (>T1), of higher TNM stage of III/IV (mainly T-staging), and of greater tumour DOI.   44  Table 2-9. Demographics and Clinico-pathological Characteristics, by Neck Treatment Group (cN0, N=447) Variables, N (%) Neck Treatment Groupa  A (N=322) B (N=125) C (N=100) D (N=89) Pb A vs. B A vs. C B vs. D Age, years; mean±SD 63.2±14.9 63.7±14.2 64.0±14.4 62.7±13.1 0.71 0.62 0.58 Age Group            <45 42 (13) 10 (8) 7 (7) 7 (8) 0.33 0.25 0.45 45-65 133 (41) 54 (43) 43 (43) 46 (52)       >65 147 (46) 61 (49) 50 (50) 36 (40)       Sex            Male 181 (56) 75 (60) 57 (57) 53 (60) 0.53 0.98 0.97 Female 141 (44) 50 (40) 43 (43) 36 (40)       Smoking History               Never 97 (30) 25 (20) 19 (19) 30 (34) <0.001 <0.001 0.04 Ever 177 (55) 92 (74) 73 (73) 58 (65)       Missing 48 (15) 8 (6) 8 (8) 1 (1)       Lesion Site Risk of Cancerc Low 62 (19) 39 (31) 30 (30) 9 (10) <0.001 0.005 0.01 Intermediate 20 (6)     5 (6)       High 240 (75) 86 (69) 70 (70) 75 (84)       Clinical T Stage               T1 230 (71) 34 (27) 31 (31) 60 (67) <0.001 <0.001 <0.001 Not T1 92 (29) 91 (73) 69 (69) 29 (33)       Clinical TNM Staging               I/II 300 (93) 97 (78) 80 (80) 84 (94) <0.001 <0.001 0.002 III/IV 22 (7) 28 (22) 20 (20) 5 (6)       Tumor Grade               I 124 (39) 40 (32) 34 (34) 24 (27) 0.4 0.74 0.71 II 173 (54) 72 (58) 57 (57) 56 (63)       III 24 (7) 12 (10) 8 (8) 9 (10)       Missing 1 (0.3) 1 (0.8) 1 (1)         DOI, mm; mean±SD 4.3±4.9 10.3±8.0 9.5±7.7 5.6±5.6 <0.001 <0.001 <0.001 DOI of 4 mm               < 4 mm 177 (55) 24 (19) 22 (22) 38 (43) <0.001 <0.001 <0.001 ≥ 4 mm 112 (35) 91 (73) 70 (70) 46 (52)       Missing 33 (10) 10 (8) 8 (8) 5 (5)       Oral Cancer Death 65 (20) 30 (24) 18 (18) 40 (45) 0.45 0.74 0.002 Time to Death or Last Follow-up, years; mean±SD 5.0±3.7 4.6±3.7 5.0±3.8 4.3±3.3 0.37 0.98 0.53 aTreatment group: A, local excision (LE)-only; B, prophylactic neck treatment (LE+END(+/- RT)); C, elective neck dissection (LE+END); D, salvage neck treatment on cN0- -pN+ patients. bDistribution for continuous variables (Student t-test) and frequency of categorical variables (Chi-square test) were compared between A versus B, A versus C, and B versus D. cLesion Site Risk of Cancer: Low, buccal mucosa, gingiva, or hard palate; Intermediate, soft palate or soft palate complex; High, tongue or floor of mouth LE, local excision; END, elective neck dissection; RT, radiotherapy; DOI, depth of invasion 45  Nodal Disease Burden in cN0 Patients Of the 447 cN0 patients 3% clinically occult nodal disease was identified at the time of END and 23% RF developed during the follow up (Figure 2-1). However, not all those who were deemed as low risk early-stage OSCCs were free of regional disease; conversely, not all of those who received neck dissection were harbouring metastatic tumour, as we summarize the outcome of nodal disease for these treatment groups in Table 2-10. Of the LE-only group (N=322), 89 (28%) patients developed RF during follow-up; of these, 8 (9%) had RR within 1.0±0.8 years after therapeutic treatment for RF. These patients remained alive and disease free. For those who received prophylactic neck treatment (N=125), 12 (10%) developed RF; of these, over half (N=9) developed RR within 0.9±0.5 years. Table 2-10. Nodal Disease Outcome among by Neck Treatment Groups (cN0, N=447)  Outcome and Timea to Outcome N (%) No Nodal Disease Time to last follow-up Occult RF Time to RF RR Time to RR LE-only, 322 (72) 233 (72) 5.2±3.7  89 (28) 1.7±1.9 8 (9) 1.0±0.8 LE + Prophylactic Neck Treatmentb,125 (28) 98 (78) 5.0±3.7 15 (12) 12 (10) 2.0±2.6 9 (75) 0.9±0.5 aTime, average time ± SD b Prophylactic neck treatment includes elective neck dissection with or without radiotherapy LE, local excision; cN0, clinically Node-negative; LE, local excision; RF, regional failure; RR, regional recurrence cN0, Neck Treatment and Survival In the following sections we will examine the impact on neck treatment on survival. We first determine if  there is survival advantage in delivering prophylactic neck treatment by comparing LE-only group to prophylactic treatment group (END +/- RT), and to END, without RT group (LE+END) (Table 2-9). Second, we investigated if prophylactic treatment has survival advantage over salvage neck treatment by comparing (END +/- RT) group to the 89 patients developed RF and received subsequent salvage neck treatment. This will help us to answer whether performing prophylactic neck dissection would benefit patient survival as opposed to wait-and-treat.     5-year Overall Survival (OS) Figure 2-7 summarizes the OS rates at 2, 3, and 5 years after initial treatment. At 5-years, LE+END (+/-RT) group had slightly lower survival rate (54.4%) compared to LE-only 46  group (61.9%) (P=0.12; Figure 2-7A). Similarly, the END group (61.7%) did not demonstrate better survival compared to LE-only group (P=0.93; Figure 2-7B). On the other hand, compared to patients who received salvage treatment, prophylactic treatment group had significantly higher 5-year OS survival rate (P=0.005; Figure 2-7C).  A. LE-only vs. Prohylactic B. LE-only vs. LE+END C. Prophylactic vs. Salvage     Estimated OS Rates % (95% CI) Treatment Groups Median, years (95% CI) 2 years 3 years 5 years P LE-only NR 79.1 (74.6-83.9) 72.4 (67.4-77.7) 61.9 (56.5-67.8)  LE+END(+/-RT) NR 71.9 (64.2-80.5) 62.0 (53.7-71.5) 54.4 (45.9-64.5) 0.12a LE+END NR 79.5 (71.6-88.1) 69.1 (60.2-79.4) 61.7 (52.3-72.8) 0.93b Salvage 2.0 (1.1-NR) 50.6 (40.9-62.6) 46.6 (37.0-58.8) 40.2 (30.6-52.8) 0.005c Figure 2-7. Kaplan-Meier Survival Curves of 5-Year OS; by Neck Treatment at Surgery (cN0, N=447) aP value is calculated by log-rank test comparing LE-only versus LE with prophylactic (LE+END(+/-RT)) bP value is calculated by log-rank test comparing LE-only versus LE with END (LE+END) cP value is calculated by log-rank test comparing prophylactic (LE+END(+/-RT)) versus salvage LE-only pN+ LE, local excision; END, elective neck dissection; RT, radiotherapy; CI, confidence interval; NR, not reached 	5-year Disease-Specific Survival (DSS) Regarding DSS rates, compared to LE-only group (80.8%), prophylactic neck treatment group had slightly lower DSS rate at 5-years (73.1%) (P=0.11; Figure 2-8A), and END group (80.3%) had almost exact the same proportion of survival as those of LE-only group (P=0.99, Figure 2-8B). Compared to salvage group, prophylactic neck treatment group had significantly higher 5-year DSS rate (P=0.002; Figure 2-8C).  47  A. LE-only vs. Prohylactic B. LE-only vs. LE+END C. Prophylactic vs. Salvage     Estimated DSS Rates % (95% CI) Treatment Groups Median, years (95% CI) 2 years 3 years 5 years P LE-only NR 87.7 (84.0-91.6) 83.0 (78.6-87.6) 80.8 (76.1-85.6)  LE+END(+/-RT) NR 81.8 (74.9-89.4) 76.7 (69.0-85.2) 73.1 (65.0-82.3) 0.11a LE+END NR 88.5 (82.0-95.5) 83.3 (75.6-91.7) 80.3 (72.0-89.4) 0.99b Salvage NR 61.5 (51.4-73.6) 58.2 (47.9-70.7) 54.0 (43.3-67.4) 0.002c Figure 2-8. Kaplan-Meier Survival Curves of 5-year DSS; by Neck Treatment at Surgery (cN0, N=447) aP value is calculated by log-rank test comparing LE-only versus LE with prophylactic (LE+END(+/-RT)) bP value is calculated by log-rank test comparing LE-only versus LE with END (LE+END) cP value is calculated by log-rank test comparing prophylactic (LE+END(+/-RT)) versus salvage LE-only pN+. LE, local excision; END, elective neck dissection; RT, radiotherapy; CI, confidence interval; NR, not reached   48  2.3.3 Regional Failure (cN0, LE-only, N=322)  In this section, we report the incidence of regional failure (RF) observed in cN0 patients who received LE-only treatment without any neck dissection (N=322; Figure 2-1).  A sub-group of these patients developed RF at follow-up (N=89, 28%; hereinafter referred to as the "cN0- -pN+" group) within only a median time of 10.8 months (95% CI, 9.6 months-1.3 years) after LE. Whereas, 233 patients remained N0, (hereinafter referred to as the "cN0- -cN0" group), throughout their follow-up period. The follow-up time for cN0- -cN0 group was long enough (5.1±3.7 years vs. 4.2±3.2 years, P=0.04) to make sure that all possible outcome of RF are captured. Figure 2-9 graphs the 5-year cumulative incidence of RF among 322 patients, by adjusting death-due-to-any-cause, as a competiting risk factor.   Regional Failure Median Time to RF, years (95% CI) Estimated Cumulative Incidence Rate, % (95% CI) 2-year 3-year 5-year cN0 LE-only (322) NR 22.9 (18.1-27.6) 25.3 (20.4-30.3) 27.9 (23.1-32.6) Figure 2-9. Cumulative Incidence of Regional Failure (cN0 LE-only, N=322) The red curve represents cumulative incidence over 5-years after adjusting for death as competing risk event (gray dotted line). The estimated rates of RF were 22.9%, 25.3%, and 28.9% at 2, 3, and 5 years after receiving LE-only treatments.  RF, regional failure; CI, cumulative incidece; NR, not reached Proportion of Regional Failure49  Table 2-11 summarizes and compares patient and clinico-pathological characteristics. There were more cN0- -pN+ patients in the age group of 45-65 or >65 (P=0.04), or with high risk site (P=0.03). Gender, smoking history, clinical T stage, or TNM staging at initial presentation were indifferent between the groups. Regarding pathological characteristics, cN0- -cN0 patients had more Grade I than Grade III tumours; whereas more cN0- -pN+ patients had either Grade II or Grade III (P<0.02). When comparing DOI, cN0- -pN+ had thicker tumours (5.7±5.6 vs. 3.7±4.5 mm, P<0.005); and almost half of cN0- -pN+ patients had DOI greater than 4mm (46% vs. 32%, P<0.001).  Table 2-11. Demographics and Clinico-pathological Characteristics (cN0 LE-only, N=322) Variables, N (%) Total(322) cN0- -cN0 (233, 72%) cN0- -pN+ (89, 28%) P Age, years, mean±SD 63.2±14.9 63.4±15.5 62.7±13.1 0.70 Age Group    0.04 <45 42 (13) 35 (15) 7 (8)  45-65 133 (41) 87 (37) 46 (52)  >65 147 (46) 111 (48) 36 (40)  Sex    0.53 Male 181 (56) 128 (55) 53 (60)  Female 141 (44) 105 (45) 36 (40)  Smoking History    0.86 Never 97 (30) 67 (36) 30 (34)  Ever 177 (55) 119 (64) 58 (65)  Missing 48 (15) 47 (20) 1 (1)  Lesion Site Risk of Cancera    0.03 Low 62 (19) 53 (23) 9 (10)  Intermediate 20 (6) 15 (6) 5 (6)  High 240 (75) 165 (71) 75 (84)  Clinical T Stage    0.34 T1 230 (71) 170 (73) 60 (67)  Not T1 92 (29) 63 (27) 29 (33)  Clinical TNM Staging    0.80 I/II 300 (93) 216 (93) 84 (94)  III/IV 22 (7) 17 (7) 5 (6)  Tumor Gradeb    0.02 I 124 (39) 100 (43) 24 (27)  II 174 (54) 118 (51) 56 (63)  III 24 (7) 15 (6) 9 (10)  DOI, mm±SD 4.3±4.9 3.7±4.5 5.7±5.6 0.005 DOI of 4 mm    <0.001 < 4 mm 177 (55) 139 (68) 38 (38)  ≥4 mm 112 (35) 66 (32) 46 (46)  Missing 33 (10) 28 (12) 5 (6)  Oral Cancer Death 62 (19) 24 (10) 38 (43) <0.001 Time to Death or Last Follow-up, years, mean±SD 4.9±3.5 5.1±3.7 4.2±3.2 0.04 aLesion Site Risk: Low, hard palate or buccal mucosa or gingiva; Intermediate, soft palate or soft palate complex; High, tongue or floor of mouth  bTumour Grade: I, well differentiated; II, moderately differentiated; III, poorly differentiated cN0, clinically node-negative; pN+, pathologically proven node-positive; DOI, depth of invasion 50  Regional Failure and Impact on Survival (cN0 LE-only, N=322) As we reported in previous section that not only there is a 28% chance of early-stage cN0 patients developing RF (cN0- -pN+), such spreading to the neck nodes happened disturbingly fast, with 50% of the RFs occurred within only 10.8 months after LE and 80% happened in 1.5 years (Figure 2-10).     Regional Failure Median Time, years (95% CI) Estimated Cumulative Incidence Rate, % (95% CI) 2-year 3-year 5-year cN0- -pN+ (89) 0.9 (0.8-1.3) 76.4 (65.7-83.8) 84.3 (74.6-90.3) 92.1 (84.0-96.1) Figure 2-10. Cumulative Incidence of Regional Failure of cN0- -pN+ (N=89) The solid curve represents cumulative incidence of RF over 5-years; dashed-line represents 95% confidence interval at correpsonding time points. The estimated rates of RF were 76.4%, 84.3%, and 92.1% at 2, 3, and 5 years after local excision. cN0- -pN+, clincially Nodal-negative patients who developed regail failure at follow-up; CI, cumulative incidece    51   Upon comparing survivals between cN0- -cN0 and cN0- -pN+ groups, there was again a clear implication that nodal disease decreases both OS (Figure 2-11) and DSS (Figure 2-12) by almost 50%. As indicated in Kaplan-Meier estimates and log-rank tests, 5-year survival rates were significantly lower for cN0- -pN+ patients.   Overall Survival Median Time, years (95% CI) Estimated Survival Rates, % (95% CI)  2-year 3-year 5-year P cN0- -cN0 (233) NR 83.9 (79.1-89.0) 80.9 (75.7-86.4) 70.1 (63.9-76.8) < 0.001 cN0- -pN+ (89) 3.6 (2.33 - NR) 67.9 (58.8-78.5) 52.7 (43.1-64.4) 43.0 (33.7-54.9)  Figure 2-11. Kaplan-Meier Survival Curve and Rates for OS, by Regional Failure (cN0 LE-only, N=322) cN0, clinically node-negative; pN+, pathologically proven node-positive; CI, confidence interval; NR, not reached   52     Disease-specific Survival Median Time, years (95% CI) Estimated Survival Rates, % (95% CI)  2-year 3-year 5-year P cN0- -cN0 (233) NR 92.3 (88.7-96.0) 91.7 (88.0-95.6) 91.0 (87.2-95.1) < 0.001 cN0- -pN+ (89) NR 77.2 (68.6-86.8) 62.5 (52.6-74.2) 56.6 (46.5-69.0)  Figure 2-12. Kaplan-Meier Survival Curve and Rates for DSS, by Regional Failure (cN0, LE-only, N=322) Survival curves are shown for cN0- -cN0 group, patients who remained N0 from initial presentation throughout study period, and for cN0- -pN+, patients who were cN0 at initial presentation but developed nodal disease during follow-up. Both groups did not receive any prophylactic neck treatment. cN0- -cN0 did not receive any radiotherapy throughout study period. Difference in curves was compared by log-rank test.  cN0, clinically node-negative; pN+, pathologically proven node-positive; CI, confidence interval; NR, not reached   53  Predictive Factors on Regional Failure (cN0, LE-only, N=322) Upon examining predictive value of patient and tumour characteristics on RF, univariate logistic regression indicated that smoking, high risk site, tumour grade III, and tumour DOI separately achieved statistical significance (Table 2-12). Of the most significant association, patients who had DOI greater than 4mm are ~2.4 times more likely to experience RF (OR, 2.64; 95% CI, 1.57-4.50, P<0.001) (Table 2-13); and patients with Grade III tumour were 4.6 times more likely (OR 4.6; 95% CI, 1.5-14.1). Similar weights of association were also seen in multivariate analysis (DOI greater than 4mm, OR, 2.20; 95% CI, 1.22-3.99; P=0.009; Grade II, OR 2.3; 95% CI 1.3-4.2; P=0.007; Grade III, OR 4.5; 95% CI, 1.4-15.1; P=0.01) Table 2-12. Univariate and Multivariate Logistic Regression Analysis for Regional Failure (cN0 LE-only, N=322)  Univariate Multivariate  OR (95% CI) P OR (95% CI) P Age Group     <45 1    45-65 2.4 (1.0-6.5) 0.05   >65 2.3 (0.9-6.1) 0.08   Gender     Male 1    Female 0.9 (0.5-1.6) 0.78   Smoking History     Never 1    Ever 1.0 (0.5-1.7) 0.88   Lesion Site Risk of Cancera     Low 1    Intermediate 2.0 (0.4-8.1) 0.34   High 2.9 (1.3-7.5) 0.01   Tumor Size     T1 1    Not T1 1.5 (0.8 - 2.7) 0.17 1.5 (0.8-2.8) 0.19 TNM Staging     I/II 1    III/IV 0.8 (0.2-2.7) 0.67   Tumor Gradeb     I 1    II 2.3 (1.3-4.2) 0.005 2.3 (1.3-4.2) 0.007 III 4.6 (1.5-14.1) 0.006 4.5 (1.4-15.1) 0.01 DOI of 4mm     < 4mm 1    ≥ 4 mm 2.4 (1.4-4.2) <0.001 2.0 (1.2-3.6) 0.01 aLesion Site Risk: Low, Buccal Mucosa, Gingiva, or Hard Palate; Intermediate, Soft Palate or Soft Palate Complex; High, Tongue or Floor of Mouth b Tumour Grade: I, well-differentiated; II, moderately-differentiated; III, poorly-differentiated DOI, depth of invasion; OR, odds ratio; CI, confidence interval 54   Given with the strong association between tumour DOI and RF, we investigated its discriminative ability by ROC analysis. From reported DOI, AUC was 64.5% with sensitivity of 71.8% and specificity of 57.3%.  