UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Cytometric detection of nuclear features associated with pre-malignancy or malignancy in human bronchial… Payne, Peter William 1997

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

Item Metadata

Download

Media
831-ubc_1997-251357.pdf [ 8.21MB ]
Metadata
JSON: 831-1.0088262.json
JSON-LD: 831-1.0088262-ld.json
RDF/XML (Pretty): 831-1.0088262-rdf.xml
RDF/JSON: 831-1.0088262-rdf.json
Turtle: 831-1.0088262-turtle.txt
N-Triples: 831-1.0088262-rdf-ntriples.txt
Original Record: 831-1.0088262-source.json
Full Text
831-1.0088262-fulltext.txt
Citation
831-1.0088262.ris

Full Text

Cytometric Detection of Nuclear Features Associated with Pre-Malignancy or Malignancy in Human Bronchial Specimens Peter Will iam Payne R.T. (Cytology), Canadian Society of Laboratory Technologists, 1987 B.Sc. (Cell Biology), The University of British Columbia, 1989 A THESIS SUBMITTED IN PARTIAL F U L F I L L M E N T OF THE REQUIREMENTS FOR THE D E G R E E OF DOCTOR OF PHILOSOPHY in THE F A C U L T Y OF G R A D U A T E STUDIES DEPARTMENT OF P A T H O L O G Y We accept this thesis as conforming to the required standard. THE UNIVERSITY OF BRITISH C O L U M B I A July 17, 1997 COPYRIGHT PETER W I L L I A M P A Y N E , 1997 In presenting this thesis in partial fulfilment of the requirements for an advanced degree at the University of British Columbia, I agree that the Library shall make it freely available for reference and study. I further agree that permission for extensive copying of this thesis for scholarly purposes may be granted by the head of my department or by his or her representatives. It is understood that copying or publication of this thesis for financial gain shall not be allowed without my written permission. Department of The University of British Columbia Vancouver, Canada DE-6 (2/88) Abstract Approximately 600,000 people in the developed world die each year from lung cancer. The five year survival rate for lung cancer has remained at 13% for the past 30 years. Currently, Stage I Non Small Cell Lung Cancer five year survival rates have been reported at 76%, Stage II 25%,, Stage Ilia and Illb less than 7%, and with Stage IV, five year survival is rare. About 80% of lung cancers are not clinically detected until they are no longer localized. It is assumed that i f more people had their lung cancers detected in Stage I rather than in more advanced stages, mortality would be reduced. The hypothesis was that the sensitivity of sputum cytology for detecting early lung cancers can be improved by exploiting the phenomenon of malignancy associated changes (MACs) by a quantitative measurement of changes in D N A distribution. In patients with lung cancer, it was demonstrated that in bronchial biopsies diagnosed as normal by conventional methods, M A C s can be detected in approximately 86% of the biopsies. M A C value did not correlate with the distance from the tumour. This M A C expression returned to normal levels after tumour resection. In historical sputum samples it was possible to detect M A C s in approximately 80% of the samples including those with no diagnostic cells by conventional criteria. These M A C s were detected approximately one year or more before cancer was detected either clinically or by x-ray. M A C expression was shown to be independent of the number of acute inflammatory cells present in the sputum. It was further demonstrated that M A C s decrease significantly after successful surgical resection, but persist in patients who develop a recurrence or metastasis within three years after surgery. Payne i i Table of Contents Abstract : ii List of Tables iv List of Figures '• v List of Abbreviations vi Nomenclature vii Acknowledgements viii INTRODUCTION 1 W H O L U N G CANCER CLASSIFICATION 3 STAGING 5 MALIGNANCY ASSOCIATED CHANGES (MACs) 18 PLANNED EXPERIMENTS 21 MATERIALS AND METHODS 22 SPECIMEN STAINING AND FIXATION 23 Staining 23 Fixation 26 Acid Hydrolysis 37 Imaging Equipment 39 HISTOLOGY SPECIMENS 41 Specimen Acquisition 41 MACs in Bronchial Biopsies 45 Bronchial Brushings 48 SPUTUM 49 Mayo Clinic Slides 49 Tokyo Samples 55 Discriminant Functions 57 RESULTS 60 1. OBSERVER VARIABILITY 60 2. M A C s IN BRONCHIAL BIOPSIES 67 Distance Effect on MACs in Bronchial Biopsies 72 3. M A C s IN BRONCHIAL BRUSHINGS 74 Results 74 4. M A C s IN SPUTUM: M A Y O CLINIC SLIDES 76 Results 76 Inflammation Effect..: 82 5. M A C s IN SPUTUM: TOKYO SAMPLES 84 Results : 84 DISCUSSION 86 Previous MAC work in Sputum 89 M A C s IN BRONCHIAL BIOPSIES 91 M A C s IN BRONCHIAL BRUSHINGS 96 M A C s IN SPUTUM: M A Y O CLINIC SAMPLES 97 MAC IN SPUTUM: TOKYO SPECIMENS 102 MAC Mechanism 103 SIGNIFICANCE OF M A C s AS A DIAGNOSTIC TOOL 105 \ iii Payne List of Tables T A B L E 1: W H O L U N G TUMOUR STAGING 6 T A B L E 2: W H O TNM 7 T A B L E 3: BRONCHIAL BRUSHINGS; PATIENTS AND SLIDES 48 T A B L E 4: MACs IN SPUTUM. SUBJECTS AND SLIDES 52 T A B L E 5: MACs IN SPUTUM. SLIDES BY DIAGNOSIS 54 T A B L E 6: T O K Y O PATIENTS' TUMOUR STAGE AND HISTOLOGY 55 T A B L E 7: T O K Y O SPUTUM. NUMBER OF SLIDES 56 T A B L E 8: DIAGNOSES BY CATEGORY FOR THREE PATHOLOGISTS 62 T A B L E 9: INTER-PATHOLOGIST AGREEMENT RATES 64 T A B L E 10: EXAMPLE OF LARGER INTER-PATHOLOGIST DISAGREEMENT 65 T A B L E 11: NEGATIVE DIAGNOSIS VS. MAJOR DISAGREEMENT 66 T A B L E 12: BIOPSY CLASSIFICATION USING M E A N MAC VALUE 70 T A B L E 13: BRUSHINGS SLIDE BY SLIDE CLASSIFICATION USING M E A N MAC VALUE 75 T A B L E 14: CLASSIFICATION MATRIX. HUMAN SELECTION 80 T A B L E 15: CLASSIFICATION MATRIX. MACHINE SELECTION 80 T A B L E 16: SLIDE BY SLIDE AND C E L L BY C E L L CLASSIFICATION FOR BRONCHIAL BIOPSIES 93 iv Payne List of Figures FIGURE 1: SPUTUM SENSITIVITY vs. TIME 14 FIGURE 2: SPUTUM SPECIFICITY VS. TIME 14 FIGURE 3: M A C EXPRESSION 19 FIGURE 4: IOD VERSUS NUCLEAR AREA (PIXELS). PERIPHERAL BLOOD LYMPHOCYTES IN P B S 29 FIGURE 5: IOD VERSUS NUCLEAR A R E A (PIXELS). PERIPHERAL BLOOD LYMPHOCYTES IN FORMALIN 30 FIGURE 6: IOD VERSUS NUCLEAR A R E A (PIXELS). PERIPHERAL BLOOD LYMPHOCYTES IN 95% ETHANOL 31 FIGURE 7: IOD VERSUS NUCLEAR A R E A (PIXELS). PERIPHERAL BLOOD LYMPHOCYTES IN 75% ETHANOL 32 FIGURE 8: IOD VERSUS NUCLEAR A R E A (PIXELS). PERIPHERAL BLOOD LYMPHOCYTES IN 50% ETHANOL 33 FIGURE 9: IOD VERSUS NUCLEAR AREA (PIXELS). PERIPHERAL BLOOD LYMPHOCYTES IN SEDFLX (50% ETHANOL PLUS 2% POLYETHYLENE GLYCOL, A.K.A. SACCOMANNO'S FIXATIVE) 34 FIGURE 10: HYDROLYSIS CURVE FOR SPUTUM IN SEDFIX: EPITHELIAL NUCLEI AND LYMPHOCYTE NUCLEI. 38 FIGURE 11: IMAGING SYSTEM STABILITY. IOD vs. TIME 40 FIGURE 12: NORMAL BRONCHIAL BIOPSY FROM NORMAL PATIENT 46 FIGURE 13: NORMAL BRONCHIAL BIOPSY FROM NORMAL PATIENT. IMAGED NUCLEI 47 FIGURE 14: MACs IN BRONCHIAL BIOPSIES BY PATIENT GROUP 69 FIGURE 15: MAC VALUE vs. DISTANCE FROM TUMOUR 73 FIGURE 16: MAC SCORE vs. INFLAMMATION SCORE 83 FIGURE 17: ROC CURVE EXAMPLES 98 FIGURE 18: ROC CURVE FOR MACs IN M A Y O CLINIC SPUTUM 99 Payne v List of Abbreviations SCLC Small Cell Lung Cancer N S C L C Non-Small Cell Lung Cancer D N A Deoxyribonucleic acid R N A Ribonucleic acid C C D Charge-coupled device H & E Haematoxylin and Eosin ROI Region of Interest M A C Malignancy Associated Change IOD Integrated Optical Density ROC Receiver Operator Characteristics CIS Carcinoma in situ PCR Polymerase Chain Reaction W H O World Health Organization NCI National Cancer Institute SEER Surveillance, Epidemiology and End Results Reporting of the NCI L O H Loss of heterozygosity L S D Least Significant Difference Payne Nomenclature Bronchial Specimens: Applied to both tissue acquired during bronchoscopy, and to cytological specimens which originated in whole or in part from the bronchial epithelium whether acquired by brushing, washing, or sputum cytology. Malignancy Associated Change (MAC): Sub-visual or nearly sub-visual changes to the chromatin arrangement of visually normal nuclei; the changes are correlated to the presence of a tumour in the patient. Pre-malignant: Visually apparent. Used interchangeably with dyskaryosis in cytological specimens or dysplasia in histological specimens. Has no reference to the biological behavior of a lesion unless explicitly stated. Dysplasia: Visually apparent. Refers to the terminology applied to the light microscopic appearance of a tissue section showing abnormal growth patterns and/or abnormal looking cells. May be broken down into various degrees of dysplasia. For example, a 'moderately dysplastic lesion' may ultimately progress, persist or regress. Atypia: Visually apparent. Atypia is used when a cited article uses this term rather than dyskaryosis or dysplasia. Bronchial Dysplasia: Visual disturbances of varying degrees in the bronchial epithelium as seen in adults. This term is distinguished from bronchopulmonary dysplasia, a chronic lung disease. Sensitivity: (true positive)/(true positive + false negative) x 100% Specificity: (true negative)/(true negative + false positive) x 100% Payne vii ACKNOWLEDGMENTS I would like to thank Dr. Branko Palcic without whom the opportunity for me to pursue this line of study may never have arisen. Dr. Jean LeRiche provided training, experience and focus on clinical relevance that cannot be surpassed. Dr. Stephen Lam's unending energy kept me going as the work piled higher. Dr. Calum MacAulay, Dr. David Garner, Dr. Alexei Doudkine and Yvonne Zheng were always available for practical guidance on analysis of data, the day-to-day problems of equipment, chemicals and general chaos. Dr. Norihiko Ikeda of Tokyo Medical College Hospital was extremely interested in this project and helped by acquiring many bronchial brushing specimens. Also from Tokyo Medical College Hospital Dr. Toichiro Katsumi worked on all phases of the M A C s in sputum before and after surgery portion of this thesis; from acquiring the specimens to reviewing cell images. I would also like to thank Paul Lam for his tremendous help in handling the staining and de-staining of the large number of slides. Last but not least, I thank my wife Lisa, and children Simone and Spencer. Payne viii Introduction It is estimated that in Canada there were 20,000 new cases of lung cancer diagnosed, and 16,800 lung cancer deaths in 1995 (1), while in 1997 in the United States there were an estimated 178,100 new cases of lung cancer and 160,400 deaths. In both countries it is also the leading cause of cancer death (1,2). It is estimated that world-wide there are 900,000 cases annually and it is projected that by the year 2000 the number of cases worldwide wil l exceed two million (3,4). In North America the number of lung cancer deaths exceeds the total number of cancer deaths for cervix, larynx, brain, ovary, pancreas, uterine body, stomach and melanoma combined. It is the leading cause of cancer death for both men and women. There are so many cases of lung cancer that a 2% increase in the lung cancer cure rate for North America would save more life-years than a 100% cure for Hodgkin's disease (5). The five year survival rate for Canadian males with lung cancer is approximately 15% for all ages and 20% for females of all ages, with survival decreasing with age. In the United States the five year survival rate is 13% for all lung cancer patients of all ages. Despite numerous advances in diagnostic aids, surgical, and radiotherapeutic treatments, the overall five year survival of lung cancer patients has remained essentially unchanged for thirty years (6,7). There have been advances in the chemotherapeutic approaches to treating lung cancers, most notably in the small cell lung cancer variant, but this has not affected the five year survival rate as the median survival for small cell lung cancer is only ten months with treatment (8,9). 1 Patient survival is closely associated with the extent of the disease as determined by staging as well as by the histological sub-type. Lung cancer can be broadly grouped into Small Cell Lung Cancer (SCLC) and Non Small Cell Lung Cancer (NSCLC). For SCLC the five year survival rate is 5.2%. When broken down into extent of disease at diagnosis, the five year rate for localized SCLC is 18.6%, for regional disease 9.5% and with distant metastases 1.7%. For NSCLC localized disease five year survival is 49.9%, for regional disease 18.5%, and for distant metastases 1.8% (7). Unfortunately, only 16% of N S C L C and 7% of SCLC are localized at the time of diagnosis. Since survival rates are typically reported separately for different lung cancer sub-types and different stages a further discussion of classification methods is warranted. These methods include those outlined by the World Health Organization (10, 11). Payne 2 WHO Lung Cancer Classification The major histological classifications of malignant epithelial lung tumours: 1. Small cell carcinoma a. Oat cell carcinoma b. Intermediate cell type c. Combined oat cell carcinoma 2. Squamous cell carcinoma (Epidermoid carcinoma) Variant: a. Spindle cell (squamous) carcinoma 3. Adenocarcinoma a. Acinar adenocarcinoma b. Papillary adenocarcinoma c. Bronchiolo-alveolar carcinoma d. Solid carcinoma with mucus formation 4. Large cell carcinoma Variant: a. Giant cell carcinoma b. Clear cell carcinoma 5. Adenosquamous carcinoma 6. Carcinoid tumor 7. Bronchial gland carcinomas a. Adenoid cystic carcinoma b. Mucoepidermoid carcinoma c. Others 8. Others The reported incidence of each of these types varies from 30-40% for squamous, 20-45% for adenocarcinoma, 9-20% for large cell carcinoma, and 16-26% for SCLC (12, 13, 14, Payne 3 15). Squamous cell carcinoma has been reported as the most common form of lung cancer, although this is not true for all populations (17). There is also an increase in the incidence of adenocarcinoma in North America (13, 169). The present staging and treatment of squamous, adeno, and large cell carcinomas are essentially the same; hence their grouping as NSCLC. SCLC tends to be advanced at clinical presentation and receives a different treatment protocol hence its separation in classification. Payne Staging The T N M staging scheme is based on three main parameters: T is a description of the primary tumor, N is based upon regional lymph node status, and M describes the presence of distant metastases if any. These three parameters can be assessed by a number of