The cut-off point at maximum sensitivity and specificity was 1.8 mm with 45.5% sensitivity and 79.3% specificity (Figure 2-13). We also performed sensitivity test on variables that showed strong association to RF in logistic regression analysis (Table 2-13). DOI of 4mm only had sensitivity of 54.8% and specificity of 67.8% (PPV, 41.1%; NPV, 78.5%). Both tumour grade and lesion site had higher sensitivity, but lower specificity, than DOI of 4 mm.   Figure 2-13. Receiver Operating Characteristic (ROC) Curve for Tumour DOI for the Diagnosis of Nodal Status ROC was performed for tumour DOI (N=322; n=89 patients developed RF and n=233 patients remained N0). Area under the curve (AUC) for DOI was 0.63 (63%; 95% CI, 56%-71%). The red cross indicates the best threshold value of DOI (2.4 mm) to achieve highest sensitivity (53.5%) and specificity (69.1%).    55  Table 2-133. Sensitivity and Specificity of Tumour Attributes for RF Variables High risk defined as Sensitivity Specificity PPV NPV DOI (mm) 4.0 mm 54.8 67.8 41.1 78.5 Tumor Gradea II or III 73.0 42.9 32.8 80.6 Site Risk Highb 84.3 29.2 31.3 82.9 aTumour Grade: II, moderately-differentiated; III, poorly-differentiated bHigh risk site, Tongue or Floor of Mouth DOI, depth of invasion; PPV, positive predictive value; NPV, negative predictive value.    We further examined the proportion of [cN0- -cN0] and [cN0- -pN+] patients who would be over-treated or under-treated at their corresponding DOIs (Figure 2-14). We noticed that if END was performed on all patients who had tumour ≥4mm DOI tumour, only 45% would have occult metastatic neck disease treated, and 25% would have occult neck disease dismissed, as the primary tumour had a DOI of <4mm.   Figure 2-14. Over-treating or Under-treating cN0 Patients using Depth of Invasion Shaded bar: Number of cN0- -pN+ patients at corresponding DOI. White bar: Number of cN0- cN0 patients at corresponding DOI. Open circles: Percentage of over-treatment: total number of cN0- -cN0 at and above the corresponding cut-off divided by the total number of treated patients at and above the corresponding DOI; Closed circles: Percentage of under-treatment: total number of cN0- - pN+ below the corresponding cut-off divided by the total number of patients below of the corresponding DOI.  56  2.4 Discussion The present study is the largest population-based cohort of oral cancer patients in the literature with comprehensive clinico-pathological information, treatment types, and long-term follow-up.  To our knowledge, this is the largest cohort of cN0 patients investigating the efficacy of prophylactic neck treatment, and to examine the natural history of the disease without neck treatment.   2.4.1 Disease Burden In this cohort, nearly half of the patients (~42%) had nodal disease either at the time of initial presentation (N=238), at time of END (N=15), or within 5-year follow-up (N=122). This echoes previously published data from other groups. Presence of nodal disease at any stage of OSCC is an independent risk factor for OS and DSS. TNM staging retains its prognostic value for survival of OSCC. Old age (>65 years old) is also an independent prognostic factor on OS, but not on DSS.  Our patient cohort showed high incidence of LR, RF, RR, or the development of SC or DM. The result on the incidence of SC supported the SEER (Surveillance, Epidemiology, and End Results) data in that oral cavity cancer has a high incidence of developing SC.8 Moreover, we observed a small portion of (3%) of patients developed DM during follow-up. These results highlighted the importance of rigorous follow-up after diagnosis of OSCC, and clinicians must be aware of the risk of not only relapse to local-regional regions, but also other organ sites.  2.4.2 Effectiveness of Neck Treatment Contrary to expectation we observed that deliverance of preventative neck treatment at surgery did not seem to make a difference in survival. However, compared to cN0 patients who actually developed regional metastasis and received neck treatment (the "Salvage" treatment group), prophylactic neck treatment group showed significant survival benefit. However, this is ignorant at the potential harm and quality of life impact using prophylactic neck treatment. Identifying nodal disease risk in this particular 'low-risk' group has become the mandate for clinicians and researchers, and is what we explored further in Chapter 3. 57  There have been several randomized clinical trials (RCTs) suggesting benefit to elective management of the cN0 neck. A cochrane meta-analysis, including 4 small sample-size RCTs with a pooled total of 283 patients, claimed that END reduced risk of disease-specific death by approximately 40%.The authors admitted that the observed benefits of END for cN0 patients were mostly contributed from relative outdate studies. With continuous advancement in technology improving early detection of clinically occult disease, and increasing refinement salvage treatment, a newer study68  did not observe the benefits in DSS between END and no END groups.  There remains clinical equipoise in this area. The most recently published RCT on this topic enrolled 500 T1 and T2 OSCC patients suggested that END significantly improved OS and disease-free survival (DFS).156 It is noted that the study reported prematurely with 25% of patients having less than 16 months follow-up. The authors chose DFS, defined as the interval between the date of randomization and the first documented evidence of any disease or death from any cause as the endpoint for the trial. This is clinically less meaningful than DSS and lead to false conclusions. For those in the control group that developed nodal disease at follow-up, DFS does not capture the impact on survival since the first time of nodal disease will be counted as an event. For those in the END group, an exceptionally high 24% incidence of clinically occult disease at the time of END did not be counted as an event till re-occurrence of nodal disease. Therefore the statistically significant changes in DFS between the therapeutic and END groups may not reflect significant changes in OS. Alternatively, the different results from our study may due to other confounding factors such as geographical difference of the studied populations or the sensitivity of preoperative assessment of clinically occult nodal disease with advanced imaging.  Although not a randomized trial, we included a population-based cohort of 583 patients with surgery as the primary intent-to-cure regime, accounted for all T1 to T4 tumors and all anatomical sites of oral cavity, and had a longer follow-up time.  Thus, the results are more likely to represent “real-life” situations and may be more generalizable.  58  2.4.3 To Dissect or Not to Dissect Necks Nodal disease, regardless the timing of its diagnosis or treatment, is a significant independent prognostic factor on survival. There was almost a 50% reduction in survival among low-risk early-staged tumours (i.e., cN0, LE-only). However, just over a quarter of cN0s (28%) developed nodal diseases during follow-up. Thus, based on this cohort, prophylactic neck treatment would result in over treating 72% of cN0 patients. Unnecessary treatment may involve significant cost and morbidity. Hence, we really need to be able to identify patients who are at risk of RF without performing END on all patients.   Although our data concurs with the notion that higher tumor grade associates with RF, its low specificity, the intrinsic inter-observer variability, and sampling errors make it an unreliable predictor. Preoperative means to measure a true DOI are only approximations at best, and surgical samples are formalin-fixed and in most cases are only available after the initial surgery. Therefore, DOI, like tumor grade, cannot be used as a “real-time” clinical decision making aid for pre- or intra-operative planning. Without level 1 evidence on the utility of DOI, multiple centers continue to use a pre-specified threshold of DOI to justify a prophylactic neck treatment. In this scenario, a second surgery with or without RT will result hospitalization, associated costs, patient morbidity, and patient convalescence.  We have observed some smaller tumors will develop RF yet some thicker tumors will not. This evidently shifts the idea that smaller tumors have less risk of developing nodal disease. Furthermore, our data have shown near plateau numbers of under- and over-treatment at various DOI cut-offs.  There is an urgent need for objective real-time biomarkers to predict nodal disease for treatment planning prior to the initial surgery. Cancer metastasis to regional lymph node has been recognized as one of the ‘hallmarks’ of carcinogenesis and involves a complicated, multi-step processes. We found that patients who developed RF at follow-up showed decreased survival irrespective of neck treatment.  Moreover, a subset of patients was refractory to prophylactic neck treatment as evidenced by 12 patients who developed pN+ during follow-up despite having received neck treatment. This observation combined with unreliability of DOI suggests that there are subgroups within OSCC which are biologically and behaviorally different.  59  Eighty percent of cN0 patients in this study developed RF within 30 months after initial LE with a short median time of 10.8 months. Additionally, for cN0s who had END, 12% of cN0 patients had occult nodal disease. The results also highlighted the importance in closely monitoring these patient post-surgery at least for the first 2.5 years.    2.4.4 Study Limitations This study is limited by several factors and all contributed to that it was retrospective in nature. Firstly, all collected data relies on the accuracy of written records that are almost impossible to authenticate. This is especially the case concerning pathology reports. Interpretation of tumour pathology and diagnosis were often dictated by one pathologist. Rarely were there cases with consultation for second opinion. As is to be expected, there is inherent error from intra-observer variability. Secondly, there was no control or standardized protocol for pre-operative clinical staging for these patients. Therefore, it is possible that presence of nodal disease was missed. Thirdly, even after exhaustive chart review, there was random missing data because they were not recorded. Examples of missing data were smoking, tumour grade, or tumour depth of invasion. However, after adjusting for missing data by excluding patients with missing values, the conclusions that we drew from comparing sub-groups of patients did not change. Lastly, there is bias in comparing survival between neck treatment groups decision for prophylactic neck dissection was not a randomized decision; and patients who received prophylactic neck treatment were in more advanced stage of disease.  With these limitations, we are unable to establish a cause-and-effect relationship between neck dissection and survival, or between tumour depth of invasion and regional failure. However, based on that this was a population-based study that covered over 7 years, the perspective of ‘real-life’ situation (i.e. no control) is well presented.  2.4.5 Next Steps Taken from what we learned and collected from this retrospective review, there is an urgent need to identify a more objective risk assessment tool for nodal disease. From the established prospective COOLS surgical trial, which recruits 400 early-stage OSCC patients who are treated with curative surgery, we can ask questions surrounding the risk and benefit of prophylactic neck treatment. For each patient there is comprehensive quality 60  of life and direct or indirect costs survey associated with type of neck treatment received. For these patients, we can apply the findings on disease or survival outcomes and construct a decision model taking into account the quality of life and cost survey collected from the COOLS trial. Most importantly, we can study the primary tumours of these patients and correlate with QTP features in order to better predict who require END.    61          CHAPTER 3 3 NOVEL COMPUTATIONAL IMAGE ANALYSIS TO IDENTIFYTHE RISK OF NODAL METASTASIS IN ORAL CANCERS 62  3.1 Hypothesis and Objectives  From Chapter 2, we have seen that almost a quarter the nodal disease developed within a very short time frame after initial surgical treatment. Thus, we hypothesize that a group of tumors are biologically different which are phenotypically presented. We can study these phenotypes with the application of Quantitative Tissue Pathology (QTP).  The hypothesis is: • QTP features of tumuor nests, from primary tumour of patients with nodal disease (N+), are different from those from patients without any nodal disease (N0). Specific objectives are: i. Identification and collection of specimen of primary tumour from OSCC patients with known nodal status. ii. Acquisition of digital images of stained slides, using Hemotoxylin & Eoisin (H&E) and Feulgen-Thionin staining.   iii. Identification of tumour nests and extraction QTP features.  iv. Analysis of QTP features by comparing between N+ and N0 status 3.2 Methodologies 3.2.1 Patient and Surgical Specimen  A total of 15 OSCC cases were identified from a surgical trial. In order to ensure the nodal status, especially for N0 group, cases assigned with N0 status had at least 5 years of post-surgical follow-up. These patients and their samples were thoroughly annotated. Demographics, clinical-pathological information, and nodal disease outcome were obtained from the trial database. Surgical tissue specimens were formalin-fixed and paraffin-embedded, and blocks containing tumours were selected, serial sectioned (4µm in thickness) and mounted on charged slides. For each case, one section was stained with H&E and the other with Feulgen-Thionin.  We summarized the workflow of QTP analysis in Figure 3-1. Each step is described in the following Sections 3.2.2 to 3.2.4.  63   Figure 3-1. QTP Analysis Workflow Schematic 3.2.2 Feulgen-Thionin Staining  Thionin stain (0.5%, pH 1.4-1.7) was formulated with 0.5g of Thionin acetate powder (Sigma-Adrich® Co. LLC, Germany) in boiled (5 min) and cooled deonized water, 1N hydrochloric acid, ter-butanol, and sodium bisulphate. Mixed solution were rested overnight and filtered prior to use.   Staining procedure was performed at room temperature and as follows. Slides were first fixed in Bohm-Springer fixative (methanol, formaldehyde, and acetic acid, 16:3:1 ratio; 45 minutes), rinsed (distilled water, 15 seconds), hydrolyzed in 5N hydrochloric acid (one hour), rinsed (distilled water, 15 seconds), immersed in Thionin stain (one hour), and rinsed (distilled water until clear of stain). Subsequently, slides were rinsed thrice in 0.5% sodium bisulphate solution (pH 7) in distilled water and hydrochloric acid, each time separated with water rinses. Finally, slides were dehydrated by immersion in 3 separated baths of 100% Serial sectioning of tumour sampleHE stainingFeulgeun-ThioninstainingTumour nestsdemarcation Image acquisitionThionin image segmentation and processingTissue ArchitectureMean (52); Variance (52)QTP feature extraction and outputTumour nest ROILayered nest ROIAnalysisVoronoi Diagram (4)Delaunay Triangulation (4)Nuclear PhenotypesMorphology(10)Optical Density (6)Chromatin Texture (36) Discrete Texture (20)MarkovianTexture (5)Fractal Texture (1)Run Length(10)Mean (8); Variance (8)Quantitative Tissue Pathology Workflow64  ethanol (15 seconds each time), soaked in xylene (30 minutes), mounted with Cyto-Seal (Thermo ScientificTM, MA, USA), and coverslipped.   3.2.3 Image Analysis Image Segmentation and Processing  Stained slides were imaged using Cyto-SavantTM imaging system under bright-field microscopy with charge-coupled digital camera, which had resolution of 1280 by 1024. The effective pixel was 0.116µm2. Images were captured with an objective of 40 and 6 planes of focus. Stained objects were focused in-frame and illuminated at wavelength of 600±5 nm. Digitized images of tissue sections were subsequently analyzed with segmentation software (Unit Display Program©, Integrative Oncology, BCCA) (Figure 3-2). OD for segmented object was then obtained. Because the distribution of DNA in a nucleus is heterogeneous, we first calculated the OD for each pixel followed by summation of all ODs to give integrated OD corresponding to the DNA content of the entire tissue section.    Figure 3-2. Example of a grey-level image of a nucleus with its corresponding mask Nuclei (left) with a Mask Object (right).  Each segmented image of an object is called a 'seed' and it represents a cell nucleus in a tumour nest. After segmentation, all seeds are manually evaluated visually and filtered for two reasons: exclude those that were not squamous epithelial cells, such as granulocytes, lymphocytes or part of keratinized pearl (Figure 3-3, top panel); and exclude ‘bad’ seeds that were overlapping, fused together, incorrectly segmented, or out-of-focus (Figure 3-3, bottom panel). The remaining objects after filtration process should represent only cancerous squamous cells.    65  A. “Bad” objects       Touching Overlapping Out of focus     B. “Good” objects, but excluded cell types        Squamous Cell Lymphocytes Fibroblast Abnormal Mitotic figure Mitotic (Telephase) Mitotic (Cytokinesis) Figure 3-3. Examples of Excluded Cell Types and ’Bad’ Nuclei after Segmentation. The bottom panel gives examples of situations where semi-automatic segmentation for cell nucleus would be unsatisfactory; the top panel gives examples of possible cell types exist in an image that are not included in the analysis.  3.2.4 QTP Features Extraction - Nuclear Phenotypes  Nuclear phenotypes collectively describe the structure of a seed and the spatial relationship among sub-cellular components within a seed. There are three general categories describing nuclear phenotypes, namely, 1) Morphology, 2) Optical Density, and 3) Chromatin Texture. Each category and associated features are described as follows. (A) Nuclear Phenotype - Morphology  Morphology category includes the size, shape, and boundary variations of a nucleus. The feature Area is the total pixel area of a seed; the Mean_radius and Max_radius measure the mean and maximum, respectively, lengths from the centre of a seed to the boundary; var_radius describes the variation of radius; sphericity measures the roundness of a seed, with 1 being a perfect circle; eccentricity measures the ratio of major axis to minor axis of the best-fit ellipse (i.e. how a seed deviates from a circle, with 0 being a perfect circle). The Morph_fft features measures freq_low_fft (frequency of low boundary harmonics), or the low harmonic variations in the boundary of the nucleus; and freq_high_fft (frequency of high boundary harmonics), the high harmonic variations in the boundary of the nucleus.  66  (B) Nuclear Phenotype - Optical Density (OD) DNA content distribution is measured as the total “shades” of the stain at any pixel of an image and is proportional to the OD over the entire seed.  The Photometric category consists features describe the OD of seeds. They include Area (total pixel area), OD_variance, OD_skewness, and OD_kurtosis.  (C) Nuclear Phenotype - Chromatin Texture  Cancer cells exhibit aberrant chromatin remodeling and alterations. These changes can be indicative of the etiology of cancer and are seen as consequences of genetic and epigenetic changes. There are four subcategories describing chromatin texture. 1). The Fractal dimension (FD) lands itself as one of the most extensively studied phenotypes in computational image analysis for cancer progression.157-160 FD infers to the amount of details and complexity of the geometrical structure of each pixel within a seed. The level of details you can measure depends on the scale used for the measurement (Figure 3-4), and fractal dimension changes based on the slopes of change as measurements increase or decrease when the scale of measurement becomes larger or smaller.  Figure 3-4. Fractal Dimension Measurement Fractal dimension of an object can be calculated from the sum of the areas of all ‘exposed’ surfaces of the bars (vertical walls and horizontal tops), where each bar represents the optical density of each pixel.161  67   2). The Run Length Texture features measure the frequency of chromatin distribution in terms of the lengths of lines comprised of consecutive pixels with the same gray level value. Short_Run features describe the frequency of small chromatin clumps in the nucleus, whereas Long_Run features describe that of larger chromatin clumps. 3). Markovian system describes the similarity and variation gray-level intensity between a pixel and its neighbours. Markovian measurements include Entropy/Energy, Contrast/Homogeneity, and Correlation.  Briefly, Entropy measures the disorder of transition in gray-level where large values mean disorganization and large variation in intensity. The negatively correlated feature is Energy, of which large values correspond to uniform intensity. The Contrast feature measures the differences in intensity between neighboring pixels; thus, the larger the contrast, the higher magnitude of density variation between chromatin components. Conversely, homogeneity depicts the smoothness of image intensity, where large value means same pair of pixels is frequently found. Lastly, the Correlation feature measures the size and intensity of chromatin clumps. A large value Correlation means a nucleus with large regions of condensed chromatin which have uniform intensity and large gray-level differences between adjacent chromatin components.  4). The Discrete Texture category is based on segmenting a nucleus into discrete regions of low, medium, and high chromatin condensation (Figure 3-5). Each of these regions is defined for each slide by setting two global thresholds based on the OD profile of the entire nucleus against some internal diploid control cells. The high to medium density borderline would fall under the OD profile of a picnotic nucleus such as that of leukocytes or lymphocytes; and the low to medium density borderline would fall two standard deviations below the average of the picnotic nuclei.  68   Figure 3-5.  Chromatin Discrete Regions in a Cell Nuclei  Discrete texture measurements include the Area (area of pixels), Amount (IOD of each region), Fraction (proportion of each region against the others), Compactness, and Average Distance from the nucleus of each discrete region. The Compactness measures the circularity of each chromatin state, where 1 compactness equals a perfect circle, and >1 compactness leans towards irregular shape. The distance from the nucleus for each discrete region is measured by the Average Distance between the centre of the nucleus and all pixels that make up each region (...av_dist). If all pixels of a condensation state were located at the nuclear border, the av_dst would be maximum value of 1.   Highly condensed chromatinMedium density chromatinLow density chromatin69  Morphology (size and shape)         small / large regular Irregular    Optical Density       Low High   Chromatin distribution    Clumps scattered Connected along the peripheral   Chromatin Texture    Smooth Coarse Figure 3-6. Examples of Nuclear Feature    70  3.2.5 QTP Features Extraction - Tissue Architecture As described in Section 1.3.2, tissue architecture features measure the organization of a population of cells, either locally or globally. There were 16 of such features and they are listed in Appendix E. Error! Reference source not found. illustrates cartooned examples of these features.  H&E image of tumour nests A  Tumour Nest A B  Tumour Nest B Voronoi Diagram  C  D  Delaunay Triangulation E  F  Figure 3-7. Examples of Tissue Architecture Features of Tumour Nests The H&E images of tumou rnests A and B are shown in the top panel. The middle (C and D) and bottom panel (E and F) shows the tissue architecutre. Fig. 3-7 C and D are the Voronoid Diagram of nest A and B, repsecitvely. Nest A (Fig. 3-7 C) has larger Vornoi polygon area, larger perimeter, and smaller desnely; whereas Nest B (Fig. 3.7 D) has smaller Voronoi polygon ara, smaller perimeter, and higher density. Fig. 3-7 E and F shows the Delaunay Triangulation of tumour nest A and B, respectively. Nest A (Fig. 3.-7 E) has smaller number of neighbours or nearest neighbours and larger distance between neighbours; whereas Nest B (Fig. 3-7F) has more niehgbours and small distance between nieghbours.  71  3.2.6 Region of Interest (ROI) for QTP Analysis Tumour Nests as ROIs  We defined a tumour nest as a group of cells clustering together that falls out from the epithelium and invades into the tissue. We studied tumour nests that had well-defined boundary.  Delineation of tumour nests started off by first identifying them on H&E stained section where the study pathologist clearly demarcated and drew the exact boundaries around each nest on scanned H&E images. These marked areas (i.e. tumour nests) are labeled as region of interests (ROIs). After identifying all the ROIs on H&E slides, the pathologist matched these ROIs on the Thionin stained slides. Features described in Section 3.2.4 and listed in Appendix E were extracted from each seed within a tumor nest. Using tumour nests as ROIs, we compared each feature between N+ and N0 nests.  Tumour Nest Layers as ROIs In addition to tumour nests, we also conducted layer-based analyses (Figure 3-8). Our intention in studying these layers was motivated by the increasing evidence that tumour cells constantly interact with surrounding stromal cells and extracelluar matrix – the tumour microenvironment - in promoting the initiation, tumour growth, and tumour spread. The Voronoi diagram allowed us to separate each nest into layers with polygons at the edge of each nest border assigned as Layer#1 (outer most layer); Layer#2 consisted polygons that are not in Layer#1 and had a neighbour in Layer#1; Layer#3 consisted of polygons that were not in Layer#1 or Layer#2 and had a neighbour in Layer#2; and so forth until there are no polygons to be accounted for. Each layer is thus a distinct ROI from which we can extract and analyze the morphology of nuclei within this a layer as well as the architecture of each layer using features described in Section 3.2.4. In reality, tumour nests are 3-dimensional structures and there are confounding factors from how tissues are tangentially sectioned. To address this issue, we also calculated features using combination of layers, e.g., layer#1-2 and layer#2-3 included them in the analysis. 72  A  B  Figure 3-8. Layered Tumour Nest Fig. 3-8A shows Voronoi diagram divides a tumour nest into layers (e.g. 11 layers for this nest) with polygons touching the outer nest boundary (dashed line) as Layer #1 (blue); Layer #2 (green) are polygons of immediate neighbors inner to Layer #1; and Layer #3 (red) are polygons of immediate neighbors inner of Layer #2. Successive layers are incrementally calculated. Fig.3-8B is the simplified cartoon of Fig.3-8A.  3.2.7 Statistical Analysis Data collection and statistical analysis were performed using R packages (R V3.2.3). Each QTP feature was described regarding their distribution with mean and variance which are graphically presented with box-whisker plots. Each feature was first compared separately between groups of nodal status, adjusted for patients’ effect by using nested analysis of variance (ANOVA). For evaluate the ability of each feature in separating ROIs into N0 or N+, we first determine the optimal feature(s) using linear discrimination model with forward stepwise selection procedure. The selected features were the subjected for classification performance analysis by ROC curve, with AUC represents the probability that an N+ ROI will be correctly labelled as N+ than a randomly selected N0 ROI.   3.3 Results 3.3.1 Patient Set for QTP Analysis The pilot study set included 5 OSCC patients (8 N0 vs. 7 N+) (Table 3-1). The average age was 59 years old, with more females (10, 68%), and more non-smokers (9, 60%). Majority of the patients had tongue tumour (14, 93%) and most were Grade II (11, 73%). The average DOI was 6.1±3.6 mm with 87% of the tumours greater than 4 mm. Layer #1Layer #2Layer #3Nest Boundaryouterinner73  There was no difference in these characteristics or follow-up time between N0 and N+ patients.  