procedures including physical examination, imaging, endoscopy, and surgical exploration. Based upon the T N M status, a cancer can be placed into a Stage. A summary of the staging system is provided in Tables 1 and 2. Stage 0 for example is carcinoma in situ or TisNOMO, while Stage I is a tumor with no nodal involvement or metastases and may or may not extend into hilar regions or invade the visceral pleura or have partial atelectasis. Stage II is the same as Stage I but with local nodal involvement. Stage Ilia is a complex mixture T N M where the tumor may be limited but have more extensive nodal involvement which does not extend to the opposite lung nor does it have distant metastases, or it may be a larger tumor involving local structures such as the chest wall or diaphragm with or without nodal involvement. Stage nib is similar but may involve the heart or great vessels or other critical structures such as the trachea. Stage IV is a cancer which has distant metastases. Payne 5 Table 1: WHO Lung Tumour Staging Stage Grouping Occult T X NO MO Stage 0 Tis NO MO Stage I T l NO MO T2 NO MO Stage II T l NI MO T2 NI MO Stage ILIA T l N2 MO T2 N2 MO T3 NO, N I , N2 MO Stage IILB Any T N3 MO > T4 Any N MO Stage IV Any T Any N M l 6 Payne Table 2: WHO TNM T N M Definitions T X Positive Cytology T l Tumour less than or equal to 3 cm T2 Tumour greater than 3 cm/ extends to hilar region/invades visceral pleura/partial atelectasis T3 Tumour extends to chest wall, diaphragm, pericardium, mediastinal pleura, total atelectasis T4 Tumour involves mediastinum, heart, great vessels, trachea, oesophagus, malignant effusion N l Peribronchial, ipsilateral hilar N2 Ispilateral mediastinal N3 Contralateral mediastinal, scalene or supraclavicular MO No distant metastasis M l Distant metastasis Using this system and currently available treatments Stage I N S C L C five year survival rates as high as 76% have been reported, Stage II 25%, Stage Ilia and Illb less than 7%, and with Stage IV, five year survival is rare. SCLC is typically reported only as 'limited' or 'extensive' stage. Limited stage SCLC has a two year survival rate of approximately 40%, and extensive stage disease is less than ten months (7, 8, 9). Payne 7 There are a number of strategies available for attacking the problem of lung cancer. One strategy is to reduce its incidence by identifying and limiting the causes of the disease either through reduced exposure or chemopreventive intervention (18). Another method is to improve the method of treatment for the disease. Several major causes of lung cancer have been identified. The association of lung cancer with certain occupations has long been known (19). Among the documented occupational lung carcinogens are arsenic (smelters), asbestos (shipyard workers), beryllium (some electronics), chromium (pigment), hydrocarbons (coal gas workers), nickel (refiners), radiation (uranium miners) (20, 21, 22, 23, 24, 25, 26). Recognition of these work place hazards has lead to efforts to prevent or reduce exposure to them. Additionally, the severity of atmospheric pollution has also been positively correlated with an increase in lung cancer rates (27). It is clear however that the major cause of lung cancer in the 20th Century is cigarette smoking. Prior to this time lung cancer was considered a particularly rare disease: "...clusters of lung cancer were distinctly unusual at the turn of the century, and physicians training in those early decades were still called to the postmortem table to observe an unusual disease process, lung cancer." (28). and "...there is nearly complete consensus of opinion that primary malignant neoplasms of the lung are amongst the rarest forms of disease." (29). Payne 8 As far back as 1665 it was noted that distilled tobacco oil would quickly ki l l a cat (30). By the 1940s and 1950s more scientific works began to note that smokers had a greater risk of lung cancer than non-smokers (31, 32). It was demonstrated shortly thereafter that cigarette 'tar' could produce tumours in mice (33). Similar evidence continued to mount and culminated with two major reports, the first by the Royal College of Physicians in 1962, and the second by the U.S. Surgeon General in 1964 which made the conclusion that smoking was associated with lung cancer and other diseases (34, 35). Since that time many other reports have been published that have established several components of cigarette smoke as carcinogens including but not limited to; benzene, 2-naphthylamine, 4-aminobiphenyl, and polonium-210. These components are Category 1 carcinogens meaning that they are confirmed human carcinogens as determined by the International Agency for Research on Cancer of the World Health Organization. Many others have been identified as Category 2a which means that there is sufficient evidence in animals and some evidence in humans or unequivocal evidence in mammalian cells. It is estimated that more than 90 percent of pulmonary carcinomas occur in heavy smokers based upon the change in the lung cancer mortality rate for males from 5/100,000 to 79/100,000 between 1930 and 1995 (36). This change is associated with the increase in smoking rates from 1900 to present. In 1900 the per capita consumption of cigarettes was 54 and had risen to a peak of 4,345 in 1963 (37). A n obvious strategy to control lung cancer is therefore to reduce smoking. Many efforts have been made to reduce smoking and have had some success, reducing the per capita smoking rate to about 2700 in 1990 (Smoking Prevalence and Lung Cancer Death Rates NTH, 1991). Unfortunately recent Payne 9 data shows that this decrease has leveled off and that 28% of men and 25% of women still smoke (2). In fact some demographic groups, in particular young females are smoking more. While this discussion is limited to North America, smoking is on a dramatic increase in developing nations. United States exports of tobacco have increased 275% between 1983 and 1995. In particular, exports have increased to Japan by 800%, and to South Korea and to countries of the former Soviet Union by 300% (2). Significant risk of lung cancer persists for long terms smokers even after quitting. Approximately 50% of lung cancer patients in North America are former smokers and 30% of them have quit smoking for five years or more (193). It is apparent that lung cancer due to smoking and industrial exposure wil l remain a major health concern for many years and no significant fall in incidence is expected for many years (38). A n alternative to removing the causes of lung cancer is to improve the treatment. With a five year survival rate of about 13%, it is clear that existing treatments and cancer control programs are inadequate. In treatment there are three main modalities: surgery, radiotherapy, and chemotherapy. These therapies can be used in various combinations. For the most part, surgical treatment has been the treatment for early stage 'resectable' lung cancers but combination chemotherapy and adjuvant therapy with surgical resection shows promise although of borderline statistical significance (38). The goal of any therapy is a complete cure. Regardless of the therapy chosen the best survival rates are consistently achieved when the cancer is detected at an early stage. This is true both of N S C L C and of SCLC. With NSCLC the best five year survival is for Payne 10 patients with Stage 0 or Stage I, T1N0M0 disease and the worst is for patients with metastatic disease (39). With SCLC the survival rate from the time of diagnosis is less than for NSCLC, but patients with limited disease survive significantly longer that patients with extensive disease. Although there are new treatment regimes constantly being developed and some have promising early results, it is clear that detection of more lung cancers at an early stage will have a significant impact even with present treatments. Even as new and better treatments become available, any new treatment would likely benefit i f used on a less advanced disease. Detection of lung cancer can take many forms. The most common is when a patient presents with symptoms. Patel and Peters report that initial symptoms and signs in patients are: cough 75%, dyspnea 60%, chest pain 45%, hemoptysis 35%, and other pain 25% (40). Lung cancer can be suspected based solely upon clinical presentation. A common presentation is an intractable productive cough, typically in long time heavy smokers or persons in occupations associated with an increased incidence of lung cancer. When a patient is first suspected of having lung cancer due to symptoms, their five year survival is approximately 5% (103). This method was the standard until the use of roentgenograms became available. The first trials on early detection of lung cancer were carried out during the 1950s with the Philadelphia Pulmonary Neoplasm Project, The South London Study and the North London Screening Program which used high dose photofluorgrams for detection Payne 11 (41, 42, 43). The sensitivity of the test was approximately 60% with little impact on improving patient survival as only 20% of the cancers detected were resectable. In the 1960s the American Cancer Society and the Veterans Administration reported results of more modern, low dose x-rays and sputum cytology. The x-ray method showed a 42% sensitivity while the sputum cytology was just 21% sensitive. However, there was little overlap between the cancers detected by the two methods such that the combined methods had a sensitivity of about 60% (44). In the 1970s, methods for detecting early lung cancer were explored by the U.S. NCI. Three participating centers, the Mayo Clinic, the Memorial Sloan-Kettering Institute, and the Johns Hopkins University each enrolled approximately 10,000 smoking males over the age of 45 years (45, 46, 47). Depending on the study group, the participants received either regular x-rays or regular x-rays plus sputum cytology. The sputum cytology had a sensitivity of 22-49% and the x-rays showed about 45% sensitivity. The specificity of the tests was not stated, but in a review of the published results Eddy estimated that the false positive rates for the Johns Hopkins patients was 0.14% and 18% for sputum and x-ray respectively (168). There was no improvement in the mortality rate of the patients. In the 1980's the Osaka Lung Project enrolled about 19,000 participants who received x-ray and sputum screening for lung cancer (48). The sputum cytology had 32% sensitivity (97.2% specificity), with most of the detected cancers being central lesions. Unfortunately only 10 cases were detected by sputum alone, all others being detected by x-ray or x-ray plus sputum. Payne 12 A randomized prospective study of lung cancer detection was started in Czechoslovakia in 1976 and reported in 1985 (49). This study enrolled 6364 high-risk men. The study compared semiannual radiologic and sputum screening to screening at a three year interval and to no screening. The screened groups had more early stage disease detected and better 5-year survival rates than the unscreened group, but there was no difference in survival rates between the two screened groups. In 1987, Ebeling and Nischan reported on an x-ray tuberculosis screening programme that was evaluated for its effect on lung cancer mortality. Their conclusion was that the screened individuals did not have a reduced risk of dying from lung cancer as compared to those who were not screened (166). The sensitivity and specificity of sputum cytology has been reported by many investigators (50-102). The various reported sensitivities are summarized in Figure 1 as sputum sensitivity versus time. The figure illustrates that there is no obvious trend of improvement in the sensitivity of sputum cytology. The range of sputum sensitivities is from 22% to nearly 100%. One might expect a steady increase in the sensitivity as experience with sputum cytology is gained and implemented into cytology training programs. Payne 13 Figure 1: Sputum Sensitivity vs. Time > c 100 90 80 70 60 50 40 30 20 10 0 • • * • • • • • • • t • • • • • *< • • m A S • i \ *• • • • • • 1930 1940 1950 1960 1970 1980 1990 2000 Year of Report For the reports which include the specificity of sputum cytology, the specificity value was very high (Figure 2) . Figure 2: Sputum Specificity vs. Time - • • » 1 • • • * • • w 1930 1940 1 950 1 960 1970 1980 1990 2000 Year of Report Payne The average sensitivity and specificity from a review of the literature was 64.5% and 97.9% respectively (103). One possible obstacle to improving the sensitivity of sputum is the present classification system. The present system is based upon years of careful observation of patients who developed lung cancer and is derived partially from similar observations in uterine cervix (104, 105). Individuals are trained to recognize abnormal cells and place them into discrete categories. These categories are often based upon those developed by Saccomanno which include seven categories: i) Normal, ii) squamous cell metaplasia, iii) squamous cell metaplasia with mild atypia, iv) squamous cell metaplasia with moderate atypia, v) squamous cell metaplasia with marked atypia, vi) carcinoma in situ, and vii) invasive carcinoma (105). Payne 15 To illustrate the subjective nature of the classification system Saccomanno's definition of squamous metaplasia with moderate atypia is provided: 1) Cells vary moderately in size; some are smaller than in mild metaplasia. 2) Nuclei vary significantly in size. 3) Nuclear to cytoplasmic ratio varies moderately. It may be higher or lower than normal. 4) Nuclear material is fine and powdery in most areas, but nuclear masses are abundant, particularly along the nuclear membrane. 5) Nuclear lobulations, crevices, and nodules are present. 6) Cytoplasm may be basophilic, but acidophilia predominates. 