Table 3-1. Demographics and Clinico-pathological Characteristics of QTP Set (N=15) Variables Total N=15 N0, N=8 (53%) N+ N=7 (47%) P Age, years, mean ± SD 58.7±11.9 63.1 53.3 0.15 Age Group    0.26 <45 2 (13)  2 (29)  45-65 10 (67) 6 (75) 4 (57)  >65 3 (20) 2 (25) 1 (14)  Gender    1.0 Male 5 (33) 3 2  Female 10 (67) 5 5  Smoking    0.75 Never 9 (60) 4 (50) 5 (71)  Ever 6 (40) 4 (50) 2 (29)  Lesion Site Riska    1.0 Low 1 (7) 1 (12)   Intermediate     High 14 (93) 7 (88) 7 (100)  Tumour Grade (differentiation)    0.13 I (well) 3 (20)  2 (29)  II (moderately) 11 (73) 6 (75) 4 (57)  III (poorly) 1 (7) 2 (25) 1 (14)  DOI, mm, mean±SD 6.1±3.6 6.4±4.5 6.7±3.3 0.89 DOI of 4 mm    1.0 < 4mm 2 (13) 1 (13) 1 (14)  ≥ 4mm 13 (87) 7 (87) 6 (86)  Survival Status    1.0 Alive 11 (73) 6 (75) 5 (71)  Dead due to any cause 1 (7)  1 (14)  Dead due to oral cancer 3 (20) 2 (25) 1 (14)  Years of Follow-up Median (25-75 percentile) 4.1 (2.8-5.0) 4.1 (3.3-4.7) 4.5 (2.3-5.0) 0.98 aLesion Site Risk: Low, buccal mucosa, hard palate, or gingival; Intermediate, soft palate or soft palate complex; High, tongue or floor of mouth DOI, depth of invasion 3.3.2 Tumour Nests, Tumour Nest Layers, and Cell Nuclei for N0 and N+ Cases Among the 15 cases, a total of 45 tumour nests were identified, with 23 (51%) from N0 and 22 (49%) from N+. These tumour nests composed of 410 nest layers (N0, 199; N+ 211) and 45,253 segmented seeds (hereinafter referred to “cell nuclei”) for analysis. Table 3-2 summarizes the distribution of nests, nest layers, and cell nuclei by nodal status to demonstrate that they were representative for nodal group. There was no difference in the 74  average number and pixel area of tumour nest per patient, or the average number layers per nest, or the average cell nuclei per nest between N0 and N+ groups (Figure 3-9). Table 3-2. ROIs and Segmented Cell Nuclei  Total, 45 (%) N0, 23 (51) N+, 22 (49) P Tumour nest per patient 3 2.9 3.1  Tumour nest area  (pixel area, µm2) 191,944.6 ± 256,088 238,211.3 ± 312,791.3 143,574.9 ± 173,536.8 0.22 Layers 410 199 211 0.54 Layers per nest  9.6 10.6  Cell nuclei 45,253 22,996 22,257 0.97 Cell nuclei per Nest  999.8 1011.7   A.  Tumour Nest Area  B. Layers per Nest  C. Cell Nuclei per Nest      Figure 3-9. Box-and-whisker Plots of Distribution of Tumour Nest Area, Layers per Nest, and Cell Nuclei per Nest for N0 and N+ Cases Box-Whisker plots are shown for A) tumour nest area, B) number of layers per tumour nest, and C) number of segmented cell nuclei per tumour nest. The horizontal line within the box indicates the median, boundaries of the box (bottom and top) indicate the 25th and 75th percentile, and the whiskers indicate the highest and lowest values of the results, excluding outliers (open circles)  3.3.3 QTP Features Overview A total of 120 features (nuclear phenotype, 104; tissue architecture, 16) were quantified from 45,253 cell nuclei. Table 3-3 summarizes the total number of features measured (each including its mean and variance)..    75  Table 3-3. QTP Features Measured Category Number of Features measured Nuclear Phenotype (104)  Morphology 10 x 2 (mean and variance) Optical Density 6 x 2 (mean and variance) Chromatin Texture  Discrete 20 x 2 (mean and variance) Markovian 5 x 2 (mean and variance) Fractal 1 x 2 (mean and variance) Run_length 10 x 2 (mean and variance) Tissue Architecture (16)  Voronoi Diagram 4 x 2 (mean and variance) Delaunay Triangulation 4 x 2 (mean and variance)  Although many of these features describe similar aspects of either nuclear phenotype or tissue architecture, comparison was conducted independently for each individual feature, i.e. we performed 120 times of unpaired ANOVA analysis independently of other features. Thus, we did not adjust p-values with correction methods, e.g. Bonferroni correction. In addition, comparison analysis was adjusted for multiple nests per patient to avoid over estimation of association. Distribution of each feature in groups N0 or N+ are graphically presented in Box-plots and variance in mean was reported with p-values.   We used tumour nests (Section 3.2.6) and tumour nest layers (Section 3.2.6) as experimental units for comparison between N0 and N+ groups.  3.3.4 Tumour Nests - Association of QTP with Nodal Status (N0 vs. N+) There were 14 features significantly different (P<0.05) between N0 and N+ groups, as summarized in Table 3-4 with features first sorted by categories  and then by ascending order of uncorrected p-values. Distributions of each of these 14 features were presented using box-whisker plots (Figure 3-11). We also plotted the mean and standard error to visualize the differences of these features (Appendix F).  Nuclear phenotype feature:  only chromatin discrete texture features showed statistical significant difference between N0 and N+ tumor nests (Figure 3-10). We observed higher fraction of high- or medium condensation region of chromatin (P<0.05) in N+ group.  76  Tissue architecture: N+ tumour nests was associated 1) with smaller tumor nest area and circumference (VorAreaMean, VorPeriMean; P< 0.05), 2) with each cell nucleus having a shorter distance to its neighbours (DelNeaNeiMean, Del3NeiDistMean, P< 0.05), and 3) more densely packed cell nuclei (Density, P=0.01) (Figure 3-11). These characteristics appear to be common among all N+ tumour nests, as there was less variation was observed among them (Del3NeiDistStdv, VorAreaStdv, and DelNeaNeiStdv; P<0.01).  Table 3-4. QTP Features of Significant Difference Between N0 and N+ Tumour Nests  QTP Category QTP Sub-category Feature Name P Nuclear Phenotype (6)  Chromatin Texture Discrete Texture Hi_av_dstMean 0.04 *  Chromatin Texture Discrete Texture High_den_objMean 0.01 **  Chromatin Texture Discrete Texture High_den_objStdv 0.04 *  Chromatin Texture Discrete Texture Lowvshigh_dnaMean 0.02 *  Chromatin Texture Discrete Texture Lowvsmed_dnaMean 0.003 **  Chromatin Texture Discrete Texture Lowvsmh_dnaStdv 0.04 * Tissue Architecture (8)  Delaunay Trianulation  Del3NeiDistMean 0.04 *  Delaunay Trianulation  Del3NeiDistStdv 0.007**  Delaunay Trianulation  DelNeaNeiMean 0.03 *  Delaunay Trianulation  DelNeaNeiStdv 0.008 **  Voronoi Diagram  Density 0.01 *  Voronoi Diagram  VorAreaMean 0.02 *  Voronoi Diagram  VorAreaStdv 0.007 **  Voronoi Diagram  VorPeriStdv 0.008 **  77        Figure 3-10. Box-and-whisker Plots of Distributions of Nuclear Phenotypes between N0 (gray) and N+ (red) Tumour Nests The horizontal line within the box indicates the median, boundaries of the box (bottom and top) indicate the 25th and 75th percentile, and the whiskers indicate the highest and lowest values of the results, excluding outliers (closed circles) Top row, left to right: Hi_av_dstMean (P=0.04), High_den_objMean (P=0.01), High_den_objStdv (P=0.04) Bottom row, left to right: Lowvshgh_dnaMean (P=0.02), Lowvsmed_dnaMean (P=0.003), Lowvsmh_dnaStdv(P=0.04) 78            Figure 3-11. Box-and-whisker Plots of Distributions of Tissue Architecture Between N0 (gray) and N+ (red) Tumour Nests The horizontal line within the box indicates the median, boundaries of the box (bottom and top) indicate the 25th and 75th percentile, and the whiskers indicate the highest and lowest values of the results, excluding outliers (closed circles) Top row, left to right: Del3NeiDistMean (P=0.04), Del3NeiDstStdv (P=0.007), DelNeaNeiMean (P=0.03), DelNeaNeiStdv (P=0.008) Bottom row, left to right: Density (P=0.01), VorAreaMean (P=0.02), VorAreaStdv (P=0.007), VorPeriStdv (P=0.008) 79  3.3.5 Tumour Nest Layers - Association of QTP with Nodal Status (N0 vs. N+) The 45 tumour nests were made up of 410 layers. Table 3-5 shows the number each nth layer in N0 or N+ group.  Table 3-5. Number of nth Layer in 45 Tumour Nests nth layer Total N0 N+ 1 45 23 22 2 45 23 22 3 45 23 22 4 44 22 22 5 44 22 22 6 37 18 19 7 31 15 16 8 20 9 11 9 17 7 10 10 16 6 10 11 12 5 7 12 9 4 5 13 8 4 4 14 5 2 3 15 4 1 3 16 3 1 2 17 3 1 2 18 3 1 2 19 2 1 1 20 2 1 1 21 2 1 1 22 2 1 1 23 2 1 1 24 2 1 1 25 2 1 1 26 1 1 0 27 1 1 0 28 1 1 0 29 1 1 0 30 1 1 0 Table 3-6 summarizes the features that were found to be significantly different for each of the first three layers between N0 and N+ groups Nuclear feature:  features describe the low, medium, and high DNA condensation state separated well each layer of N0 from N+ groups (Figure 3-12 A-E). For example, N0 80  layers exhibited higher ratio of IOD between medium to low condensation state (lowvsmed_dnaMean), meaning that there were higher fractions of medium-density DNA as opposed to low-density DNA. Furthermore, there were more high-density DNA components dispersed in N+ layers as opposed to that of N0 layers (High_den_objMean and High_den_objStdv). Tissue architecture: we observed a trend of increasing cell area (VorAreaMean) going from outer (Layer#1) to inner layers (Layer#3) in N0 group (Figure 3-12F). For N+ group, cell area was consistently smaller across the three layers (VorAreaMean, P<0.05).  Table 3-6. QTP Features Different Between N0 and N+, Layer#1, or Layer#2, or Layer#3  Category Feature Feature Name Layer#1 Layer#2 Layer#3 Nuclear Phenotypes  Discrete Hidnacomp HidnacompMean 0.01    Discrete High_den_obj High_den_objMean 0.004    Discrete High_den_obj High_den_objStdv 0.001    Discrete Low_av_dst Low_av_dstMean  0.03   Discrete Lowvsmed_dna Lowvsmed_dnaMean 0.005    Discrete Meddnacomp MeddnacompMean  0.03   Discrete Mhdnacomp MhdnacompMean  0.04   Fractal Fractal_dimen Fractal_dimenMean  0.005   Fractal Fractal_dimen Fractal_dimenStdv 0.001    Morph_fft Freq_high_fft Freq_high_fftStdv 0.04    Photometric Od_kurtosis Od_kurtosisStdv   0.001  Photometric Od_skewness Od_skewnessMean  0.04 0.01  Photometric Od_variance Od_varianceMean  0.03 0.02  Run_length Gray_level1 Gray_level1Mean 0.04    Run_length Long_runs2 Long_runs2Mean  0.03   Run_length Long_runs2 Long_runs2Stdv  0.004   Run_length Run_length1 Run_length1Mean 0.002 0.009   Run_length Run_length2 Run_length2Mean 0.02   Tissue Architecture   VorArea VorAreaStdv   0.01  81  A. Nuclear Phenotypes, Chromatin Texture (Discrete Texture)         Top row, left to right: HighdnacompMean, High_den_objMean, High_den_objStdv, Low_av_dstMean Bottom row, left to right: Lowvsmed_dnaMean, MeddnacompMean, MhdnacompMean 82  B. Nuclear Phenotypes, Chromatin Texture (Fractal Texture)     Left to right: Fractal_dimenMean andFractal_dimenStdv C. Nuclear Phenotype, Morphology (Size / Shape)      Freq_high_fftStdv 83  D. Nuclear Phenotype, Optical Density (Photometric)     Left to right: Od_kurtosisStdv, Od_skewnessMean, Od_varianceMean  84  E. Nuclear Phenotype, Chromatin Texture (Run Lengths)         Top row, left to right: Gray_level1Mean, Long_runs2Mean, Long_runs2Stdv, Run_length1Mean Bottom row: Run_length2Mean 85  F. Tissue Architecture, Voronoi Diagram      VorAreaStdv    Figure 3-12. Box-and-whisker Plots of Distributions of QTP Features among Layer#1’s, Layer#2’s and Layer#3’s, by N0 (gray) and N+ (red) From Fig 3-12A to F, chromatin discrete texture (A), chromatin fractal texture (B), morphology (C), optical density (D), chromatin run lengths texture (E), Voronoi polygon area (F) The horizontal line within the box indicates the median, boundaries of the box (bottom and top) indicate the 25th and 75th percentile, and the whiskers indicate the highest and lowest values of the results, excluding outliers (closed circles) 86  Combining Tumour Nest Layers Considering the possibility of tangential section of the tumour nests presented on the 2-dimensional cell nuclei distribution and low number of segmented nuclei per layer in some cases, we decided to combine nest layers to try to overcome these issues. For example, we treated Layer#1 and #2 (Layer#1-2) or Layer#2 and #3 (Layer#2-3) as single layer by grouping   nuclei together. Table 3-7. QTP Features Different between N0 and N+ Layers#1-2 or Layers#2-3  QTP Category QTP Sub-category Feature Name Layer#1-2 P Layer#2-3 P Nuclear Phenotype  Chromatin Texture Discrete Lowvsmed_dnaMean <0.001 0.01   Discrete Lowvshigh_dnaMean 0.03    Discrete High_den_objMean 0.004 0.02   Discrete High_den_objStdv 0.02    Run_length Run_length1Mean 0.008    Run_length Run_length2Mean 0.03  Table 3-7 lists the features that were significantly different in combined layers, Layer#1-2 and Layer#2-3, between N0 and N+ groups with box plots for distribution in Figure 3-12 and Figure 3-14, respectively.  Layer#1-2  (Table 3-7 and Figure 3-12): we observed significantly larger fraction of medium- and high-density DNA condensation states in N+ group, as reflected by the higher averaged fraction of medium-density and high-density DNA content (Lowvsmed_dnaMean, 2.6±0.15; P< 0.001; lowvshigh_dnaMean, 4.0±0.0.8; P=0.03). There was also more high-density DNA regions dispersed in N+ group (High_den_objMean, 2.2±1.0; P=0.003); however, this feature seemed to carry bigger variability among layers in N+ group as we observed larger error bars in High_den_objMean and significantly higher value of High_den_objStdv (1.6±0.7; P=0.02) (Figure 3-13).  Layer#2-3 (Table 3-7 and Figure 3-14): Similar observation was also seen with higher medium-density DNA (Lowvsmed_dnaMean, 2.5±0.2 vs 2.2±0.2; P=0.01) and increase of high-density DNA content (High_den_objMean, 1.9±1.0 vs 1.2±0.6; P=0.02).  87  Recall that the run length features measure the size of chromatin clumps in a nucleus, and the larger the Run Length value corresponds to bigger size of clumps and vice versa. Both the average of Run_length1 and Run_length2 of N+ group were larger than that of N0 group (Run_length1Mean, 151.4±43.0 vs. 134.6±26.0; P =0.008; and Run-_length2Mean, 39.1±8.6 vs. 34.5±7.9; P=0.03).  None of tissue architecture features showed significance when using combined layers.         Figure 3-13. Box-and-whisker Plots of Distributions of QTP Features among Layer#1-2, by N0 (gray) and N+ (red) Top row, left to right: Lowvsmed_dnaMean (P<0.001), Lowvshigh_dnaMean (P=0.03), High_den_objMean (P=0.004); High_den_objStdv (P=0.02) Bottom row, left to right: Run_length1Mean (P=0.008), Run_length1Mean (P=0.03)    88     Figure 3-14. Box-and-whisker Plots of Distributions of QTP features among Layers#2-3 by N0 (gray) and N+ (red) Left to right: Lowvsmed_dnaMean (P=0.01), High_den_objMean (P=0.02)  3.3.6 Discriminative Ability of QTP Features on Nodal Status The observed differences in the features provided initial evidence in supporting QTP analysis for differentiating N0 and N+ groups. Next, we investigated whether these features obtained from tumour nests or tumour nest layers can be used to diagnose N+ status. Linear discrimination model with forward stepwise selection procedure was performed to determine the features that can differentiate nodal status among A) tumour nests, B) tumour nest layers, or C) combined tumour nest layers. We stopped the selection process at two features to avoid overfitting the regression model from too many predictors on a relatively medium-size sample set. After the selection process, classification performance of the fitted coefficients from the model was assessed by ROC curve.  Tumour Nests as Experimental Unit  Stepwise selection of 120 features showed that combination of Lowvsmed_dnaMean + VorPeriStdv exhibited significant ability for classification and discrimination between tumour nests of N0 or N+ status, with an AUC of 85% (Figure 3-14A). The coefficient threshold, from the fitted model of these two features, was 0.41 with maximum sensitivity of 82% and specificity of 78%. Results again suggested that N+ tumours display higher fraction of medium-density chromatin region and smaller perimeter of cell boundary (Figure 3-14B).   89      Two feature selection AUC (%) Model Threshold Sensitivity (%) Specificity (%) Tumour Nests Lowvsmed_dnaMean + VorPeriStdv 85 0.41 82 78 Figure 3-15. Receiver Operating characteristic (ROC) Curve and Scatter Plot for QTP Features for Discrimination of N0 vs. N+, among Tumour Nests Two QTP features, Lowvsmed_dnaMean and VorPeriStdv were selected with highest performance for the discrimination of nodal status among tumour nests. Left: ROC curve (black) of QTP features selected from tumour nests (N=45). Model threshold (gray cross) is the coefficient derived from the fitted model with maximum sensitivity and specificity.  Right: Scatter plot showing the possible relationship between Lowvsmed_dnaMean (y-axis) and VorPeriStdv (x-axis) to nodal status among tumour nests.  AUC, area under curve; N0, node negative; N+, node positive.     90  Tumour Nest Layers as Experimental Unit We also applied the same approach to analyze the performance ability for features measured from Layer#1, #2, and #3 (Figure 3-15), or from combined layers, Layer#1-2 or Layer#2-3 (Figure 3-16).  With similar trends, comparisons among Layer#1s and Layer #2s revealed that N+ group exhibited higher fraction of medium or high density chromatin and smaller nuclear size.   Layer Two feature selection AUC (%) Model Threshold Sensitivity (%) Specificity (%) Layer#1 Lowvsmed_dnaMean + High_den_objStdv 84 0.31 71 87 Layer#2 Short_runs2Stdv + Lowvsmed_dnaMean 82 0.62 75 93 Layer#3 AreaStdv + Freq_low_fftStdv 88 0.69 82 90 Figure 3-16. Receiver Operating Characteristic (ROC) Curve Analysis for QTP Features for Discrimination of N0 vs. N+, among Layers#1s, or Layers#2, or Layers#3 Blue ROC curve is graphed QTP features selected among Layer #1; Green ROC curve is graphed from QTP features selected among Layer#2; Red ROC curve is graphed from QTP features selected from Layer#3. Model thresholds (black crosses) are the coefficients from the fitted model with maximum sensitivity and specificity.  AUC, area under curve    91  Layer#1-2 showed the highest AUC of 94%, with combination of Lowvsmed_dnaMean and Mh_av_dstStdv, in stratifying the outermost layers into N0 and N+ (sensitivity, 100% and specificity, 75%) (Figure 3-17). There was no difference in AUC between ROC curve derived from Layer#1-2 or Layer#2-3 (P=0.17).   Layer  Two feature selection AUC (%) Model Threshold Sensitivity (%) Specificity (%) P Layer #1-2 Lowvsmed_dnaMean + Mh_av_dstStdv 94 0.28 100 75 0.17 Layer #2-3 Lowvsmed_dnaMean + MeddnaamtStdv 82 0.28 94 70  Figure 3-17. Receiver Operating Characteristic (ROC) Curve for QTP Features for Discrimination of N0 vs. N+, among Layer#1-2 or Layer#2-3 Left: ROC curves for QTP features selected from Layer#1-2 (cyan) and ROC curve for QTP features selected from Layer#2-3 (yellow). Right: Scatter plot showing the possible relationship between Lowvsmed_dnaMean (x-axis) and Mh_av_dstStdv (y-axis) to nodal status among Layer#1-2. Bottom panel: model threshold (crosses) is the coefficient from the fitted model with maximum sensitivity and specificity in each ROC curve. AUC for Layer#1-2 is 94% and for Layer#2-3 is 82%, P=0.17. AUC, area under curve    92  3.4 Discussion 3.4.1 Discussion of the Results Clinical evaluation of patient’s risk in nodal metastasis has always been difficult for clinicians as OSCC is heterogeneous in nature and ability to spread disregards conventional histological characteristics, as we have seen in Chapter 2. A recurring theme across global oral cancer research groups is the goal to uncovering the phenotypic map in OSCC patients. Computational image analysis has found applications in medical and biological fields since mid-90s’, with gradual derivatives of mathematical descriptors correlated with pathology and diagnosis of neoplasm. In this pilot study, the potential capacity of QTP image analysis to detect metastatic oral cancer was investigated. To our knowledge, this is the first attempt to use this technology of its kind in oral cancer research to solve this timely clinical problem.  Advantages of computational phenotyping such as QTP analysis comes in multi-fold. First and foremost is the objectivity towards describing detailed phenotypic characteristics, which mostly are previously unquantifiable characteristics. Second, its systematic workflow allows reproducible results while minimizing errors from inter- or intra- observer variability. Third, preparation of samples for QTP analysis is relatively similar and easy to perform as for immunohistochemistry. Generally, only up to two 4µm thick slides need to be sectioned, one for H&E and one for DNA staining. Subsequent steps are semi-automatic with minimum demands for human efforts in image scanning, segmentation, or analysis. This provides a benefit for clinical usage as QTP has a relatively fast turnover rate compared to genomic testing. Lastly, large scale genomic studies such as the genome-wide association studies (GWAS) seeks to find correlation between genes and outcomes; and results have been extensively deciphered. However, there is much to do in correlating these findings with the 'phenome' domain. QTP allows us to study alterations of cancer cells in microscopic scale at where we can begin to correlate the phenotypes to ‘nanoscopic’ genetic and epigenetic alterations; ultimately, understand the biology behind onset of nodal disease. In this pilot study, QTP analysis highlights 6 nuclear phenotype and 14 tissue architectural features in N+ OSCC versus N0 OSCC. The features concerning chromatin condensation stood out as different between N0 and N+ tumour nests or layers, suggesting 93  that these features may serve as indicators of nodal metastatic risk. In particular, we observed a higher ratio of condensed to less condensed chromatin in N+ than in N0 tumour nest or nest-layers. This finding agrees with other researches of aggregates of heterochromatin in cancer cells.148,150,162,163 Nuclear chromatin alteration has been extensively studied as one of the governors for genomic activity and functions, which result in aberrant chromatin remodeling associated with human diseases or cancers.150,164,165 During normal cell cycle, chromatin de-condenses and becomes loosely packed, euchromatin, exposing DNA and allowing gene activity and transcription. Whereas, condensed chromatin, heterochromatin, is tightly packed and corresponds to less gene activity. Altogether, changes in condensation states and relocation of chromatin to nucleus membrane occur throughout cell division. Therefore, it could be hypothesized that the observed higher fraction of condensed chromatin reflects enhanced and consistent proliferation of cells in N+ groups. Nonetheless, we also observed an increase in heterochromatin content in the outermost layer of the N+ tumour nests. Perhaps the N+ group tumour cells in these layers are associated with inflammation-related hypermethylation. Further investigations are warranted to include the tumour microenvironment and study the relationship between these tumour layer nuclei and cells in the tumour microenvironment, such as inflammatory cells. Another hypothesis on the observation of higher fraction of packed chromatin in metastatic tumour cells is that measurements from these discrete regions of low, medium, or high IOD not only infer chromatin condensation state, but are also indicative of DNA ploidy. The IOD is simply absorbance of light transmitted according to the amount of the absorbance material. However, ploidy can only be measured on intact cells; therefore, establishment of a panel of diploid standards from cytology imaging of non-metastatic or metastatic OSCC cells may aid in estimation of ploidy from analyzing tissue sections. With regards to tissue architecture, we observed a profound correlation between density and N+ status, as characterized by smaller surface area and shorter distance between neighbours. From this observation, we may infer the NC (nucleus-cytoplasm) ratio by associating nuclear area (size) with Voronoi polygon area. In normal circumstances, the NC ratio decreases as cells mature, which suggests that a larger NC ratio indicates atypical growth pattern, as seen in premalignant or cancer cells. In our pilot cohort, the average 94  nuclei area (or nuclei radius) did not differ between N+ and N0. However, the average polygon (cell) area was smaller in N+ group. If we take N0 group as the reference group, i.e. with NC ratio of 1:1, a smaller cytoplasm area with the same nuclear area would mean larger NC ratio. Analogously, we suggest that the NC ratio is associated with nodal status, which was supported by Student’s t test (P=0.01), but not by nested ANOVA analysis (P=0.11) (Figure 3-18). A similar relationship was also observed nest layers.   Figure 3-18. Scatter Plot and Box-whisker Plot of NC Ratio in N0 and N+ Tumour Nests Left: Scatter plot of relationship between VorAreaMean (x-axis) and AreaMean (y-axis) to nodal status.  Right: Box-whisker plot of distribution of NC ratio in N0 (gray) and N+ (red) tumour nests. The horizontal line within the box indicates the median, boundaries of the box (bottom and top) indicate the 25th and 75th percentile, and the whiskers indicate the highest and lowest values of the results, excluding outliers (closed circles)    It is important to be reminded that that all of the analyzed cell nuclei were cancerous squamous cells. Therefore, in this pilot study on population of exclusively cancer cells, little cell diversity is expected. As such, distinctive features between cancerous and normal or pre-cancerous cells that were reported as remarkable in other studies may not be reflected here. Indeed, one of the interesting findings of this study was the lack of statistically significant difference in nuclear fractal dimension (FD).  FD measures the extent of irregularity and complexity in the structural design of an object, in our cases, a cell nucleus. There is a wealth amount of researches dedicated to investigate FD as a prognostic factor in cancer reported higher FD is associated with poor prognosis.157-160 However, in this 95  study cohort, no remarkable differences were seen in FD; thus, we hypothesized that nuclear complexity in terms of FD was a consistent event at the state of OSCC.  The most important finding of our study is however, the value of QTP analysis as predictive factors and computation image analysis aids in pathology. Combination of measuring both chromatin texture and cell perimeter from tumour nests were shown to be sensitive and specific at differentiating N+ from N0 tumour nests (AUC, 85%). In addition, as we see in measurements from Layer#1-2, chromatin texture and variation in distance from the center of the nucleus were shown to be highly sensitive at differentiating N+ from N0 tumour nests with AUC 94%. To our knowledge, this pilot study is the first ever investigation of cell sociology of tumour nests as indication of nodal disease in OSCC. These nests are bordered with well-defined boundary; and immediately inside the boundary are most likely where hyperactive aggressive tumour cells reside. Indeed, the progression to cancer and metastasis require not only morphological changes in tumour cells themselves, but also cascades of events and responses with local microenvironment leading to loss of polarity, disruption in epithelial compartment, invasion and angiogenesis. Thus, with this novel approach to quantitatively analyzing morphological changes and architectural characteristics in the outer layers of tumour nests can potentially be useful as a clinical grading tool for predicting metastatic potential of primary OSCC tumours.  3.4.2 Study Limitations This pilot study was limited by the small number of patients included in this study. Although there were no differences seen in demographics and clinico-pathological characteristics, confirmation of the observed associations and predictive ability of QTP features in nodal disease requires confirmation in a larger sample size study with controlled and matched variables. Nevertheless, despite the small number of N0 and N+ patients, the sample size of  tumour nests and nest layers were considered large enough for statistical analysis, with adjustments for the possibility of multiple observations from the same patient.  Another limitation may came unadjusted histological characteristics such as differentiation, depth of invasion, growth type, pattern of invasion, presence of lymphovascular or perineural invasion, or presence of inflammation. However, as this is a 96  small pilot study to investigate the applicability of using QTP analysis on tumour nests, we did not performed analysis of association features with these histological factors. Nevertheless, we recognize such analyses are warranted in future studies with increasing sample size and detailed histology annotations. 