7) Cells usually occur in sheets, but an increase in single cells is found. Although based upon extensive observation and with reasonable correlation to histology, classification is difficult to repeat as there is room for variance in interpretation based upon training and experience. This variability in sputum cytology has been studied extensively and demonstrated that agreement is good for the normal and malignant categories but poor for the intervening categories (106, 107, 108). Among the reasons for the interobserver variability in categorizing specimens is that observers place emphasis on different features. This may also account for the difference in reported spontaneous regression of lesions rate which vary from 20% to 80% spontaneous regression for marked atypia (109, 172). Furthermore there are studies which indicate Payne 16 that the ultimate behavior of the lesion cannot be predicted by the present classification (110, 111, 112). In summary, the low sensitivity of sputum for detecting lung cancer is due at least in part to the subjectivity of the classification system. It would be advantageous if a test could be devised which could 1) take advantage of the ease of acquiring a sputum sample, 2) not require the presence of classical 'diagnostic' cells, and 3) be assessed objectively. It was believed that such a test could be developed to increased the sensitivity of the sputum test. Payne 17 Malignancy Associated Changes (MACs) The phenomenon of Malignancy Associated Changes (MACs) has been observed . for many years. It may have first been described by Gruner in 1916, but the term is most commonly associated with the work of Nieburgs (113, 114). M A C s were first described by Nieburgs in the late 1950s in buccal mucosa. They have been described by several authors since, but the reproducibility of the observation had been poor until the advent of a method of objectively measuring the M A C change (115, 116, 117, 118, 119, 120, 121, 122, 123, 124, 125, 126, 127, 128). M A C s are described as subtle changes in the chromatin arrangement of normal cells which are in the vicinity of a tumor (Figure 3). Whether these changes are a response to factors released by the neighbouring tumour or due to genetic damage has not been investigated. Payne 18 Figure 3: MAC Expression ooocP boooc Normal Cell In Tissue Nearby Cancerous Cells Nearby Dysplastic Cells DNA Distribution: Normal DNA Distribution: MAC. Sub-Visual M A C s have been described in a number of tissues including breast, colon, oral mucosa, and uterine cervix (114, 120, 121, 122). Nieburgs claimed that it was possible to detect i f there was any malignancy in the body by observing changes in white blood cells, but this was evaluated subjectively and has not been repeated (115). Vetrani employed the subjective definition of MACs developed by Nieburgs and applied them to sputum specimens in asbestos exposed workers with some success. Unfortunately Vetrani's study focused on detecting asbestos exposure and hence the true cancer sample size was very small including only one case of carcinoma in situ (118). A report by van Oppen Toth investigated MACs and compared the results of three studies demonstrating that as with dysplasia grades, M A C results had a low reproducibility (119). Payne 19 As digital cameras and computers became affordable it became possible to apply these technologies as a quantitative measure of histological and cytological specimens. Eventually these technologies were applied to detect MACs and met with success (120, 127, 123, 122, 121). To date quantitative investigation of the M A C phenomenon in lung histological or cytological specimens has not been investigated. Payne 20 Planned Experiments The hypothesis was that the sensitivity of sputum cytology for detecting early lung cancers can be improved by exploiting the phenomenon of malignancy associated changes (MACs) by a quantitative measurement of changes in D N A distribution. To test this hypothesis several preliminary experiments were conducted. The first of these experiments involved evaluation of the experimental setup including the specimen fixation and the stability of the imaging equipment used. Next bronchial biopsies and bronchial brushing were examined for the presence of MACs. The advantage of these specimen types over sputum was that it was possible to be certain that the cells examined were from the bronchial epithelium. The effect of the distance between the sampled site and the primary tumour on M A C s was examined. The final experiments were to test for M A C s on 1) a small set of sputum samples which had long term follow-up and, 2) on a larger set of sputum samples with shorter follow-up. Payne 21 Materials and Methods A n image cytometer was used to acquire high resolution digital images of cell nuclei (Oncometrics Imaging Corporation, Vancouver, Canada). These images were analyzed to detect subtle changes in D N A distribution of the normal bronchial cells of cancer patients versus normal bronchial cells from non-cancer patients. These changes in D N A distribution are reported as MACs. Several preliminary experiments were conducted. The first of these were to establish the limits of the specimen preparation methods and equipment. The next experiment was to determine i f MACs exist in cells derived from bronchial epithelium by testing for M A C s in bronchial biopsies. This required imaging of histologically normal bronchial biopsies from cancer and non-cancer patients. Since a histopathologist must choose the normal biopsies it was necessary to know the error of this step. The inter-observer variability of diagnosis has been previously reported for several tissue types but not for bronchial biopsies, hence this variability was investigated. Whether or not these M A C changes could be detected in cytological specimens was first investigated by examining bronchial brushing specimens. The first sputum experiment was to examine historical sputum from the Mayo Clinic portion of the National Cancer Institute Cooperative Early Lung Cancer Detection Program for the presence of MACs . The final experiment was conducted on more recent sputum specimens from Tokyo Medical College Hospital to gain some insight into the M A C mechanism. Payne 22 Specimen Staining and Fixation Many of the staining and digital imaging procedures are common to all experiments. The variations in technique are due to the specimen type or the purpose of each study. The specimen types used are tissue sections from bronchial biopsies collected during bronchoscopy and cytological specimens. This section details common elements. The materials and methods which are unique to each of the subsequent chapters or sections wil l be dealt with in those sections Staining The three staining procedures used in the following experiments are: 1. hematoxylin and eosin (H&E). 2. a modified Papanicolaou method (PAP). 3. a Feulgen reaction with thionin stain (F-T). H&E The stain used for studying inter-observer variability in diagnosing bronchial biopsies was haematoxylin and eosin (H&E) which is commonly employed for histological specimens. The haematoxylin portion of the H & E and PAP stains is the basophilic component responsible for the majority of nuclear staining, while eosin is an acidophilic stain which primarily stains cytoplasmic components (129). The exact Payne 23 mechanism of the hematoxylin stain has been controversial for some time. Bettinger and Zimmermann have greatly enhanced our understanding of this mechanism, but conflicting results remain (130). Additionally there are a number of different formulations and procedures for haematoxylin based stains (131, 132, 133). Regardless of the formulation used it is well established that haematoxylin is a non stoichiometric stain that forms an electrostatic complex with D N A which is sensitive to several conditions, such as pH of water-baths, which may vary widely between laboratories. PAP The staining used for the conventional diagnosis of the Mayo Clinic sputum specimens was a modified Papanicolaou (PAP) procedure (134). This method is among the most common and well established in conventional cytology as it provides good nuclear detail and cytoplasmic differentiation (131, 132). This popular stain has however been developed as a method intended for human visual observation and although H & E and PAP are very familiar to most pathologists they are not suitable for M A C analysis due to their variability and non-stoichiometric characteristics (135, 136, 137). F-T The staining method used for nuclear image analysis was a modified Feulgen procedure which is D N A specific (138). The central hypothesis of this thesis is that very subtle changes in the D N A distribution of normal cells may indicate a neighbouring Payne 24 tumor. The outcome of the investigations therefore depends on staining of D N A in a quantitative and reproducible manner. A stain which is specific for D N A was sought. A commonly used D N A specific stain is based upon the Feulgen reaction (138). This involves an acid hydrolysis step which causes de-purination of the D N A (removes adenine and guanine) leaving an aldehyde group exposed. The Schiff reagent then binds covalently to the sugar backbone of the D N A molecule. This stain results in a magenta color in the nucleus. This particular stain is sub-optimal i f one wishes to re-stain the specimen with a traditional counter stain to examine the cytoplasm of the cells. These counter stains typically contain eosin which stains the cytoplasm a pinkish color resulting in poor contrast between nucleus and cytoplasm. A second consideration is the sensitivity of the imaging system. The CCD present in the imaging system used is less sensitive in this range. There are other stains which could be used, one of the most commonly used being pararosanaline. After experimentation with several quantitative stains and techniques, Feulgen-Thionin was chosen for several reasons which were demonstrated by Tezcan (139): 1. It is specific for D N A . 2. The coefficient of variation of thionin is not statistically significantly different from that of pararosanaline. 3. Feulgen-Thionin fades more slowly than pararosanaline. It give a bluish color to the nucleus which approximates that of haematoxylin thus facilitating use of common counter stains as needed. Payne 25 Fixation Several experiments where conducted in the Cancer Imaging Laboratory by Paul Lam (M.Sc. Biochemistry) and Dave Garner (Ph.D. Chemistry) on the optimal hydrolysis times, temperatures and staining times to develop an extremely consistent staining procedure. Since most of these experiments were conducted for air-dried uterine-cervical smears, and sputum samples are normally fixed specimens, it was felt necessary to repeat the experiment for future fixation of cytological specimens. In addition it was desirable to know if the Saccomanno fixative used on the historical Mayo Clinic sputum had good characteristics for image cytometry. The first item investigated was the effect of fixation on nuclear staining with the Feulgen-Thionin stain. Many different fixation methods are employed by different laboratories for cytological samples, thus an experiment on thirteen of the more common fixatives, plus a reference in phosphate buffered saline were conducted. These tests were conducted on lymphocytes separated from peripheral blood of four different individuals by the Accuspin, System-Histopaque-1077 method (Sigma Diagnostics). The integrated optical density, which correlates with D N A amount, and the nuclear area were considered critical or 'base' features as many of the nuclear texture features used for M A C analysis are dependent on these features. Payne 26 It was felt that the minimum requirements of a fixative should be: • Little variation in apparent D N A amount from cell to cell. • Minimal cell clumping (single cell spread). • Minimal cell loss. • Consistent nuclear size for the same cell type. • Minimal variation from slide to slide. The fourteen common solutions tested were: 1. Phosphate buffered saline 2. 100% ethanol 3. 95% ethanol 4. 75%) ethanol 5. 50% ethanol 6. 50% ethanol plus 2% polyethylene glycol (SedFix™ or Saccomanno's Fixative) 7. 20% ethanol 8. 20% ethanol plus 1% acetic acid (Hussein's fixative) 9. 100% isopropyl alcohol 10. Formalin (37% w/v formaldehyde, neutral, phosphate buffered) 11. Paraformaldehyde 12. Histochoice (Amaresco) 13. Cytolyte (Cytyc Corporation) 14. Cytolyte plus PreserveCyt (Cytyc Corporation) Payne 27 Some representative plots of IOD versus nuclear area are presented in figures 4 to 9. The concentrations of the solutions are the final concentrations after lymphocytes were added. This was achieved by adding a small volume of concentrated lymphocytes to a large excess of fixative. Specifically, 500 pi of lymphocytes were added to 10 milliliters of test solution, mixed by vortexing and allowed to fix for 30 minutes at room temperature. The specimens were then centrifuged for 10 minutes at 1000 g and excess fixative removed. The cells were resuspended by vortexing and cell counts performed. Volumes were adjusted using each respective fixative such that 10,000 lymphocytes were expected on the final cytospin slides. The suspensions were then vortexed again and 0.5 milliliters were used for Cytospin™ (Shandon) preparations. Payne 28 Figure 4: IOD versus Nuclear Area (pixels). Peripheral blood lymphocytes in PBS. Payne Figure 5: IOD versus Nuclear Area (pixels). Peripheral blood lymphocytes in formalin. 40 DO 3&)0 3000 Payne 30 Figure 6: IOD versus Nuclear Area (pixels). Peripheral blood lymphocytes in 95% ethanol. 4000 ^ 3500 1 3000 1 j CO I** < CD 35* a$oo 2<XK) I 1500 X 100Q J 500 1 800 300 1000 I O D Payne Figure 7: IOD versus Nuclear Area (pixels). Peripheral blood lymphocytes in 75% ethanol. 