3.4.3 Next Steps The study demonstrated the potential of QTP as an adjunct tool for pathologists in screening patients of highest risk to develop nodal disease. Although with relatively high sensitivity and specificity, there still remains a lot of ground work before we can verify the predictive value of QTP measurement. Ongoing efforts are happening to achieve such verification, involving expanding size of OSCC patient cohort and number of tumour nests in order to train and to demonstrate reproducibility of the results presented. Moving forward, if verified, a validation on an independent patient cohort is warranted. Furthermore, we also would like to take into account QTP features of inflammatory cells surrounding the tumour nests, especially those at the invasive front. This may help us explain the higher state of chromatin condensation seen in N+ group as well as elucidate the possible anti-cancer or pro-cancer interaction between the N+ and tumour microenvironment. The current QTP system has the capacity to analyze immune stained tissue specimens.  97         CHAPTER 4  4 CONCLUSIONS AND FUTURE DIRECTIONS  98  The key finding of this thesis is the recognition and a reminder that nodal disease is a real threat for BC population.166  The urgency for a new marker of risk in RF cannot be understated.  Our pilot data demonstrated that QTP is a potential marker to assess the risk of nodal disease.  Currently, most clinicians advocate prophylactic neck treatment on patients diagnosed with localized clinically node-negative disease. However, balancing between potential survival benefit through early intervention (~28%) and reduction of quality of life (~72%), prophylactic neck treatment should not be applied to all cN0 patients. Due to the short time frame in developing nodal disease after initial treatment, it is obvious that there are biologically and behaviourly different subgroups among early-stage OSCC. We have observed that some of the small tumour developed nodal disease yet some of the thicker tumour did. Therefore, the ability to predict which primary tumours are capable of metastasis would enable more individualized and early treatment to be delivered to patients at higher risk of regional failure. As we learned from Chapter 2, there is no shortage of evidence that advanced staged N+ OSCC is very aggressive and appears to be independent prognostic factor for survival. Our study has also confirmed that nodal disease is associated with a significant decrease in survival even in the face of early-stage disease. Prophylactic neck treatment including END did not improve patient survival. Reliance upon DOI of ≥4mm is assoicated with a higher incidence of RF but has poor sensitivity and specificity as a predictor for nodal disease for early-stage cN0 patients. Moreover, conventional pathological diagnosis comes with inherent limitations to differentiate and calling out microscopic abnormalities if attempting to determine metastatic nature of primary tumour. There is an alarming need to identify other effective markers, perhaps biomarkers, in order to improve survival. In Chapter 3, we demonstrated that computer assisted measurement of morphological changes or the disorder of architectural organization of tumour nests supports the hypothesis that QTP features have the potential as new prognostic factors of nodal disease in clinically node-negative OSCC patients. Additional studies are warranted, with increased number of patients, tumour nest and layers tumour to form an assistive system of high reliability, reproducibility, and objectivity. Finally, QTP feature measurements from diagnostic preoperative biopsy tissues may also be needed to deliver 99  prophylactic treatment at earliest time possible. In turn, we may find solutions to prognostic challenges that have been lingering for early-stage (cN0) OSCC patients who continue to battle with unpredictable risk of disease relapse and survival.  With the success of patient recruitment in the COOLS trial, we have established a prospective cohort of at least 200 patients from BC. With detailed information on treatment and clinical follow-up, this independent set of patient will allow us to validate our findings in Chapter 2, especially the efficacy of prophylactic neck treatment and histological prognostic factors in the development of nodal disease. In addition, correlation of QTP features with these clinico-pathological factors may also elucidate a group of patients who need individualized therapy. Moreover, our lab has banked high quality and well annotated biospecimens, including fresh frozen and formalin-fixed paraffin-embedded tissues, blood samples, and normal epithelial brushing samples. These valuable samples will allow us to incorporate next generation sequencing technologies to explore other genetic markers through studying biological behaviour of nodal disease. With this, we have the opportunity to provide more precise management for OSCC patients as well as prevent over- and under-treatment.  The ultimate goal is to identify biomarkers that can be applied on initial biopsy samples and thereafter design treatments depending on the risk of nodal disease, in real-time fashion. For example, patients with no risk of nodal disease, local excision only treatment is needed and an appropriate measure; whereas for patients with a risk for nodal disease, local excision with prophylactic neck treatment need to be done at earliest time as possible. In addition, we have the potential to identify tumours that are highly aggressive in developing distant metastasis. For these patients, designing counter treatment should be considered. This will eventually improve disease burden from patient’s and societal point of view, including direct and indirect costs of patient and health care from unnecessary treatment, hospitalization, resource utilization and treatment morbidities results from the uncertainty of the nodal disease.  In the era of personalized medicine and precision medicine, we are able to provide better care for our patients and ultimately better outcomes can be achieved.   100  BIBLIOGRAPHY 1. Sanderson RJ, Ironside JAD. Squamous cell carcinomas of the head and neck. Brit Med J. 2002;325(7368):822-827. 2. Canadian Cancer Statistics 2015. Toronto: Canadian Cancer Society;2015. 0835-2976. 3. Ferlay J, Soerjomataram I, Dikshit R, et al. Cancer incidence and mortality worldwide: sources, methods and major patterns in GLOBOCAN 2012. International journal of cancer. Journal international du cancer. 2015;136(5):E359-386. 4. Johnson NW, Jayasekara P, Amarasinghe AAHK. Squamous cell carcinoma and precursor lesions of the oral cavity: epidemiology and aetiology. Periodontol 2000. 2011;57:19-37. 5. Pathology and Genetics of Head and Neck Tumours. Lyon, France: IARC Press; 2005. Barnes L, Eveson JW, Reichart P, Sidransky D, eds. 6. Wenig BM. Squamous cell carcinoma of the upper aerodigestive tract: Precursors and problematic variants. Modern Pathol. 2002;15(3):229-254. 7. Cavenee WK. Genetic driver events in premalignancy: LOH validated for marking the risk of oral cancer. Cancer prevention research. 2012;5(9):1073-1074. 8. SEER Cancer Statistics Review, 1975-2012. Bethesda, MD: National Cancer Institute;2014. 9. Awojobi O, Newton JT, Scott SE. Why don't dentists talk to patients about oral cancer? Brit Dent J. 2015;218(9). 10. Boffetta P, Mashberg A, Winkelmann R, Garfinkel L. Carcinogenic Effect of Tobacco Smoking and Alcohol Drinking on Anatomic Sites of the Oral Cavity and Oropharynx. International Journal of Cancer. 1992;52(4):530-533. 11. Moore C, Catlin D. Anatomic Origins and Locations of Oral Cancer. Am J Surg. 1967;114(4):510-&. 12. Mashberg A, Samit AM. Early Detection, Diagnosis, and Management of Oral and Oropharyngeal Cancer. Ca-Cancer J Clin. 1989;39(2):67-88. 13. Brown B, Barnes L, Mazariegos J, et al. Prognostic Factors in Mobile Tongue and Floor of Mouth Carcinoma. Cancer. 1989;64(6):1195-1202. 14. Mashberg A, Meyers H. Anatomical Site and Size of 222 Early Asymptomatic Oral Squamous-Cell Carcinomas - Continuing Prospective-Study of Oral Cancer .2. Cancer. 1976;37(5):2149-2157. 15. Chong V. Oral cavity cancer. Cancer imaging : the official publication of the International Cancer Imaging Society. 2005;5 Spec No A:S49-52. 16. Zhang L, Cheung KJ, Jr., Lam WL, et al. Increased genetic damage in oral leukoplakia from high risk sites: potential impact on staging and clinical management. Cancer. 2001;91(11):2148-2155. 17. Liu W, Shi LJ, Wu L, et al. Oral Cancer Development in Patients with Leukoplakia - Clinicopathological Factors Affecting Outcome. Plos One. 2012;7(4). 18. Chaturvedi AK, Anderson WF, Lortet-Tieulent J, et al. Worldwide Trends in Incidence Rates for Oral Cavity and Oropharyngeal Cancers. J Clin Oncol. 2013;31(36):4550-4559. 19. Rivera C, Venegas B. Histological and molecular aspects of oral squamous cell carcinoma (Review). Oncol Lett. 2014;8(1):7-11. 101  20. Pai SI, Westra WH. Molecular pathology of head and neck cancer: implications for diagnosis, prognosis, and treatment. Annual review of pathology. 2009;4:49-70. 21. Edge SB, Compton CC. The American Joint Committee on Cancer: the 7th Edition of the AJCC Cancer Staging Manual and the Future of TNM. Ann Surg Oncol. 2010;17(6):1471-1474. 22. Bailey BJ, Calhoun KH, Healy GB, et al., eds. Head and Neck Surgery - Otolaryngology. 3rd ed: Wolters Kluwer; 2001; No. 1. 23. Kurita H, Nakanishi Y, Nishizawa R, et al. Impact of different surgical margin conditions on local recurrence of oral squamous cell carcinoma. Oral Oncol. 2010;46(11):814-817. 24. Bernier J, Domenge C, Ozsahin M, et al. Postoperative irradiation with or without concomitant chemotherapy for locally advanced head and neck cancer. New Engl J Med. 2004;350(19):1945-1952. 25. Bernier J, Cooper JS, Pajak TF, et al. Defining risk levels in locally advanced head and neck cancers: A comparative analysis of concurrent postoperative radiation plus chemotherapy trials of the EORTC (#22931) and RTOG (#9501). Head Neck-J Sci Spec. 2005;27(10):843-850. 26. Cooper JS, Pajak TF, Forastiere AA, et al. Postoperative concurrent radiotherapy and chemotherapy for high-risk squamous-cell carcinoma of the head and neck. New Engl J Med. 2004;350(19):1937-1944. 27. Carvalho AL, Magrin J, Kowalski LP. Sites of recurrence in oral and oropharyngeal cancers according to the treatment approach. Oral diseases. 2003;9(3):112-118. 28. Gonzalez-Garcia R, Naval-Gias L, Roman-Romero L, Sastre-Perez J, Rodriguez-Campo FJ. Local Recurrences and Second Primary Tumors from Squamous Cell Carcinoma of the Oral Cavity: A Retrospective Analytic Study of 500 Patients. Head Neck-J Sci Spec. 2009;31(9):1168-1180. 29. Rennemo E, Zatterstrom U, Boysen M. Outcome of local failures after oral cancer - recurrence vs. second primary. J Oral Pathol Med. 2010;39(9):657-661. 30. Luryi AL, Chen MM, Mehra S, et al. Treatment Factors Associated With Survival in Early-Stage Oral Cavity Cancer Analysis of 6830 Cases From the National Cancer Data Base. Jama Otolaryngol. 2015;141(7):593-598. 31. Leibel SA, Scott CB, Mohiuddin M, et al. The Effect of Local-Regional Control on Distant Metastatic Dissemination in Carcinoma of the Head and Neck - Results of an Analysis from the Rtog Head and Neck Database. Int J Radiat Oncol. 1991;21(3):549-556. 32. Hahn SS, Spaulding CA, Kim JA, Constable WC. The prognostic significance of lymph node involvement in pyriform sinus and supraglottic cancers. International journal of radiation oncology, biology, physics. 1987;13(8):1143-1147. 33. Mamelle G, Pampurik J, Luboinski B, et al. Lymph node prognostic factors in head and neck squamous cell carcinomas. Am J Surg. 1994;168(5):494-498. 34. Woolgar JA, Scott J, Vaughan ED, et al. Survival, metastasis and recurrence of oral cancer in relation to pathological features. Annals of the Royal College of Surgeons of England. 1995;77(5):325-331. 35. Godden DRP, Ribeiro NFF, Hassanein K, Langton SG. Recurrent neck disease in oral cancer. J Oral Maxil Surg. 2002;60(7):748-753. 102  36. Mayland MK, Sessions DG, Lenox J. The influence of lymph node metastasis in the treatment of squamous cell carcinoma of the oral cavity, oropharynx, larynx, and hypopharynx: N0 versus N+. Laryngoscope. 2005;115(4):629-639. 37. Sano D, Myers JN. Metastasis of squamous cell carcinoma of the oral tongue. Cancer Metast Rev. 2007;26(3-4):645-662. 38. Yuasa-Nakagawa K, Shibuya H, Yoshimura R, et al. Cervical lymph node metastasis from early-stage squamous cell carcinoma of the oral tongue. Acta Oto-Laryngol. 2013;133(5):544-551. 39. Shah JP. Patterns of Cervical Lymph-Node Metastasis from Squamous Carcinomas of the Upper Aerodigestive Tract. Am J Surg. 1990;160(4):405-409. 40. Wong SY, Hynes RO. Lymphatic or hematogenous dissemination: How does a metastatic tumor cell decide? Cell Cycle. 2006;5(8):812-817. 41. Teichgraeber JF, Clairmont AA. The Incidence of Occult Metastases for Cancer of the Oral Tongue and Floor of the Mouth - Treatment Rationale. Head Neck Surg. 1984;7(1):15-21. 42. Yuen APW, Wei WI, Wong YM, Tang KC. Elective neck dissection versus observation in the treatment of early oral tongue carcinoma. Head Neck-J Sci Spec. 1997;19(7):583-588. 43. Cunningham MJ, Johnson JT, Myers EN, Schramm VL, Thearle PB. Cervical Lymph-Node Metastasis after Local Excision of Early Squamous-Cell Carcinoma of the Oral Cavity. Am J Surg. 1986;152(4):361-366. 44. Ho CM, Lam KH, Wei WI, Lau SK, Lam LK. Occult Lymph-Node Metastasis in Small Oral Tongue Cancers. Head Neck-J Sci Spec. 1992;14(5):359-363. 45. Robbins KT, Medina JE, Wolfe GT, et al. Standardizing Neck Dissection Terminology - Official Report of the Academy Committee-for-Head-and-Neck-Surgery-and-Oncology. Arch Otolaryngol. 1991;117(6):601-605. 46. Robbins KT, Shaha AR, Medina JE, et al. Consensus statement on the classification and terminology of neck dissection. Arch Otolaryngol. 2008;134(5):536-538. 47. Eibling DE, Snyderman CH, Weber PC, Johnson JT, Steed DL. Internal Jugular-Vein Reconstruction in Bilateral Radical Neck Dissection. American journal of otolaryngology. 1995;16(4):260-264. 48. Robbins KT, Ferlito A, Shah JP, et al. The evolving role of selective neck dissection for head and neck squamous cell carcinoma. Eur Arch Oto-Rhino-L. 2013;270(4):1195-1202. 49. Eibling DE. Neck Dissection. In: Myers EN, ed. Operative Otolaryngology: Head and Neck Surgery. 2nd ed. Philadelphia, PA: Saunders Elsevier; 2008:30. 50. Shah JP, Strong E, Spiro RH, Vikram B. Neck Dissection - Current Status and Future Possibilities. Clin Bull. 1981;11(1):25-33. 51. Terrell JE, Welsh DE, Bradford CR, et al. Pain, quality of life, and spinal accessory nerve status after neck dissection. Laryngoscope. 