4000 ^ 3S0Q 1 2500 < {g 2000 : CD J2J 1000 i $00 I Payne Figure 8: IOD versus Nuclear Area (pixels). Peripheral blood lymphocytes in 50% ethanol. Payne Figure 9: IOD versus Nuclear Area (pixels). Peripheral blood lymphocytes in SedFix (50% ethanol plus 2% polyethylene glycol, a.k.a. Saccomanno's Fixative). AQQQ 3$G0 3GQ0 to t AS o 3 2500 1 1500 Payne 34 Figure 4, lymphocytes in PBS, shows several different populations of nuclei. It can be noted that nuclei less than about 300 pixels in area, as outlined in box ' A ' , consistently show a positive slope. Other investigators have made this observation and have attributed this to either increased condensation of chromatin resulting in poor penetration of the stain, hydrolysis effect and/or a glare effect in which glare from small dark objects creates an artificially low IOD (154). Box B denotes nuclei which visually appear to be poorly preserved, enlarged and have ill-defined nuclear borders. Their generally low IOD suggests loss of nuclear material available for staining. Box C is characterized by two overlapping nuclei (doublets) which the image cytometer treated as one object. Objects which are plotted to the right of Box C are groups of three or more nuclei. The ellipse denoted by the ' D ' are lymphocyte nuclei which appear visually normal. Although there are a range of sizes, the IOD is very consistent. Figure 5 shows lymphocytes fixed in formalin. It is immediately obvious that the nuclei are considerably smaller in area, virtually all below 300 pixels. Although the nuclei appear visually very well preserved and with good nuclear detail this results in a range of IOD values for the same nuclei. This profile is also seen in Figure 6, 95% ethanol. In Figure 7, 75% ethanol fixation, the profile is closer to the minimum fixation requirements previously described. If nuclei smaller than 300 pixels are discarded, there is little variation in IOD between singlet nuclei, the size of the nuclei is fairly consistent, and there was very little cell loss. There remains notable cell aggregation. Payne 35 Figure 8, lymphocytes in 50% ethanol, looks similar to Figure 4, lymphocytes in PBS. The main exception is that the very poorly preserved nuclei seen in Box B of Figure 4 are not present. Figure 9 is lymphocytes in SedFix, which is a combination of 50% ethanol and 2% polyethylene glycol in water. This fixative is frequently referred to as Saccomanno's fixative. It was thus very encouraging to see that what is one of the most widely used fixatives for sputum has excellent characteristics for image cytometry; i f small nuclei are excluded, as with the previous fixatives, there is very little variation in IOD from nucleus to nucleus (coefficient of variation < 4%), there is little cell loss and there is minimal cell clumping. The standard deviation of the IOD values for lymphocyte nuclei between slides prepared from the four volunteers was also calculated. The lowest standard deviation for i the illustrated fixatives was that of SedFix (1, 4), the highest was for 75% ethanol (5, 4). As Saccomanno's fixative was used to fix the Mayo Clinic sputum slides and SedFix (chemically the same as Saccomanno's) has been shown to be a good fixative for lymphocytes as determined by the criteria outlined earlier, it was felt that lymphocytes could be used to normalize between the Mayo Clinic slides. Payne 36 Acid Hydrolysis The Feulgen reaction involves an hydrolysis step in concentrated hydrochloric acid. This step makes aldehyde groups available on the deoxyribose backbone of D N A to bind the thionin stain. This is an optimization process; too little hydrolysis and not all potential binding sites become available, too much hydrolysis and the D N A is degraded. Hydrolysis had been investigated previously under different conditions by different groups (173-176). Previous results on air-dried smears suggested that a 45 minute hydrolysis at room temperature was optimal. The hydrolysis experiment was repeated on SedFix preserved sputum. Acid hydrolysis was performed at room temperature in five minute steps for two hours as shown in Figure 10. The beginning of the plateau region is after 30 minutes. By two hours, the optical density of the epithelial cells begins to decline suggesting over hydrolysis. In the plateau region, lymphocyte IOD was approximately 10% less than epithelial IOD. This has been noted previously, but the cause is unclear (194, 195). It was decided that a forty-five minute hydrolysis is inside the plateau and was an acceptable time. Payne 37 Figure 10: Hydrolysis Curve for Sputum in SedFix: Epithelial Nuclei and Lymphocyte Nuclei. 0 20 40 60 80 100 120 Acid Hydrolysis Time (minutes) Payne 38 Imaging Equipment Cytology Specimens The samples were imaged on a Cyto-Savant™ (Oncometrics Imaging Corp., Vancouver) which was developed in conjunction with the Cancer Imaging Department of the British Columbia Cancer Research Centre (177-179). Slides were robotically loaded onto an motorized stage and automatically identified by a bar-code. The specimens were scanned using a high resolution digital camera mounted on an Olympus BH-2 microscope with a Nikon 20x PlanApo objective with a numerical aperture of 0.75. A 600nm +/- 10 nm filter was used as this corresponds to the region of peak absorption of the thionin nuclear stain. As the CCD camera used in the system has a pixel size of 6.8 um x 6.8 um, the effective pixel size with a 20x objective lens is 0.34 x 0.34 um. Individual nuclei were captured in an automated and unattended manner using unique focusing and segmentation algorithms as the robotic stage moves the slide under the objective. Nuclei were stored in an image file which is automatically named to match the slide's bar code. The cytometer stores the microscope stage coordinates for each object enabling the operator to revisit each nucleus in the computer file and examine it under the microscope. Periodically, a test was run overnight to check the imaging system stability. This involved having the cytometer re-image the same scene to produce a single image file of over 50 objects (lymphocyte nuclei). It then continued to re-image the same scene, but saved those images in sequentially numbered files. The data was then analyzed and a Payne 39 typical result is shown in Figure 11. The variance is on the order of half of a percent which corresponds to the bright field noise characteristics of the camera. Figure 11: Imaging system stability. IOD vs. Time. 120 00 100 00 80 00 D O 60 00 40 00 20 00 0 00 JL : M ean: ; Variance: 102.6 0.67 100 200 300 400 500 60O 70G 800 Time (M inutes) Payne Histology Specimens The histology specimens were imaged on a similar system. Due to the crowding of nuclei in the bronchial biopsy specimens, imaging could not be conducted in the fully automated mode; segmentation of nuclei had to be corrected by hand. Only nuclei which appeared to be single and non-overlapping were selected. Approximately 150-200 epithelial nuclei were collected per biopsy specimen and 50-150 lymphocyte nuclei were collected as an internal control. The second variation is that for the bronchial biopsies used in the malignancy associated changes study, a 1.25x projection lens was present in the light path resulting in a magnification of 25X, and an effective pixel size of approximately 0.27 microns. Specimen Acquisition Bronchial Biopsies The bronchial biopsies specimens are obtained during a bronchoscopy procedure. A typical scenario was that a flexible fiberoptic bronchoscope was inserted into the conducting portion of the lung and used to visually examine the bronchial epithelium either under white light or in combination with the Light Imaging Fluorescence Endoscope (LIFE) which increases the sensitivity of bronchoscopy to dysplastic lesions (165). When an area of interest was visualized by the bronchoscopist, biopsy forceps were passed through a channel in the bronchoscopic device and a small tissue biopsy was taken. The biopsy was then placed in a balanced saline solution and taken to the laboratory for processing. The specimen was fixed in buffered formalin (37-40% w/v Payne 41 formaldehyde in water, phosphate buffer) and embedded in paraffin. The paraffin block i was cut into 5 micron sections, placed on microscope slides, de-waxed, hydrated and stained with haematoxylin and eosin. The specimens were evaluated by a pathologist. Among the sources of error are a failure to biopsy significant areas of the bronchi, cell loss during processing, and failure to recognize diagnostic regions of the biopsy on histological examination. These errors are controlled largely by the skill of the operators of each of these steps. Sources of error for image cytometry include the consistency of the section thickness, fixation and staining. The fixation and embedding were performed by an automated tissue processor so that all specimens were handled the same. In the sections obtained during bronchoscopy, which are typically one millimeter or less, uniformity of section thickness within the specimen does not appear to be as much a problem as in larger section such as breast biopsies or biopsies taken from larger surgical specimens (181). Consistency between specimens is important asreported by Bacus who concluded that the most important aspect in consistent sections was correct angle of the microtome blade (180). The use of internal controls for section thickness is paramount and one must keep in mind the limitations of assessing cells with nuclei > 5um which have been cut into slices which are thinner than their smallest dimension. To reduce error in the histologic specimens used in this study all sections were cut on the same microtome by the same operator. Lymphocytes within the section were imaged as an internal sectioning control and peripheral blood lymphocytes from normal volunteers working in the laboratory were used as external staining control. Payne 42 Observer Variability Bronchial biopsy specimens taken during bronchoscopy in 1990 and 1991 were fixed in formalin, embedded in paraffin and cut at 5 microns on a Leitz 1512 Rotary Microtome. Section were stained with H & E . They were then assigned a diagnosis by a pathologist as per clinical procedure at the British Columbia Cancer Agency, Vancouver. After reviewing more than 200 pathology reports, 113 slides were then chosen from 48 patients. The choice of slides was based upon a clearly stated diagnosis and a range of diagnosis from normal to malignant. A standardized form was prepared based upon comments reviewed on the pathology report forms. The slides were chosen such that a minimum of eleven examples for each category on the form were present (CIS and Carcinoma treated as one category). The form contained the following categories: 1. Negative 2. Mi ld dysplasia 3. Moderate dysplasia 4. Severe/Marked dysplasia 5. Carcinoma in situ 6. Carcinoma. A sample of the form used is in Appendix A. A l l 113 slides in the study set were originally given a diagnosis which fit into one of these categories. Payne 43 The slides were randomized and placed into two boxes. The original slide identification was covered with a bar-code label. The boxes were circulated through three pathologists at the B.C. Cancer Agency, Vancouver along with blank report forms. The pathologists viewed the slides independently. The diagnoses were then compiled into a database and triple checked for transcription errors. ' Inter-observer variation was calculated by two methods. The first method was based upon the number of exact matches and matches within one category (107). The second method was by kappa statistic according to Fleiss, 1981, as calculated by the statistical software package B M D P (146, 147). The advantage of the kappa statistic is that it allows for some randomness in the agreement and disagreement rates. Payne 44 M A C s in Bronchial Biopsies Specimens For this study 152 bronchial biopsies from 82 patients were used. A l l of these biopsies were originally given a diagnosis of 'normal'. A l l the biopsies were reviewed by an experience pathologist to confirm the diagnosis, thus two pathologists agreed on a diagnosis of 'normal'. Sections were cut at 5 microns. The biopsies were divided into five groups based upon broncoscopic examination: 1. Subjects without cancer or dysplasia; 36 normal sections from 14 non-smokers and 27 normal sections from 14 current or ex-smokers. 2. Subjects with dysplasia but no cancer; 6 normal sections from 4 patients who had severe dysplasia at least one other biopsy site, 17 normal sections from 8 patients who had at worst a moderate dysplasia at least one other biopsy site, and 17 normal sections from 9 patients who had mild dysplasia at least one other biopsy site, and 2 normal sections from 2 subjects who had squamous metaplasia. 3. Subjects with Stage III invasive carcinoma; 14 normal sections from 12 patients (8 patients had squamous carcinoma, 2 had adenocarcinoma, 1 had small cell carcinoma, and 1 had large cell carcinoma). 4. Subjects with carcinoma in situ (CIS) or microinvasive cancer; 14 normal sections from 7 subjects. 5. Subjects with Stage I, resected non-small cell lung cancer; 19 normal sections from 12 patients. 45 Payne Imaging The imaging equipment has been described earlier. Images were collected in a semiautomated fashion in that the operator was responsible for checking and correcting segmentation of the nuclei while the system performed automated focusing. Only nuclei which did not appear to overlap and which appeared to be nearly whole nuclei in the opinion of the operator were selected for image collection. Images of approximately 250 normal bronchial epithelial cells and >40 lymphocytes per slide were collected. Approximately equal numbers of basal and columnar cells were collected. A sample image of a normal bronchial biopsy is provided in Figure 12. Figure 12: Normal Bronchial Biopsy from Normal Patient Figure 12 is a digital image of a portion of a bronchial biopsy diagnosed as normal by two pathologists. A 20x objective with a 0.75 numerical objective and a 1.25x projection lens were used to project an image onto the CCD camera with 6.8 x 6.8 pixel size. Magnification from cell to image on printed page is lOOOx. Payne 46 Figure 13: Normal Bronchial Biopsy from Normal Patient. Imaged Nuclei. Figure 13 is the same image as in Figure 12. The objects outlined in red are epithelial nuclei which would be collected for use in analysis. The object covered by a blue mask is a lymphocyte which would be used in the normalization procedure. Magnification lOOOx. Payne 47 Bronchial Brushings ( Brushing samples can be obtained from the respiratory tract during bronchoscopic examination. A small brush is inserted through a channel in the bronchoscope while the sampled area is visualized by the bronchoscopist. The brush is retracted and a smear is made on a microscope slide which is immediately placed in Saccomanno's fixative. The brush-smear was used for this experiment while the brush with any remaining sample was used for conventional cytology. The samples used for M A C analysis were those which appeared to be from normal appearing areas according to conventional white light bronchoscopy and LIFE bronchoscopy. Only samples with a conventional cytological diagnosis of 'normal' were used. This resulted in 79 samples from 74 patients a shown in Table 8. Table 3: Bronchial Brushings; Patients and Slides Normal Mild Moderate Carcinoma Invasive Recurrence Benign Severe in situ Carcinoma Patients 20 8 14 8 22 2 Slides 24 8 14 8 23 2 The normal and benign patients were combined to form the 'normal' training group, and the carcinoma in situ and invasive carcinoma patients formed the 'cancer' training group. The expected result was that the normal and cancer groups should have the best separation, while the intervening dysplasia groups should have intermediate separation. The recurrence patients should be clustered with the cancer group. Payne 48 Sputum Mayo Clinic Slides The sputum used in this investigation was from the Mayo Clinic portion of the National Cancer Institute Cooperative Early Lung Cancer Detection Program which ran during the late 1970's and early 1980's (45, 46, 47). This sputum was collected by either induction or the three-day pooled method. Specimens were prepared by the Saccomanno method and stained by the modified Papanicolaou method. The slides were then reviewed by a cytotechnologist and/or cytopathologist to assign a diagnosis. Sputum is the product which is expelled via the oral cavity after a deep cough. It is a product of all the passageways through which the airflow has traveled including the gas exchange portions and conducting portions of the respiratory tract including the oral cavity. A sputum sample can thus be composed of a variety of cellular and non-cellular material. The make up of these components determines the diagnosis; hence, acquiring an adequate sample is of paramount importance in the diagnosis. A n adequate 'deep cough' sputum is defined as one that contains pulmonary alveolar macrophages. The presence of these cells is an indicator that at least a portion of the sputum sample has come from deep in the respiratory tract (140). These cells often contain material which they have phagocytized which is frequently particles which were suspended in the air, hence the name 'dust cells'. Payne 49 The patient is asked to cough deeply and expectorate the product into a container. The container may contain any of a number of fixatives for preserving the cells or inactivating infectious agents. The sputum may be a single sample or pooled over several days. Previous investigators have demonstrated that a three day or more pooled sample can increase the sensitivity of the sputum examination (141, 142). For patients who have difficulty producing sputum, its yield may be enhanced by induction (143). Induction is typically by inhalation of an aerosolized balanced saline solution or sterile that helps to induce a deep cough. Agitation of the chest may be used to loosen cells and mucus. This agitation may take the form of some physical activity or of pneumatic chest oscillation devices. The sputum sample in a container must then be transferred onto microscope slides. This material is then spread between two or more slides and stained to produce a preparation suitable for microscopic examination. Sputum is a very mucoid specimen and a number of techniques are used to break up the mucus prior to preparing the slides in the belief that this produces a more representative sample. These methods include mechanical, enzymatic and chemical treatments all of which add time and cost to the preparation. A n example of a mechanical method is the Saccomanno method in which the sputum sample is placed in a blender and subjected to shearing forces (144). This breaks up the mucus to free some trapped cells. A criticism of this method is that although this increases the yield of squamous carcinoma cells, it may decrease the yield of other more fragile carcinoma cells such as those seen in SCLC (145). After the specimen is on the slide it is most frequently stained according to modified Papanicolaou Payne 50 staining procedures and then reviewed by a trained cytotechnologist or cytopathologist (131). A diagnosis is made according to the material seen. Some sources of false negative diagnoses are: 1. Inadequate sampling; a deep cough is not produced. 2. Inadequate sub sampling; diagnostic cells are not placed on slides. 3. Cell loss/destruction in processing. 4. Failure to see or recognize diagnostic cells. Sources of false positive diagnoses include misinterpretation of cells present particularly when degenerated, and plant cells which are contamination from the oral cavity. Kato has noted several plant cell types as particular sources of error, hence adequate cleaning of the oral cavity is required to reduce the sources of false positive results (140). The method of sputum acquisition, fixation, and preparation, used by the Mayo Clinic during the NCI trial was state-of-the-art circa 1975. One possible source of false negative specimens according to current knowledge is that fragile carcinoma cells such as SCLC, may be destroyed by vigorous blending during the Saccomanno method. Only specimens of patients with squamous cell carcinoma were used in the present work. In addition the sensitivity of sputum cytology reported by the Mayo Lung Project was just 22% which is one of the lowest reported in the literature compared to an average of 64.5 % (103). Payne 51 Specimens Three day pooled sputum collected by the Mayo Clinic during a long term controlled trial of radiological and cytological screening sponsored by the National Cancer Institute were used in this study. Subjects were recruited from 1971 to 1976 and screened every four months by both x-ray and sputum cytology and were followed for five years or more. Initially enrolled were 10,933 male outpatient smokers 45 years of age or older. After initial examination, 91 cancers were detected (prevalence); 9,211 men met criteria for continued screening (47). During the study, 366 lung cancers were detected (incidence) (26). We examined 73 slides from 20 different subjects. Nine patients (40 slides) were from the group who eventually developed squamous carcinoma of the lung. Eleven subjects (33 slides) were from the group who did not develop lung cancer during follow up (Table 8). Table 4: MACs in Sputum Subjects and Slides GROUP Number of Subjects Number of Specimens Developed Cancer 9 40 Did not develop Cancer 11 33 In the group that did not develop lung cancer, all 33 slides were originally diagnosed as having no cells diagnostic for carcinoma and were categorized as: i) negative (29), ii) negative with metaplastic cells or some slightly atypical cells (3), iii) Payne 52 moderately atypical cells present (0), and iv) markedly atypical cells present (1) (Table 9). In the group who did eventually develop squamous carcinoma, the 40 slides available were diagnosed as: i) negative (8), ii) metaplastic cells or some slightly atypical cells present (9), iii) moderately atypical cells present (7), iv) markedly atypical cells present (5), and v) squamous carcinoma cells present (11). A l l specimens were fixed in 50% ethanol and 2% polyethylene glycol, prepared by the Saccomanno method and smeared onto non-frosted microscope slides. Payne 53 Table 5: MACs in Sputum. Slides by Diagnosis Number of Sputum Slides in each Diagnostic Group. Only normal epithelial nuclei were used in the analysis. GROUP Negative or Slight Atypia Moderate or Marked Atypia Carcinoma Cells Present Developed Cancer 17 12 11 Did not develop Cancer 32 1 0 Staining: The specimens which had originally been stained with Papanicolaou stain were de-stained and then re-stained by the Feulgen-Thionin method. A l l staining was performed in one batch. Payne 54 Tokyo Samples Two groups of patients were selected from the files of the Tokyo Medical College Hospital. A l l patients had been diagnosed with non-small cell primary lung cancer between 1988 and 1994. The 'good prognosis' group consisted of 19 patients, aged 62 +/- 5 years, who had not developed any metastases nor recurrence after surgical resection with at least 30 months of follow-up (14 with over 60 months of follow-up). The 'bad prognosis' group consisted of 12 patients, 64 +/- 4 years of age, who did develop a recurrence and/or metastases within 36 months of surgical resection. The stages and histologic types of the tumours are summarized in Table 14. The two stage IV cases were not suspected pre-operatively; post-operative examination showed intrapulmonary metastasis. Table 6: Tokyo Patients' tumour stage and histology. Good Prognosis Group Bad Prognosis Group Stage I 16 2 Stage H 1 1 Stage m 2 7 Stage IV 0 2 Squamous carcinoma 8 3 Adenocarcinoma 10 8 Mixed Adeno-squamous 1 I Two sputum specimens were collected from each patient 2-4 weeks prior to surgery by the three-day pooled specimen method. Patients took home a plastic bag containing a modified Saccomanno fixative and coughed and expectorated into the bag over a three day period. The fixative modification was that 1% thymol (l-methyl-3-Payne 55 hydroxy-4-isopropylbenzyne) was added as an antiseptic. Patients who did not show signs of recurrence or metastasis typically had one follow-up sputum. The patients who did have recurrence had additional follow-up sputum as shown in Table 15. Table 7: Tokyo Sputum. Number of slides. Good Prognosis Group # Patients (# slides) Bad Prognosis Group # Patients (# slides) Pre-operative sputum 19 (37) 12 (24) Post-operative sputum 19(21) 12 (25) Slides were destained and restained according to the Feulgen-thionin method. The slides were scanned on the automated image cytometer set to collect 1500 normal epithelial nuclei. A l l images were reviewed by a cytotechnologist. Only normal epithelial nuclei in good focus and properly segmented were used in the analysis. Forward step-wise linear discriminant function analysis was used to create the M A C function as previously described. Payne 56 Discriminant Functions The method of discriminant function analysis and significance testing of such functions warrants commentary. Two group discriminant function analysis is analogous to multiple regression and is also known as 'Fisher linear discriminant analysis'. In a test of two normally distributed, randomly sampled groups a single feature might be used to discriminate between the groups. A simple test statistic, such as Student's t-test, can then be applied to determine if these groups are significantly separated by the examined feature. There are also many cases when multiple groups, and or multiple features are examined. In these cases, other analyses may be required such as analysis of variance, the Student-Newman-Keuls procedure, and Fisher' Protected Least Significant Difference (LSD) among others. Choice of the appropriate test depends on the characteristics of the samples and the assumptions made (158, 197 ). In discriminant functions several features may be combined into a single score used to separate two groups. These features may or may not be normally distributed, or a mixture of both. In addition, individual features typically are weighted to maximize the separation between the two initial groups (training groups) used to create the discriminant function. This creates a situation where some common tests cannot be applied to the training groups used to create the discriminant function, but can be used on another group or groups (test groups) which were not used to create the discriminant function. Payne 57 An artificial method of producing a test group when the available sample size is small and no actual test group is available is referred to as 'jackknife' analysis (also known as 'leave-one-out' analysis). In this method, an available sample is excluded from the training set used to produce the discriminant function. The discriminant function is then created from the remaining samples and applied to this single sample to test i f it is correctly classified. This can be repeated for all single samples in a step wise manner. The classification matrix for the case when all samples are included in creating the discriminant function is then compared to the jackknife results. Addition of more features would likely improve the classification matrix, but at the expense of reproducibility since these additional features take advantage of random effects in the training groups. The jackknife results are reported in this thesis. This creates a classification result which is not as good as would be obtained by the addition of more features, but is more likely to be repeatable in prospective studies. The test statistics which are suitable for multivariate analysis as applied to cytological data collected by image cytometry have been discussed by Bartels (196). The suggested tests are Hotelling's T 2 and Wilks' lambda. One of the assumptions for these tests, is equality of variance-covariance matrices. Bartels asserts that only extreme deviations affect these tests. Wilks' lambda has been calculated for all the analyses, and converted to an F value such that a p-value can be reported for the discriminant function. It is the experience of the author and other investigators that the jackknife classification is more informative of the behaviour of the discriminant function than the p-value. Payne 5 8 In the M A C s in bronchial biopsy section, there are six groups. The discriminant function is calculated from two groups, thus the remaining groups which were not used to create the function can serve in the capacity of test groups upon which several standard tests can be performed. The assumptions made in discriminant function analysis are: 1) normal distribution of the variables; this not a critical assumption (198). 2) homogeneity of variances and covariances. Only affected by extreme deviations. 3) no correlation between means and variances. 