2000;110(4):620-626. 52. Inoue H, Nibu K, Saito M, et al. Quality of life after neck dissection. Arch Otolaryngol. 2006;132(6):662-666. 53. Nibu K, Ebihara Y, Ebihara M, et al. Quality of life after neck dissection: a multicenter longitudinal study by the Japanese Clinical Study Group on Standardization of Treatment for Lymph Node Metastasis of Head and Neck Cancer. Int J Clin Oncol. 2010;15(1):33-38. 103  54. Shah S, Har-El G, Rosenfeld RM. Short-term and long-term quality of life after neck dissection. Head Neck-J Sci Spec. 2001;23(11):954-961. 55. Stuiver MM, van Wilgen CP, de Boer EM, et al. Impact of shoulder complaints after neck dissection on shoulder disability and quality of life. Otolaryng Head Neck. 2008;139(1):32-39. 56. Yasui A, Okada Y, Mataga I, Katagiri M. An Analysis of Distant Metastases in Oral Squamous Cell Carcinoma. J Hard Tissue Biol. 2010;19(1):27-31. 57. Myers JN, Greenberg JS, Mo V, Roberts D. Extracapsular spread. A significant predictor of treatment failure in patients with squamous cell carcinoma of the tongue. Cancer. 2001;92(12):3030-3036. 58. McMahon J, O'Brien CJ, Pathak I, et al. Influence of condition of surgical margins on local recurrence and disease-specific survival in oral and oropharyngeal cancer. Brit J Oral Max Surg. 2003;41(4):224-231. 59. Pathak KA, Das AK, Agarwal R, et al. Selective neck dissection (I-III) for node negative and node positive necks. Oral Oncol. 2006;42(8):837-841. 60. Fakih AR, Rao RS, Borges AM, Patel AR. Elective Versus Therapeutic Neck Dissection in Early Carcinoma of the Oral Tongue. Am J Surg. 1989;158(4):309-313. 61. Lim YC, Lee JS, Koo BS, et al. Treatment of contralateral N0 neck in early squamous cell carcinoma of the oral tongue: Elective neck dissection versus observation. Laryngoscope. 2006;116(3):461-465. 62. Keski-Santti H, Atula T, Tornwall J, Koivunen P, Makitie A. Elective neck treatment versus observation in patients with T1/T2 NO squamous cell carcinoma of oral tongue. Oral Oncol. 2006;42(1):96-101. 63. Alkhalil M, Smajilagic A, Redzic A. The lymph node neck metastasis in oral cancer and elective neck dissection as method of choice. Med Glas. 2007;4(2):94-95. 64. Il Jang W, Wu HG, Il Park C, et al. Treatment of patients with clinically lymph node-negative squamous cell carcinoma of the oral cavity. Jpn J Clin Oncol. 2008;38(6):395-401. 65. Fasunla AJ, Greene BH, Timmesfeld N, et al. A meta-analysis of the randomized controlled trials on elective neck dissection versus therapeutic neck dissection in oral cavity cancers with clinically node-negative neck. Oral Oncol. 2011;47(5):320-324. 66. Vandenbrouck C, Sanchogarnier H, Chassagne D, et al. Elective Versus Therapeutic Radical Neck Dissection in Epidermoid Carcinoma of the Oral Cavity - Results of a Randomized Clinical-Trial. Cancer. 1980;46(2):386-390. 67. Kligerman J, Lima RA, Soares JR, et al. Supraomohyoid Neck Dissection in the Treatment of T1/T2 Squamous-Cell Carcinoma of Oral Cavity. Am J Surg. 1994;168(5):391-394. 68. Yuen AP, Ho CM, Chow TL, et al. Prospective randomized study of selective neck dissection versus observation for N0 neck of early tongue carcinoma. Head & neck. 2009;31(6):765-772. 69. Liu TR, Chen FJ, Yang AK, et al. Elective neck dissection in clinical stage I squamous cell carcinoma of the tongue: Does it improve regional control or survival time? Oral Oncol. 2011;47(2):136-141. 104  70. D'Cruz AK, Siddachari RC, Walvekar RR, et al. Elective neck dissection for the management of the N0 neck in early cancer of the oral tongue: need for a randomized controlled trial. Head & neck. 2009;31(5):618-624. 71. Arya S, Rane P, Deshmukh A. Oral cavity squamous cell carcinoma: Role of pretreatment imaging and its influence on management. Clin Radiol. 2014;69(9):916-930. 72. Boland PW, Watt-Smith SR, Hopper C, Golding SJ. Magnetic resonance imaging predicts survival and occult metastasis in oral cancer: a dual-centre, retrospective study. Brit J Oral Max Surg. 2013;51(8):696-701. 73. August M, Nguyen M. Evaluation of Metastatic Neck Disease by Computed-Tomography. Int J Oral Max Surg. 1994;23(5):290-293. 74. Jank S, Robatscher R, Errishoff R, et al. The diagnostic value of ultrasonography to detect occult lymph node involvement at different levels in patients with squamous cell carcinoma in the maxillofacial region. Int J Oral Max Surg. 2003;32(1):39-42. 75. Yoshida H, Yusa H, Ueno E, Tohno E, Tsunoda-Shimizu H. Ultrasonographic evaluation of small cervical lymph nodes in head and neck cancer. Ultrasound Med Biol. 1998;24(5):621-629. 76. Stuckensen T, Kovacs AF, Adams S, Baum RP. Staging of the neck in patients with oral cavity squamous cell carcinomas: a prospective comparison of PET, ultrasound, CT and MRI. J Cranio Maxill Surg. 2000;28(6):319-324. 77. Woolgar JA. Pathology of the N0 neck. Brit J Oral Max Surg. 1999;37(3):205-209. 78. Merritt RM, Williams MF, James TH, Porubsky ES. Detection of cervical metastasis - A meta-analysis comparing computed tomography with physical examination. Arch Otolaryngol. 1997;123(2):149-152. 79. Haberal I, Celik H, Gocmen H, et al. Which is important in the evaluation of metastatic lymph nodes in head and neck cancer: Palpation, ultrasonography, or computed tomography? Otolaryng Head Neck. 2004;130(2):197-201. 80. Akoglu E, Dutipek M, Bekis R, et al. Assessment of cervical lymph node metastasis with different imaging methods in patients with head and neck squamous cell carcinoma. J Otolaryngol. 2005;34(6):384-394. 81. Schoder H, Carlson DL, Kraus DH, et al. F-18-FDG PET/CT for detecting nodal metastases in patients with oral cancer staged N0 by clinical examination and CT/MRI. J Nucl Med. 2006;47(5):755-762. 82. Seitz O, Chambron-Pinho N, Middendorp M, et al. F-18-Fluorodeoxyglucose-PET/CT to evaluate tumor, nodal disease, and gross tumor volume of oropharyngeal and oral cavity cancer: comparison with MR imaging and validation with surgical specimen. Neuroradiology. 2009;51(10):677-686. 83. Torabi M, Aquino SL, Harisinghani MG. Current concepts in lymph node imaging. J Nucl Med. 2004;45(9):1509-1518. 84. Werner JA, Dunne AA, Myers JN. Functional anatomy of the lymphatic drainage system of the upper aerodigestive tract and its role in metastasis of squamous cell carcinoma. Head Neck-J Sci Spec. 2003;25(4):322-332. 85. Lengele B, Hamoir M, Scalliet P, Gregoire V. Anatomical bases for the radiological delineation of lymph node areas. Major collecting trunks, head and neck. Radiother Oncol. 2007;85(1):146-155. 105  86. Werner JA, Lippert BM, Folz BJ, Nabavi A, Rudert H. Regional metastases in head and neck cancer: pathways, diagnostics, surgical therapy (with video presentation). Brit J Cancer. 1998;77:18-18. 87. Lindberg R. Distribution of cervical lymph node metastases from squamous cell carcinoma of the upper respiratory and digestive tracts. Cancer. 1972;29(6):1446-1449. 88. Ross G, Shoaib T, Soutar DS, et al. The use of sentinel node biopsy to upstage the clinically N0 neck in head and neck cancer. Arch Otolaryngol. 2002;128(11):1287-1291. 89. Chikamatsu K, Kamada H, Ninomiya H, et al. A preliminary study on sentinel lymph node biopsy: feasibility and predictive ability in oral cavity cancer. Ann Nucl Med. 2004;18(3):257-262. 90. Hoft S, Maune S, Muhle C, et al. Sentinel lymph-node biopsy in head and neck cancer. Brit J Cancer. 2004;91(1):124-128. 91. Civantos FJ, Zitsch RP, Schuller DE, et al. Sentinel Lymph Node Biopsy Accurately Stages the Regional Lymph Nodes for T1-T2 Oral Squamous Cell Carcinomas: Results of a Prospective Multi-Institutional Trial. J Clin Oncol. 2010;28(8):1395-1400. 92. Stefanicka P, Profant M, Duchaj B, et al. Sentinel lymph node radiolocalization and biopsy in oral cavity and oropharynx mucosal squamous cell carcinoma. Bratisl Med J. 2010;111(11):590-594. 93. Gurney BAS, Schilling C, Putcha V, et al. Implications of a positive sentinel node in oral squamous cell carcinoma. Head Neck-J Sci Spec. 2012;34(11):1580-1585. 94. Alkureishi LWT, Ross GL, Shoaib T, et al. Sentinel Node Biopsy in Head and Neck Squamous Cell Cancer: 5-Year Follow-Up of a European Multicenter Trial. Ann Surg Oncol. 2010;17(9):2459-2464. 95. Civantos FJ, Stoeckli SJ, Takes RP, et al. What is the role of sentinel lymph node biopsy in the management of oral cancer in 2010? Eur Arch Oto-Rhino-L. 2010;267(6):839-844. 96. O'Brien CJ, Lauer CS, Fredricks S, et al. Tumor thickness influences prognosis of T1 and T2 oral cavity cancer - But what thickness? Head Neck-J Sci Spec. 2003;25(11):937-945. 97. Breslow A. Thickness, Cross-Sectional Areas and Depth of Invasion in Prognosis of Cutaneous Melanoma. Ann Surg. 1970;172(5):902-&. 98. Moore C, Kuhns JG, Greenberg RA. Thickness as Prognostic Aid in Upper Aerodigestive Tract Cancer. Arch Surg-Chicago. 1986;121(12):1410-1414. 99. Almangush A, Bello IO, Keski-Santti H, et al. Depth of invasion, tumor budding, and worst pattern of invasion: Prognostic indicators in early-stage oral tongue cancer. Head Neck-J Sci Spec. 2014;36(6):811-818. 100. Asakage T, Yokose T, Mukai K, et al. Tumor thickness predicts cervical metastasis in patients with stage I/II carcinoma of the tongue. Cancer. 1998;82(8):1443-1448. 101. Byers RM, El-Naggar AK, Lee YY, et al. Can we detect or predict the presence of occult nodal metastases in patients with squamous carcinoma of the oral tongue? Head Neck-J Sci Spec. 1998;20(2):138-144. 102. Chen YW, Yu EH, Wu TH, et al. Histopathological factors affecting nodal metastasis in tongue cancer: analysis of 94 patients in Taiwan. Int J Oral Max Surg. 2008;37(10):912-916. 106  103. Fukano H, Matsuura H, Hasegawa Y, Nakamura S. Depth of invasion as a predictive factor for cervical lymph node metastasis in tongue carcinoma. Head Neck-J Sci Spec. 1997;19(3):205-210. 104. Hayashi T, Ito J, Taira S, Katsura K. The relationship of primary tumor thickness in carcinoma of the tongue to subsequent lymph node metastasis. Dento maxillo facial radiology. 2001;30(5):242-245. 105. Hosal AS, Unal OF, Ayhan A. Possible prognostic value of histopathologic parameters in patients with carcinoma of the oral tongue. European archives of oto-rhino-laryngology : official journal of the European Federation of Oto-Rhino-Laryngological Societies. 1998;255(4):216-219. 106. Ebrahimi A, Gil Z, Amit M, et al. Primary Tumor Staging for Oral Cancer and a Proposed Modification Incorporating Depth of Invasion An International Multicenter Retrospective Study. Jama Otolaryngol. 2014;140(12):1138-1148. 107. Jones KR, Lodgerigal RD, Reddick RL, Tudor GE, Shockley WW. Prognostic Factors in the Recurrence of Stage-I and Stage-Ii Squamous-Cell Cancer of the Oral Cavity. Arch Otolaryngol. 1992;118(5):483-485. 108. Jung J, Cho NH, Kim J, et al. Significant invasion depth of early oral tongue cancer originated from the lateral border to predict regional metastases and prognosis. Int J Oral Max Surg. 2009;38(6):653-660. 109. Kim HC, Kusukawa J, Kameyama T. Clinicopathologic parameters in predicting cervical nodal metastasis in early squamous cell carcinoma of the oral cavity. The Kurume medical journal. 1993;40(4):183-192. 110. Kurokawa H, Yamashita Y, Takeda S, et al. Risk factors for late cervical lymph node metastases in patients with stage I or II carcinoma of the tongue. Head Neck-J Sci Spec. 2002;24(8):731-736. 111. Lim SC, Zhang SC, Ishii G, et al. Predictive markers for late cervical metastasis in stage I and II invasive squamous cell carcinoma of the oral tongue. Clin Cancer Res. 2004;10(1):166-172. 112. Mohittabatabai MA, Sobel HJ, Rush BF, Mashberg A. Relation of Thickness of Floor of Mouth Stage-I and Stage-Ii Cancers to Regional Metastasis. Am J Surg. 1986;152(4):351-353. 113. Nakagawa T, Shibuya H, Yoshimura R, et al. Neck node metastasis after successful brachytherapy for early stage tongue carcinoma. Radiother Oncol. 2003;68(2):129-135. 114. Nathanson A, Agren K, Biorklund A, et al. Evaluation of Some Prognostic Factors in Small Squamous-Cell Carcinoma of the Mobile Tongue - a Multicenter Study in Sweden. Head Neck-J Sci Spec. 1989;11(5):387-392. 115. O-charoenrat P, Pillai G, Patel S, et al. Tumour thickness predicts cervical nodal metastases and survival in early oral tongue cancer. Oral Oncol. 2003;39(4):386-390. 116. Sparano A, Weinstein G, Chalian A, Yodul M, Weber R. Multivariate predictors of occult neck metastasis in early oral tongue cancer. Otolaryng Head Neck. 2004;131(4):472-476. 117. Spiro RH, Huvos AG, Wong GY, et al. Predictive Value of Tumor Thickness in Squamous Carcinoma Confined to the Tongue and Floor of the Mouth. Am J Surg. 1986;152(4):345-350. 107  118. Steinhart H, Kleinsasser O. Growth and Spread of Squamous-Cell Carcinoma of the Floor of the Mouth. Eur Arch Oto-Rhino-L. 1993;250(6):358-361. 119. Urist MM, Obrien CJ, Soong SJ, Visscher DW, Maddox WA. Squamous-Cell Carcinoma of the Buccal Mucosa - Analysis of Prognostic Factors. Am J Surg. 1987;154(4):411-414. 120. Yuen APW, Lam KY, Lam LK, et al. Prognostic factors of clinically stage I and II oral tongue carcinoma - A comparative study of stage, thickness, shape, growth pattern, invasive front malignancy grading, Martinez-Gimeno score, and pathologic features. Head Neck-J Sci Spec. 2002;24(6):513-520. 121. Bryne M, Nielsen K, Koppang HS, Dabelsteen E. Reproducibility of 2 Malignancy Grading Systems with Reportedly Prognostic Value for Oral-Cancer Patients. J Oral Pathol Med. 1991;20(8):369-372. 122. Yoshioka Y, Yoshida K, Shimizutani K, et al. Proposal of a new grading system for evaluation of tongue hemiatrophy as a late effect of brachytherapy for oral tongue cancer. Radiother Oncol. 2001;61(1):87-92. 123. Okamoto M, Ozeki S, Watanabe T, Iida Y, Tashiro H. Cervical Lymph-Node Metastasis in Carcinoma of the Tongue - Correlation between Clinical and Histopathological Findings and Metastasis. J Cranio Maxill Surg. 1988;16(1):31-34. 124. Bryne M, Koppang HS, Lilleng R, et al. New Malignancy Grading Is a Better Prognostic Indicator Than Broders Grading in Oral Squamous-Cell Carcinomas. J Oral Pathol Med. 1989;18(8):432-437. 125. Angadi PV, Patil PV, Hallikeri K, et al. Tumor Budding Is an Independent Prognostic Factor for Prediction of Lymph Node Metastasis in Oral Squamous Cell Carcinoma. Int J Surg Pathol. 2015;23(2):102-110. 126. Crissman JD, Liu WY, Gluckman JL, Cummings G. Prognostic Value of Histopathologic Parameters in Squamous-Cell Carcinoma of the Oropharynx. Cancer. 1984;54(12):2995-3001. 127. Nakanishi Y, Ochiai A, Kato H, et al. Clinicopathological significance of tumor nest configuration in patients with esophageal squamous coil carcinoma. Cancer. 2001;91(6):1114-1120. 