4) no matrix ill-conditioning; for example, when one variable is the sum of two or more of the other variables. None of the features used in this work are of this type. This is checked within the B M D P software by a 'tolerance' setting which wil l not allow the addition of features which are correlated beyond a certain tolerance. This was left at the default setting of 0.01. Payne 59 Results 1. Observer Variability Several of the previously mentioned sources of error can be reduced by careful sample acquisition and processing. This is achieved by good technique and training of the technicians. Pathologists are highly trained in assessing the cellular and organizational components of cytological and histological specimens. The technique has been developed over more than 100 years and countless specimens. In the final analysis however a decision must be made by an individual person based upon their training and experience. That person may ask for the interpretation of others to aid in the decision making process. The working definition of M A C s in this thesis is that they are sub-visual changes in epithelial cells which are traditionally classified as normal. Presently 'normal' epithelium or cells are defined by visual observation by a pathologist. No work on the interpathologist variability of diagnosis in bronchial biopsies has been previously published, thus a study was conducted to establish that variability. Payne 60 Results Pathologist A chose to reduce the number o f dysplasia' categories from three (Mild, Moderate, and Severe/Marked) to two (Minor and Significant). This pathologist believed that the 'Minor' category was roughly equivalent to combining the ' M i l d ' and 'Moderate' categories. As shown in Table 3 this seems to be true because when ' M i l d ' and Moderate' are combined for Pathologists B, and C, their number of 'Minor' dysplasias is approximately the same as for pathologist A. Pathologist A also chose to classify one biopsy as 'Unsatisfactory' while B and C did not. A l l three pathologists chose to add the category of 'Benign'. Payne 61 Table 8 shows the number and percentage of slides in each category for each pathologist. Table 8: Diagnoses by Category for Three Pathologists. Path Unsat Neg Ben Minor Sev CIS Cancer A 1 15 38 30 12 11 6 (0.9) (13.3) (33.6) (26.5) (10.6) (9.7) (5.3) B 0 23 33 35 8 10 4 (0.0) (20.4) (29.2) (31.0) (7.1) (8.8) (3.5) C 0 29 25 26 10 15 8 (0.0) (25.7) (22.1) (23.0) (8.8) (13.3) (7.1) Path = Pathologist, Unsat = Unsatisfactory specimen, Neg = Normal, Ben = Benign changes (hyperplasia or squamous metaplasia without atypia), Minor = Minor dysplasia, Sev = Severe dysplasia, CIS = Carcinoma in situ, Cancer = all types of carcinoma. Table 8 gives the number of slides given a particular diagnosis by each of the pathologists. The numbers in brackets are the percentage of slides in each category. Comparison of the three participating pathologists with the original pathologist was not considered appropriate for two reasons: 1. The original pathologist had the benefit of additional H & E stained slide(s). 2. The original pathologist had access to clinical information which has been demonstrated to bias the diagnosis of H & E stained specimens (148). Payne 62 The data was analyzed in two ways. The first was a full analysis of all categories including kappa statistics in order to elucidate the general reproducibility of the diagnostic method. The second analysis focused specifically on the reproducibility of the 'negative' category to determine if M A C analysis of bronchial biopsies could be conducted. All Categories analysis. The first form of this analysis is simply a count of all exact matches of diagnostic categories between all pairwise comparisons of the three pathologists. Additionally, a modified count such that a 'plus or minus one category' disagreement is scored as a match is assessed. This method counts 'Negative' and 'Benign' or 'Benign' and 'Minor' as a match, but not 'Negative' and 'Minor'. The one slide deemed unsatisfactory by pathologist ' A ' was excluded from the analysis as it seemed unreasonable to count 'Unsatisfactory' as a plus or minus one category match with any particular category; analysis was thus conducted on 112 slides. The second form of analysis was using the kappa statistic for all pairwise comparisons of the three pathologists. According to Fleiss, a kappa value below 0.4 is poor agreement, between 0.4 to 0.7 is good agreement, and above 0.7 is excellent agreement (146). The values given are based upon the categories given in Table 8, excluding the one 'Unsatisfactory' slide. Results are tabulated in Table 9. Payne 63 Table 9: Inter-Pathologist Agreement rates. A B C A B Exact match 51.8 +/- 1 Category match 90.2 kappa 0.391 C Exact match 42.0 43.8 +/- 1 Category match 83.0 80.4 kappa 0.294 0.315 The +/- 1 Category match rates and kappa values illustrated in Table 9 are consistent for similar results of other anatomic sites reported in the literature (182, 183, 184). The kappa values for all three pairwise comparisons are below 0.4 and are thus poor agreement as described by Fleiss. The high agreement rate of the +/- 1 Category scores (80-90%) contrasts with the low kappa scores. This is explained by the insensitivity of the +/- 1 Category scoring method to very large variations. The diagnosis of 8 of the 112 slides analyzed are considered large variations which lower the kappa score significantly. Data for three slides which have a large range of diagnosis are shown in Table 10. Payne 64 Table 10: Example of Larger Inter-Pathologist Disagreement Slide Negative Benign Minor Significan t CIS Carcinoma 1 C -B A 2 B A C 3 A B C These results caused some concern as to how reliable a 'Negative' diagnosis would be i f such a slide were to be used for M A C analysis which requires examination of a negative slide. The next step was to examine closely the 'negative' slides and their diagnosis. Of 112 slides, 39 had at least one pathologist assigned it to the negative category. For 11 (28.2%) of these slides, all three pathologists agreed that the diagnosis was 'Negative', and as such would be acceptable for M A C analysis. The next step was to assess the probability of major disagreement i f two pathologists agreed on a diagnosis of 'Negative', and the probability of such a disagreement i f just one pathologist considered the slide 'Negative'. Various definitions of 'Major' disagreement were evaluated; a range of 'Negative' to 'Minor' or worse and of 'Negative' to 'Severe' or worse are shown in Table 11. Payne 65 Table 11: Negative Diagnosis vs. Major Disagreement Definition of 'Major" Disagreement Probability of disagreement i f two pathologists agree on 'Negative' Probability of disagreement i f one pathologist agrees on 'Negative' Negative to Minor or worse 5.1 35.9 Negative to Severe or worse 2.6 20.5 Table 11 shows that i f two pathologists agree on a diagnosis of 'Negative' in a bronchial biopsy, the probability of a 'major' disagreement by an additional pathologist drops. In conclusion, it was determined that for a slide to be used for M A C analysis that two pathologists must agree that the slide is negative. Payne 66 2. MACs in Bronchial Biopsies It was decided to test for MACs in bronchial biopsies before testing in sputum for several reasons. First, to see i f M A C can be, localized to cells from the bronchial epithelium since sputum is a mix of cells from several sources. Second, to examine if there is a distance effect on MACs such as that observed by Montag in colon (120). If in a clinical setting a patient's sputum sample shows MACs and the bronchial biopsies show only normal epithelium, it would be of value i f the strength of the bronchial biopsy M A C signal could give some indication of the tumour location. This may only be an indication that the search should be concentrated in one particular lung, or perhaps even one particular lobe. Results The mean IOD value of the lymphocytes of each slide was used to normalize the samples since several features are IOD dependent. The normalized image files then were used to calculate more than 60 nuclear features such as integrated optical density (IOD), area, shape features (e.g. elongation), continuous texture features (Markovian co-occurrence features, run-length, fractal dimension, and optical density moments), and discrete texture features (185-187). Nuclear images were separated into their respective patient groups. Forward step-wise linear discriminant analysis was then used to produce a discriminant function which was best able to separate the training groups; normal epithelial nuclei of normal subjects versus normal epithelial nuclei of patients with 67 Payne invasive carcinoma. Features to be used for discrimination were chosen according to their F statistic value (parametric) as well as according to their Mann-Whitney rank-sum method (non-parametric). These discriminant function values and nuclear population statistics can be conceptualized as a ' M A C value' from which a diagnosis of M A C positive (probable cancer) or M A C negative (probable no cancer) could be assigned to a patient. M A C values were calculated using two methods. In the first method, non-cancer patients without dysplasia whether they were smokers, non-smokers, or ex-smokers, were pooled as the 'normal' group. The 'cancer' group was the Stage III cancer patients. Approximately 70 normal epithelial nuclei per 'normal' sample, and 100 normal epithelial nuclei per 'cancer' sample were randomly selected to use in the training set. The results of this method are shown in Figure 14. In the second method, the 'normal' group used only the non-smokers, while the 'cancer' group remained the same. The first test set consisted of 5,853 cells while the second set consisted of 3,566 cells. The large number of cells in each set suggests that a large number of features can be used to create a discriminant function, however, the actual number of patients is significantly smaller. Since the within sample nuclei are likely to be highly correlated, it was decided that a smaller number of features, i.e. only three, would be used. The features selected were, in order of importance; 1) medium density chromatin average extinction ratio, 2) range in intensity between brightest local maximum in the nucleus and the darkest local minimum, and 3) the nuclear area. Payne 68 Figure 14: MACs in Bronchial Biopsies by Patient Group. Figure 14 is a plot of M A C score versus patient group with a sample decision boundary. The groups are: 1. Non-smoker patients (Low risk) 2. Current or Ex-Smoker patients (High risk) 3. Patients with dysplasia in at least one other biopsy site (red circles are for patients who have developed lung cancer on follow-up). 4. Patients with Carcinoma in situ in at least one other biopsy site 5. Patients with invasive carcinoma 6. Invasive carcinoma patients after resection (red circle is for a patient who was later shown to have carcinoma beyond the resection margin. Payne Points plotted above the decision boundary in Figure 14 indicates the patient does not display M A C , while those plotted below the boundary indicates the patient is 'positive' for MACs . The decision boundary can be moved depending on what is determined to be an acceptable false positive and false negative rate. There appears to be little or no effect on M A C score due to smoking status. A decision boundary can be drawn at any point on the graph to reduce either false positive or false negative results. If the boundary is placed at zero, then no low risk patients are M A C positive, but five carcinoma in situ patients are also M A C negative. If the boundary is moved to reduce the false negative rate these values will change. One possible results is illustrated in Table 12. Table 12: Biopsy Classification using Mean MAC value. Biopsies from Group Fraction displaying MACs (%) Sub-groups Fraction displaying MACs (%) Normal 17.5 Low risk 11.1 High risk 25.9 Invasive Cancer 78.6 CIS + microinvasive 85.7 Dysplasia 35.7 Mild 17.6 Moderate 47 Severe 66.6 Resected lung cancer 10.5 Table 7 shows the M A C results for a decision boundary at the -0.3 position. There is a trend of increasing M A C score from the low risk patients to the carcinoma groups. In the dysplasia groups, sub-groups based upon the highest grade of dysplasia seen in the Payne 70 patient also shows the same trend. The M A C score of patients with resected lung cancer is similar to that for the low risk patients. Wilks' lambda for separating the training groups was 0.199, indicating a highly significant separation of the training groups, and the p-value for the separation of the training groups using the F-approximation of Wilks' lambda is < 0.0001. The overall classification matrix was 75.2% correct before and after jackknife analysis, indicating that the features selected are robust. Applying Fisher's Protected L S D test, there is no significant different between the non-smoking, smoking, and resected groups, nor is there a difference between the CIS and Stage III groups. The dysplasia group is an intermediate group by this test. This is consistent with the expected behaviour of the groups. Using the more conservative Tukey's W test (less chance of a Type I, or alpha, error), the dysplasia group is no longer significantly different from the non-smoker, smoker and resected groups, while the CIS and Stage III groups remain significantly different. This also is consistent with the expected behaviour. Payne 71 Distance Effect on MACs in Bronchial Biopsies If a bronchoscopist acquires only normal epithelium from a bronchial biopsy, and that biopsy shows MACs, it would be of value if the strength of the M A C value could give an indication of how far away a lesion of interest might be. To this end, the M A C in bronchial biopsy data was evaluated to see if this could be determined. The mean M A C value and biopsy location for all the normal biopsies of the Stage 0, Stage I and Stage III cancer patients were entered into a database. Next the location of the actual tumor was entered into the database. The distance between the tumor and the normal biopsies which showed M A C changes was calculated by several methods: i) same vs. opposite lung, ii) adjacent lobes vs. non-adjacent lobes, and iii) upper vs. lower regions. No correlation between M A C value and distance from the tumor was observed which is in contrast to the findings of Montag (120). The simplest score used is shown in Figure 15. If the tumor was in the same lung as the negative biopsy, the distance was scored as '0 ' , while i f the tumor was in the opposite lung, the distance was ' 1'. Payne 72 Figure 15: MAC value vs. Distance from