128. Rahima B, Shingaki S, Nagata M, Saito C. Prognostic significance of perineural invasion in oral and oropharyngeal carcinoma. Oral Surg Oral Med O. 2004;97(4):423-431. 129. Kurtz KA, Hoffman HT, Zimmerman B, Robinson RA. Perineural and vascular invasion in oral cavity squamous carcinoma - Increased incidence on re-review of slides and by using immunohistochemical enhancement. Arch Pathol Lab Med. 2005;129(3):354-359. 130. Flynn CJ, Khaouam N, Gardner S, et al. The Value of Periodic Follow-up in the Detection of Recurrences after Radical Treatment in Locally Advanced Head and Neck Cancer. Clin Oncol-Uk. 2010;22(10):868-873. 131. Fagan JJ, Collins B, Barnes L, et al. Perineural invasion in squamous cell carcinoma of the head and neck. Arch Otolaryngol. 1998;124(6):637-640. 132. Kowalski LP, Medina JE. Nodal metastases: predictive factors. Otolaryngol Clin North Am. 1998;31(4):621-637. 133. Bignold LP. The mutator phenotype theory of carcinogenesis and the complex histopathology of tumours: support for the theory from the independent occurrence of 108  nuclear abnormality, loss of specialisation and invasiveness among occasional neoplastic lesions. Cell Mol Life Sci. 2003;60(5):883-891. 134. Young IT, Verbeek PW, Mayall BH. Characterization of Chromatin Distribution in Cell-Nuclei. Cytometry. 1986;7(5):467-474. 135. Zink D, Fischer AH, Nickerson JA. Nuclear structure in cancer cells. Nat Rev Cancer. 2004;4(9):677-687. 136. Nielsen B, Albregtsen F, Danielsen HE. A review of statistical nuclear texture analysis in cancer research. Cell Oncol. 2008;30(2):165-165. 137. Nielsen B, Hveem TS, Kildal W, et al. Entropy-Based Adaptive Nuclear Texture Features are Independent Prognostic Markers in a Total Population of Uterine Sarcomas. Cytom Part A. 2015;87A(4):315-325. 138. Singh H, Sekinger EA, Gross DS. Chromatin and cancer: Causes and consequences. J Cell Biochem. 2000:61-68. 139. Leonhardt H, Cardoso MC. DNA methylation, nuclear structure, gene expression and cancer. J Cell Biochem. 2000:78-83. 140. Li GH, Reinberg D. Chromatin higher-order structures and gene regulation. Curr Opin Genet Dev. 2011;21(2):175-186. 141. Guillaud M, Zhang LW, Poh C, Rosin MP, MacAulay C. Potential use of quantitative tissue phenotype to predict malignant risk for oral premalignant lesions. Cancer Res. 2008;68(9):3099-3107. 142. Guillaud M, Buys TPH, Carraro A, et al. Evaluation of HPV Infection and Smoking Status Impacts on Cell Proliferation in Epithelial Layers of Cervical Neoplasia. Plos One. 2014;9(9). 143. Mulder JWR, Offerhaus GJA, Defeyter EP, et al. The Relationship of Quantitative Nuclear Morphology to Molecular Genetic Alterations in the Adenoma-Carcinoma Sequence of the Large-Bowel. Am J Pathol. 1992;141(4):797-804. 144. Dong F, Irshad H, Oh EY, et al. Computational Pathology to Discriminate Benign from Malignant Intraductal Proliferations of the Breast. Plos One. 2014;9(12). 145. Tambasco M, Eliasziw M, Magliocco AM. Morphologic complexity of epithelial architecture for predicting invasive breast cancer survival. Journal of translational medicine. 2010;8:140. 146. Wang LW, Qu AP, Yuan JP, et al. Computer-based image studies on tumor nests mathematical features of breast cancer and their clinical prognostic value. Plos One. 2013;8(12):e82314. 147. Beck AH, Sangoi AR, Leung S, et al. Systematic analysis of breast cancer morphology uncovers stromal features associated with survival. Science translational medicine. 2011;3(108):108ra113. 148. Vanvelthoven R, Petein M, Oosterlinck WJ, et al. The Use of Digital Image-Analysis of Chromatin Texture in Feulgen-Stained Nuclei to Predict Recurrence of Low-Grade Superficial Transitional-Cell Carcinoma of the Bladder. Cancer. 1995;75(2):560-568. 149. Gao F, Jutting U, Rodenacker K, Gais P, Lin PZ. Relevance of chromatin features in the progression of esophageal epithelial severe dysplasia. Anal Cell Pathol. 1997;13(1):17-28. 150. Cherkezyan L, Stypula-Cyrus Y, Subramanian H, et al. Nanoscale changes in chromatin organization represent the initial steps of tumorigenesis: a transmission electron microscopy study. Bmc Cancer. 2014;14. 109  151. Zarella MD, Breen DE, Reza MA, Milutinovic A, Garcia FU. Lymph Node Metastasis Status in Breast Carcinoma Can Be Predicted via image Analysis of Tumor Histology. Anal Quant Cytopatho. 2015;37(5):273-285. 152. Veltri RW, Khan MA, Miller MC, et al. Ability to predict metastasis based on pathology findings and alterations in nuclear structure of normal-appearing and cancer peripheral zone epithelium in the prostate. Clin Cancer Res. 2004;10(10):3465-3473. 153. Natarajan S, Mahajan S, Boaz K, George T. Prediction of lymph node metastases by preoperative nuclear morphometry in oral squamous cell carcinoma: A comparative image analysis study. Indian J Cancer. 2010;47(4):406-411. 154. Karino M, Nakatani E, Hideshima K, et al. Applicability of Preoperative Nuclear Morphometry to Evaluating Risk for Cervical Lymph Node Metastasis in Oral Squamous Cell Carcinoma. Plos One. 2014;9(12). 155. Poh CF, Durham JS, Brasher PM, et al. Canadian Optically-guided approach for Oral Lesions Surgical (COOLS) trial: study protocol for a randomized controlled trial. BMC Cancer. 2011;11:462. 156. D'Cruz A, Dandekar M, Vaish R, et al. Elective versus therapeutic neck dissection in the clinically node negative early oral cancer: A randomised control trial (RCT). J Clin Oncol. 2015;33(15). 157. Delides A, Panayiotides I, Alegakis A, et al. Fractal dimension as a prognostic factor for laryngeal carcinoma. Anticancer Res. 2005;25(3B):2141-2144. 158. Oczeretko E, Juczewska M, Kasacka I. Fractal geometric analysis of lung cancer angiogenic patterns. Folia Histochem Cyto. 2001;39:75-76. 159. Dey P, Rajesh L. Fractal dimension in endometrial carcinoma. Anal Quant Cytol. 2004;26(2):113-116. 160. Yokoyama T, Kawahara A, Kage M, et al. Image analysis of irregularity of cluster shape in cytological diagnosis of breast tumors: Cluster analysis with 2D-fractal dimension. Diagn Cytopathol. 2005;33(2):71-77. 161. Doudkine A, Macaulay C, Poulin N, Palcic B. Nuclear texture measurements in image cytometry. Pathologica. 1995;87(3):286-299. 162. Cremer M, Kupper K, Wagler B, et al. Inheritance of gene density-related higher order chromatin arrangements in normal and tumor cell nuclei. J Cell Biol. 2003;162(5):809-820. 163. Poulin N, Frost A, Carraro A, et al. Risk biomarker assessment for breast cancer progression: Replication precision of nuclear morphometry. Anal Cell Pathol. 2003;25(3):129-138. 164. Croft JA, Bridger JM, Boyle S, et al. Differences in the localization and morphology of chromosomes in the human nucleus. J Cell Biol. 1999;145(6):1119-1131. 165. Young IT, Verbeek PW, Mayall BH. Characterization of chromatin distribution in cell nuclei. Cytometry. 1986;7(5):467-474. 166. Cancer Surveillance and Outcomes, Population Oncology. BC Cancer Agency; 2014. Accessed October 1, 2014.    110  Appendix A. International Classification of Disease-09 Version: 2015  Malignant neoplasm of oral cavity C02.0 Dorsal face of tongue C02.1 Border of tongue C02.2 Ventral surface of tongue C02.3 Anterior two-thirds of tongue C02.4 Lingual Tonsil C02.8a Overlapping lesion of tongue C02.9 Tongue, unspecified C03.0 Upper gumb C03.1 Lower gumb C03.9 Gum, site not specified C04.0 Anterior floor of mouth C04.1 Lateral floor of mouth C04.8a Overlapping lesion of floor of mouth C04.9 Floor of mouth, unspecified C05.0 Hard Palate C05.1 Soft Palate C05.2 Uvula C05.3a Overlapping lesion of palate C05.9 Palate, unspecified C06.0 Cheek mucosac C06.1 Vestibule of mouth C06.2 Retromolar area C06.8a Overlapping lesion of other and unspecified parts of mouth C06.9 Mouth, unspecified aPoint of origin cannot be classified to other category bGum, Alveolar (ridge) mucosa; gingiva cCheek mucosa, buccal mucosa, internal aspect of cheek   111  Appendix B. Patient Demographics and Consultation Report 	 112  Appendix C. Surgical Pathology Report Form   113  Appendix D. Radiotherapy or Chemotherapy Report  114  Appendix E. Quantitative Tissue Pathology (QTP) Phenotypes Phenotypes Category Sub-category QTP Feature Name Nuclear Morphology  Area (mean/stdv) Nuclear Morphology  Mean_radius (mean/stdv) Nuclear Morphology  Max_radisu (mean/stdv) Nuclear Morphology  Var_radius (mean/stdv) Nuclear Morphology  Sphericity (mean/stdv) Nuclear Morphology  Eccentricity (mean/stdv) Nuclear Morphology  Freq_low_fft (mean/stdv) Nuclear Morphology  Freq_high_fft (mean/stdv) Nuclear Morphology  Harmon01 (mean/stdv) Nuclear Morphology  Harmon02 (mean/stdv) Nuclear Optical Density  Dna_index (mean/stdv) Nuclear Optical Density  Od_maximum (mean/stdv) Nuclear Optical Density  Od_variance (mean/stdv) Nuclear Optical Density  Od_skewness (mean/stdv) Nuclear Optical Density  Od_kurtosis (mean/stdv) Nuclear Optical Density  Var_intensity (mean/stdv) Nuclear Chromatin Texture Fractal Texture Fractal_dimen (mean/stdv) Nuclear Chromatin Texture Run Length Short_runs1 (mean/stdv) Nuclear Chromatin Texture Run Length Short_runs2 (mean/stdv) Nuclear Chromatin Texture Run Length Long_runs1 (mean/stdv) Nuclear Chromatin Texture Run Length Long_runs2 (mean/stdv) Nuclear Chromatin Texture Run Length Gray_level1 (mean/stdv) Nuclear Chromatin Texture Run Length Gray_level2 (mean/stdv) Nuclear Chromatin Texture Run Length Run_length1 (mean/stdv) Nuclear Chromatin Texture Run Length Run_length2 (mean/stdv) Nuclear Chromatin Texture Run Length Run_percent1 (mean/stdv) Nuclear Chromatin Texture Run Length Run_percent2 (mean/stdv) Nuclear Chromatin Texture Markovian Texture Entropy (mean/stdv) Nuclear Chromatin Texture Markovian Texture Energy (mean/stdv) Nuclear Chromatin Texture Markovian Texture Correlation (mean/stdv) Nuclear Chromatin Texture Markovian Texture Contrast (mean/stdv) Nuclear Chromatin Texture Markovian Texture Homogeneity (mean/stdv) Nuclear Chromatin Texture Discrete Texture Lowdnaarea (mean/stdv) Nuclear Chromatin Texture Discrete Texture Meddnaarea (mean/stdv) Nuclear Chromatin Texture Discrete Texture Hidnaarea  (mean/stdv) Nuclear Chromatin Texture Discrete Texture Lowdnaamnt  (mean/stdv) Nuclear Chromatin Texture Discrete Texture Meddnaamnt  (mean/stdv) 115  Appendix E (continued from page 114) Phenotypes Category Sub-category QTP Feature Name Nuclear Chromatin Texture Discrete Texture Hidnaamnt (mean/stdv) Nuclear Chromatin Texture Discrete Texture Lowdnacomp  (mean/stdv) Nuclear Chromatin Texture Discrete Texture Meddnacomp (mean/stdv) Nuclear Chromatin Texture Discrete Texture Hidnacomp (mean/stdv) Nuclear Chromatin Texture Discrete Texture Mhdnacomp (mean/stdv) Nuclear Chromatin Texture Discrete Texture Low_av_dst (mean/stdv) Nuclear Chromatin Texture Discrete Texture Med_av_dst (mean/stdv) Nuclear Chromatin Texture Discrete Texture Hi_av_dst (mean/stdv) Nuclear Chromatin Texture Discrete Texture Mh_av_dst (mean/stdv) Nuclear Chromatin Texture Discrete Texture Lowvsmed_dna (mean/stdv) Nuclear Chromatin Texture Discrete Texture Lowvshigh_dna (mean/stdv) Nuclear Chromatin Texture Discrete Texture Lowvsmh_dna (mean/stdv) Nuclear Chromatin Texture Discrete Texture Low_den_obj (mean/stdv) Nuclear Chromatin Texture Discrete Texture Med_den_obj (mean/stdv) Nuclear Chromatin Texture Discrete Texture High_den_obj (mean/stdv) Tissue Architecture Voronoi Diagram Density Density  (mean/stdv) Tissue Architecture Voronoi Diagram VorArea VorArea (mean/stdv) Tissue Architecture Voronoi Diagram VorPeri VorPeri (mean/stdv) Tissue Architecture Voronoi Diagram VorRF VorRF (mean/stdv) Tissue Architecture Delaunay Triangulation DelNeiNb DelNeiNb (mean/stdv) Tissue Architecture Delaunay Triangulation NeaNei NeaNei (mean/stdv) Tissue Architecture Delaunay Triangulation DelNeaNei DelNeaNei (mean/stdv) Tissue Architecture Delaunay Triangulation Del3NeiDist Del3NeiDist (mean/stdv)   116  Appendix F. Plots of Mean of QTP Features for Tumour Nests, by Nodal Status             117  Appendix F. (continued from page 109)         QTP features of tumour nests that are significantly different between N0 (gray) and N+ (red) (n = 45). First row, left to right: Hi_av_dstMean, High_den_objMean, High_den_objStdv, Lowvshgh_dnaMean Second row, left to right: Lowvsmed_dnaMean, Lowvsmh_dnaStdv Third row, left to right: Del3NeiDistMean, Del3NeiDstStdv, DelNeaNeiMean, DelNeaNeiStdv Fourth row, left to right: Density, VorAreaMean, VorAreaStdv, VorPeriStdv The black dot indicates the mean, the error bars indicate standard error.  118  Appendix G. Plots of Mean of QTP Features for Tumour Nest Layers#1, #2, and #3, by Nodal Status  QTP features for layers #1, #2, #3 that are significantly different between N0 (gray) and N+ (red).A, chromatin discrete texture, B, chromatin fractal texture, C, morphology, D, optical density, E, chromatin run lengths texture, F,  Voronoi polygon area. The closed circles indicate the means. The closed circle indicates the mean; the error bars indicate standard error.  A. Nuclear Phenotypes, Chromatin Texture (Discrete Texture)         Top row, left to right: HighdnacompMean, High_den_objMean, High_den_objStdv, Low_av_dstMean, Lowvsmed_dnaMean Bottom row, left to right: MeddnacompMean, MhdnacompMean    119  Appendix G. (Continued from page 111)  B. Nuclear Phenotypes, Chromatin Texture (Fractal Texutre) C. Nuclear Phenotype, Morphology (Size / Shape)     Left to right: Fractal_dimenMean, Fractal_dimenStdv Freq_high_fftStdv     D. Nuclear Phenotype, Optical Density (Photometric)    Left to right: Od_kurtosisStdv, Od_skewnessMean, Od_varianceMean          120  Appendix G. (Continued from page 112)  E. Nuclear Phenotype, Chromatin Texture (Run Lengths)       Top row, left to right: Gray_level1Mean, Long_runs2Mean, Long_runs2Stdv, Run_length1Mean Bottom row: Run_length2Mean   F. Tissue Architecture, Voronoi Diagram   VorAreaStdv    121  Appendix H. Plots of Mean of QTP Features for Layer#1-2, by Nodal Status         QTP features for Layer#1-2 that are significantly different between N0 (gray) and N+ (red). Top row, left to right: Lowvsmed_dnaMean, Lowvshigh_dnaMean, High_den_objMean Bottom row, left to right: High_den_objStdv, Run_length1Mean The closed circle indicates the mean; the error bars indicate standard error.    122  Appendix I. Plots of Mean of QTP Features for Layer#2-3, by Nodal Status     QTP features for Layer#2-3 that are significantly different between N0 (gray) and N+ (red). Left to right: Lowvsmed_dnaMean, High_den_objMean. The closed circle indicates the mean; the error bars indicate standard error.   

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.24.1-0300166/manifest

Comment

Related Items