tumour. o 1 MAC Proximity Figure 15 shows that there was no statistically significant relationship, at the 95% confidence interval, between Mean M A C value and the proximity of the normal biopsy used for M A C analysis to the tumour. Proximity '0 ' is M A C and tumour biopsy are from the same lung and ' 1' is when the biopsies were from the opposite lung. Payne 73 3. MACs in Bronchial Brushings After demonstrating that MACs existed in bronchial biopsies, it was desirable to test i f M A C s could be detected in cytological specimen from the lung. In this situation whole cell nuclei would be evaluated rather than sections of nuclei as was the case with the biopsies. In addition, the specimens were to be fixed in Saccomanno's fixative, which was believed to be superior to formalin for image cytometry as shown in the Methods section. It was unknown if MACs would continue to be detectable under these conditions. Results As outlined in the previous section, a M A C feature was created by using discriminant function analysis. Several aspects of the M A C feature for each diagnostic group is then evaluated for its ability to discriminate the groups. This included the mean and standard deviation of the M A C feature for each group. Using a decision point which results in a 100% correct classification of the recurrence group, we get 87% correct classification of the invasive primary carcinoma group and a 33% false positives in the normal-benign group; 70% of the false positives are from the hyperplasia and squamous metaplasia groups. Payne 74 Table 13: Brushings Slide by Slide Classification using Mean MAC value Brushings from Group Fraction displaying MACs (%) Sub-groups Fraction displaying MACs (%) Normal-Benign 33 Invasive Cancer 87 CIS 63 Dysplasia 50 Mild 37.5 Moderate-Severe 57 Recurrence of lung cancei-100 One alternative decision boundary results in 70% correct classification of the invasive cancer groups and a 20% false positive rate for the normal-benign group; 80% of the false positive are from patients with squamous metaplasia in another portion of the bronchial tree. It is possible, but not demonstrated, that these patients may have dysplasia in unsampled areas of their lungs. A cautionary note is that the sample size for the 'Recurrence of lung cancer group' is only two cases. This results show that MACs can be detected in cytological samples with a similar pattern to that in bronchial biopsies. The overall jackknife classification results is 64.1 % which is not as good as that seen in the bronchial biopsy data. Wilks' lambda however remains good at 0.122, with a p-value of <0.0001 and shows that the separation of the training groups is significant. Payne 75 4. MACs in Sputum: Mayo Clinic Slides The sensitivity of sputum cytology in the original Mayo Clinic study was 22%. Specificity was not reported, but is assumed to be at approximately 95% based upon the findings of other investigators (shown in Figure 2). For this study the two end-point outcomes chosen were: i) the subject developed lung cancer as confirmed by histology, x-ray and clinical findings and, ii) the subject did not develop lung cancer within a follow-up period of five to eight years. The set of sputum examined for M A C s were from the Mayo Clinic portion of the National Cancer Institute (NCI) Cooperative Early Lung Cancer Study. The NCI study enrolled approximately 30,000 male smokers at three different centers; Johns-Hopkins, Memorial Sloan-Kettering and the Mayo Clinic. Results The 73 sputum slides were scanned on the Cyto-Savant image cytometry system. The system was set to collect 1500 objects per slide. Only 'normal epithelial nuclei were used in the analysis. 'Normal' was defined in two manners: 1) Human selected nuclei. Normal appearing, Feulgen-Thionin stained, squamous and columnar cell nuclei were selected by agreement between two Canadian Society of Laboratory Technologists certified cytotechnologists experienced in both routine cytology and image-cytometry. Payne 76 Then a plot of IOD, corresponding to D N A amount, was made and only cell nuclei which were in the first peak (G0/G1) were kept. This presumably excludes most normal appearing cells which may be in S or G2/M phase of the cell cycle and therefore may have nuclear features which could merely reflect the cycling status of the cell rather than changes in nuclear texture which we define as 'Malignancy Associated Changes'. The average number of nuclei meeting these criteria per slide was: 438 squamous cell nuclei and 28 columnar cell nuclei per slide for slides from normal patients, and 471 squamous cell nuclei and 22 columnar cell nuclei per slide for slides from patients who developed lung carcinoma. Objects excluded were a few normal nuclei outside the G0/G1 peak, inflammatory cells, some cells considered 'atypical' or 'dyskaryotic', overlapping nuclei, and debris. Payne 77 2) Machine selected nuclei. Using the human selected group of normal appearing nuclei, the image cytometer was trained so that it would recognize normal nuclei without human intervention. By determining which features of the normal nuclei that best discriminated them from all other objects, the device simply examined all the objects collected and accepted as normal only those nuclei which matched those features within one standard deviation. No person edited the machine's selection of normal nuclei which were to be used for M A C analysis. This is a required step if the device is to be used successfully as an automated post or pre-screening device. Similar numbers of normal nuclei were discovered by the machine without human intervention as were selected originally by the human image reviewers. Next, all the normal nuclei from the non-cancer group and from the cancer group were analyzed for features which could best discriminate between these two groups. To avoid biasing the training set of nuclei for determining features for detecting cancer patients, a limit of 100 normal nuclei was applied to all slides since some slides had much larger numbers of normal nuclei than others and could unduly influence the training set. This limit was applied to both the human and machine selections. Using a step-wise linear discriminant function analysis, the features which best separated the cancer and non-cancer groups were determined. To classify each biopsy into a category, sample based features were also generated. Payne 78 Non-Cancer Patients For 'normal nuclei' as determined by human observers, 26 of 33 (78.8%) of the slides from non-cancer patients were identified as being from non-cancer patients (Table 12) by the use of MACs . For 'normal nuclei' as determined by machine classification without human review of the machine's choices, 25 of 33 (75.8%) were identified correctly by the machine classification (Table 13) as being from non-cancer patients. Of the seven false positive slides, four were the same for both machine and human classification. Cancer Patients For 'normal nuclei' as determined by human observers, 31 of 40 (77.5%) cancer patients were correctly identified, while for 'normal nuclei' as determined by machine only classification, 29 of 40 (72.5%) cancer patients were correctly identified as such. Of the false negative slides, six were the same for both machine and human classification. At this point it should again be emphasized that all these diagnoses were made by analyzing only normal looking epithelial cell nuclei. A l l diagnostic cells were purposely excluded since the objective of this feasibility study was to identify cancer patients who did not present diagnostic cells in their sputa. Payne 79 Table 14: Classification Matrix. Human Selection. Classification Matrix for Human Selected Nuclei. Numbers in Parentheses are after Jackknife analysis. Nuclei selected by two cytotechnologists as 'normal' according to conventional criteria were used for discriminant function analysis. The cytotechnologists did not select M A C or Non-MAC nuclei. GROUP Percent Correct No Cancer Cancer No Cancer 81.8(78.8) 27 (26) 6(7) Cancer 77.5 (77.5) 9(9) 31 (31) Total 79.5 (78.1) 36 (35) 37 (38) Table 15: Classification Matrix. Machine Selection. Classification Matrix for Machine Selected Nuclei. Numbers in Parentheses are after Jackknife analysis. The image cytometer was trained to recognized 'normal' nuclei from a training set of ~ 4000 normal nuclei from the Non-Cancer Patients' sputum. GROUP Percent Correct No Cancer Cancer No Cancer 78.8 (75.8) 26 (25) 7(8) Cancer 75.0 (72.5) 10(11) 30 (29) Total 76.7 (74) 36 (36) 37 (37) Payne 80 Tables 12 and 13 show both the before jackknife analysis and the more conservative after jackknife analysis values. For Human classification of the slides on a case by case basis the features used were: i) the frequency of M A C cells as derived from cell by cell analysis, ii) the number of low, density objects in the nucleus and, iii) the correlation from the gray level co-occurrence matrix. For the Machine classification of the slides on a case by case basis the features used were: i) mean value of the cluster shade, ii) the discriminant function value (mean M A C value) from cell by cell analysis, and iii) the kurtosis of the optical density maxima. Although only three features were used for the human classified 'normal' nuclei, i f feature set is reduced to just two features, there is no reduction in the total number of slides correctly classified (78.1%, jackknifed value), suggesting that this discriminant is very robust. The overall jackknife classification matrix is slightly better than for the bronchial biopsy experiment, as is Wilks' lambda for the cell by cell classification (0.117). The p-value is < 0.0001 for both human reviewed and machine only experiments. Payne 81 Inflammation Effect The next step was to examine if the M A C effect is merely a reflection of an inflammatory effect. A scale of zero to three was used. 'Zero' was very little or no inflammatory component in the sputum. 'One' was a noticeable mount of inflammatory cells in the background that does not obscure observation of the epithelial cells. 'Two' was a similar amount or slightly increased amount of inflammation in the background, and areas of epithelial cells are obscured by the inflammation. 'Three' was for very large amounts of background inflammatory cells and/or large areas of epithelial cells are obscured by inflammatory cells. There was no correlation between the calculated M A C value and either inflammatory score. These results are seen in Figure 16. Payne 82 Figure 16: MAC Score vs. Inflammation Score. 5 3 ( > 1 MAC < > Score -i _3 r \ \ -j -5 < c > -0.5 0 0.5 1 1.5 2 2.5 3 3.5 Inflammation Score For Figure 16 the regression was calculated as; R = 0.067, R 2 = 0.004. M A C Score is not correlated with the amount of acute inflammation in the sputum sample. Payne 8 3 5. MACs in Sputum: Tokyo Samples The bronchial biopsy experiment demonstrated that M A C s disappeared after surgical resection of the tumour. In the cases where the M A C effect persisted, follow-up demonstrated that these patients had residual tumour. That result suggested that M A C s are a result of a message released by the tumour rather than from a field effect of general damage to the bronchial tree which would not have shown the loss of M A C s after resection. In collaboration with Dr. Toichiro Katsumi of the Tokyo Medical College Hospital, this was investigated further in sputum samples. It was believed that since M A C s are expressed more strongly with increasing grades of dysplasia in both bronchial biopsies and bronchial brushings, that a strong M A C response in sputum may give an indication as to the prognosis of the patient, thus aiding in the choice of an appropriate therapy. Results For the sputum collected before surgical resection, we were able to correctly identify patients, on a slide by slide basis, from each group with 82% accuracy. For sputum collected after surgical resection there was 83% accuracy. This is not a significant difference. Among the nuclear features which were most useful in differentiating good from bad prognosis in both pre- and post-surgical sputum were: in bad prognosis patients, optical density maximum was higher by 8%, and range in extreme values of optical 84 Payne density was lower by 7%. This should be visualized in normal epithelial cells as slightly more chromatin condensation but with no clearing. In practice it has not been possible to identify M A C cells reliably by visual methods alone. When taking into account the degree of interobserver variability in both sputum and biopsy samples for the more pronounced changes seen in dysplasia, it is certain that reliable classification of these subtle changes requires a more objective measurement. For the good prognosis patients, the M A C value was significantly different after surgery than in the pre-surgical sputum (Wines' lambda 0.092, p = 0.0005). In the bad prognosis group, there was no significant difference before and after surgery. This suggests that a comparison between pre and post operative sputum examination for M A C may give an indication of surgical success. If there is a significant drop in M A C after surgery, there is a better prognosis. For patients who do not show this change, further evaluation may be advisable. These results support the hypothesis that M A C s are a result of a soluble factor released by the tumour. Payne 85 Discussion Lung cancer has a five year survival rate, combined for all races and sex, of approximately 13% (7). This survival is related to the stage of the disease at detection with early stage disease having a significantly better survival rate. The focus of several previous studies was thus to examine i f early detection of lung cancer would lead to increased five-year survival and/or reduced mortality (41-49, 166). The common conclusion was that five-year survival is increased by early detection, but that this does not lead to reduced mortality. This has been attributed to three effects: lead-time bias, length-biased sampling, and misclassification (163, 167). Lead-time bias is a artifact which can occur from earlier detection. If a disease is detected earlier, this may have no effect on the patient outcome, only that we are aware of the disease for a longer time. This bias can lead to an increased five-year survival time which is due only to the increased time of awareness, not to a change in the disease progression. Length biased sampling: rapidly growing cancers may become symptomatic early and are detected clinically, hence they are removed from the screened population. The cancers which remain in the screened population are slower growing and may have a longer survival time naturally. In this situation, the five-year rate of the screened population may appear to increase only because the aggressive cancers have been selected out. Payne 86 Misclassification: many cancers can be classed as in situ (no invasion). These lesions may regress, or progress so slowly that the patient dies of other causes. Prolonged survival may be observed for 'lung cancer', but little real benefit to the patient has occurred. This particular source of error has been disputed based upon autopsy information and information on disease outcome from individuals with screening-detected early stage lung cancer (171). A key article by Eddy reviewed the effectiveness of various lung cancer screening strategies including sputum and x-ray (168). The observation was made that no present screening program is effective. Several agencies including the American Cancer Society "...decline to recommend lung screening with any test at any frequency" (168). The problem is thus to develop an effective screening test. A n ideal screening test would be 100% sensitive, 100% specific, and cause no harm to the person being screened. Such a test would be expensive, hence many present screening programs use a two-tiered approach. The first test is highly sensitive (but less specific), not harmful and inexpensive. Once this test has identified a small group of high risk people, a second test with high sensitivity and high specificity, but higher cost, is used for a final diagnosis. Such a two-tiered system is employed in screening for tuberculosis and human immunodeficiency virus (HIV). The previous screening system for lung cancer was similar in that sputum, x-ray, or both were used as the first screen, and bronchoscopic examination, further x-rays or computed tomography, or more invasive procedures were used as the second screen. The Payne 87 failure of this system is due to the poor sensitivity of the first screening tests. The Mayo Clinic Study reported a sputum sensitivity of just 22%. They did not report specificity, but it is assumed to be -95% based on a review by Bocking (103). In the same study, the sensitivity of x-rays was 32% and the specificity was 72%, which is consistent with earlier studies (41, 42, 43). It is clear that sensitivity of the present system is inadequate. It was believed that the sensitivity of sputum cytology might be improved by a quantitative assessment. The hypothesis of this thesis was that malignancy associated changes (MACs) can be detected as subtle nuclear texture changes in sputum samples from patients with early carcinoma of the lung and that these M A C s are detectable before conventional diagnostic methods. Experiments were first conducted to investigate some characteristics of the specimen preparation procedures including fixation and staining. As M A C s are presently investigated in cells which are visually 'normal' the reproducibility of such visual assessment was determined. Normal bronchial biopsies from cancer and non-cancer patients were then analyzed for the presence of MACs. Normal epithelial cells from bronchial brushings and sputum were then evaluated for the presence of M A C s in patients who developed non-small cell carcinoma. Payne 88 Previous M A C work in Sputum In 1959, Nieburgs described malignancy associated changes in buccal mucosa in a qualitative manner. Since then several independent investigators have observed the phenomenon in several tissue types by both qualitative and quantitative means (115, 118. 119, 120, 121, 122, 123, 124, 125, 126, 127). There have been only two studies which have previously attempted to apply M A C s to sputum, however both of these studies were qualitative. In 1975 van Oppen Toth qualitatively examined MACs in sputum of 315 patients, 160 of which had proven malignant disease, 85 of which were lung primaries (44 squamous cell, 36 SCLC, and 5 adenocarcinoma), and 155 with benign disease. This study limited their M A C investigation to squamous cells fixed in 70% alcohol and stained by routine Papanicolaou method. M A C changes were described as: "...hyperchromasia and a specific chromatin pattern made up of small curved regular chromatin bands joining small regular chromocenters. The curved bands form the limits of small circular areas, giving a sieve-like appearance to the nuclei. There is a slight increase in the nucleo-cytoplasmic ratio. The nuclear membrane may be slightly irregular. There are usually no nucleoli." (119). Of the patients with malignant disease, 130 of 160 (81.3%) showed M A C changes. Of the patients with benign disease, 54 of 155 (34.8%) were M A C positive. Payne 89 This is a sensitivity and specificity of 81.3% and 65.2% respectively. These investigators also noted that sputa were more likely to be M A C positive in older patients and that a source of false positive M A C results was non-cancer patients who have "metabolic" disorders or other benign diseases (listed by Oppen Toth as hyperlipidemia, liver cirrhosis, diabetes mellitus, lung tuberculosis, heart failure, cardiovascular diseases, and peptic ulcer). This study also compared three investigators' M A C results. The sensitivity for detecting M A C s varied from 60.9% to 81.3% for patients with tumours and the specificity was 4.8% to 34.1%, suggesting that reproducibility of qualitatively assessing M A C s was a problem. The source of the variability was not investigated in that report. A 1986 study by Vetrani also reported M A C s in sputum. That study examined 97 asbestos exposed workers, 27 of whom were non-smokers (118). The staining of these specimens was according to the modified Papanicolaou stain described by Koss (131). These specimens included adequate and inadequate specimens, although a breakdown was not provided. Twelve cases were determined to be M A C positive (eight smokers, four non-smokers). The single case of carcinoma in situ in the study was M A C positive, three of forty-two inflammatory cases and three of fifteen cases showing squamous metaplasia with or without atypia were M A C positive. No follow-up of the ultimate fate of the patients was provided. The Vetrani group concluded that since the percentage of M A C positive sputa was higher in the group with squamous metaplasia with atypia and in their one case of carcinoma in situ than without, that M A C is positively related to malignant tumors. This conclusion appears to be pre-mature since there is no follow-up i 90 Payne of the patients, and there is only one case of carcinoma in situ. If the M A C positive patients had been demonstrated to have a higher probability of eventually developing tumours it would be possible to agree with their conclusion. Additionally, the problems of the reproducibility of hematoxylin based staining and of subjective assessment of the subtle changes associated with MACs are not addressed. MACs in Bronchial Biopsies Biopsies from the conducting portion of the lung were obtained during bronchoscopic examination with the LIFE imaging system (Xillix Technologies Corp., Richmond). This bronchoscopic device was developed in conjunction with the B.C. Cancer Research Centre and has been demonstrated to have a higher sensitivity for dysplastic lesions than white light bronchoscopy alone (165, 188). Any areas interpreted as abnormal by the bronchoscopist were biopsied as were one or more normal appearing control sites. It is assumed that all abnormal areas were biopsied during the bronchoscopy procedure such that the highest degree of dysplasia present in the lung was therefore known. As the entire bronchial tree is not accessible to current bronchoscopes, the LIFE device is not 100% sensitive, it is possible that areas of higher degree dysplasia were present than were sampled, however the LIFE device is the most sensitive instrument presently available. Forward step-wise discriminant function analysis was used to develop a M A C function for separating the bronchial biopsy diagnostic groups. The number of nuclei Payne / 91 used in the analysis was over 2000 per diagnostic group. The specimen sizes were 14 (from 12 subjects) for the Stage III cancer group and 36 (from 14 non-smokers) for the normal group. A guideline for this type of analysis is that for each diagnostic group, 10 or more 'objects' are required for each additional feature. Although nuclei were used to create the function and there are large numbers of nuclei, the number of patients was fairly small, hence the feature set for the function was then revised down to three features. Table 16 shows results for three and fourteen features for comparison. Features were ranked according to their F statistic (parametric) and Mann-Whitney rank sum (non-parametric). The rank sum results could not be interpreted as several features were scored as p < 0.0000, which was the limit of significant figures in the software used (BMDP); hence it was not possible to determine which were the most effective features. Performing the calculations by hand of -60 nuclear features on approximately 3000-5000 objects was not considered practical. The features were thus chosen by their F values. The features used were: 1. medium density chromatin average extinction ration (the mean optical density of the medium density pixels divided by the mean optical density of the low density pixels). 2. range in intensity between the brightest local maximum in the nucleus and the darkest local minimum. 3. nuclear area. Using either 14 features to make a nucleus-by-nucleus classification gave similar results to the 3 features used to make a slide-by-slide classification (Table 16). 92 Payne Table 16: Slide by Slide and Cell by Cell Classification for Bronchial Biopsies. Biopsy Slide by Nucleus by Sub-Group Slide by Nucleus by Group Slide. Nucleus. Slide. Nucleus. Fraction Fraction Fraction Fraction MAC MAC MAC MAC positive positive positive positive (3 features). (14 features). (3 features). (14 features). Normal 17.5 22.2 Low-Risk 11.1 11.1 High-Risk 25.9 37 Dysplasia 35.7 Mild 17.6 35.3 Moderate 47 53 -Severe 66.6 50 CIS and 85.7 85.7 Microinvasive Invasive 78.6 100 Resected 10.5 10.5 Table 16 shows that roughly 80-85% of lung cancer patients can be identified by imaging normal regions of the bronchial epithelium; this is approximately a 15-20% false negative rate for cancer patients and 10% false positive rate for low risk non-cancer subjects. Figure 14 (page 76) shows that two patients with dysplasia who were M A C Payne 93 positive have since been demonstrated to have lung carcinoma. This suggests that M A C s may be an indicator of higher risk for developing cancer than patients with dysplasia who are M A C negative. This is the subject of an ongoing study. Figure 14 also shows that one patient who was resected for lung cancer and was M A C positive was later shown to have tumour beyond the resection margin. Montag demonstrated that M A C expression in colon decreased with distance from the tumour. Figure 15 illustrates that it was not possible to reproduce this result in bronchial biopsies. Although there was a slight trend that the M A C score was higher when the negative biopsy imaged for M A C analysis was on the same side as the tumour, the result was not statistically significant (p = 0.2). Several methods of scoring distance were attempted including; upper versus middle and lower lobes, adjacent lobes, and upper and middle lobes versus lower lobes, but none showed a significant relationship. One explanation for this difference is that the distances examined by Montag were on the order of a few millimeter, while the distances examined in the lung were on the order of 10-30 cm. There are several possible explanations for this difference. First, in the colon the M A C effect decreased with distance, but there was no reported distance at which the effect disappeared. It may be that we are detecting the same changes, but at very subtle levels. Second, the M A C effect we detected may be due to a field effect resulting from a generalized damage in all parts of the bronchial tree. The fact that M A C s disappeared after resection does not support this, but the sample size in this experiment is not sufficiently large to make that conclusion. Third, the M A C s reported by Montag may be Payne 94 biologically difference in that they are only seen in a localized area as i f by a paracrine messenger. Although formalin was considered a 'sub-optimal' fixative based upon the fixation study in this thesis, good results consistent with those of the ethanol-polyethylene glycol fixed specimens were obtained. It is now believed that fixative specific discriminant functions can be obtained. It has been demonstrated that MACs can be detected in bronchial biopsies and that the M A C effect decreases proportionally as the degree of dysplasia in the patient decreases. The following sections address increasing the sensitivity of the sputum test by using MACs, and provides evidence that MACs are not a result of a field effect nor a locally restricted messenger. Payne 95 MACs in Bronchial Brushings. A central purpose of this study was to examine i f the findings of the bronchial biopsy study could be reproduced in cytological specimens with different fixation. The results of this study mirrored very closely those of the biopsy findings. It was expected that as a result of using whole nuclei preserved in what was believed to be a superior fixative that the results should be superior to those of the biopsy experiment. As this is not observed, it appears that consistent specimen treatment is more important than the specific fixation method for the detection of the nuclear texture features specific to MACs . Other nuclear characteristics such as D N A amount are also of interest in M A C studies because M A C s are defined as being in normal G0/G1 nuclei. A fixative which gives consistent D N A amount values, such as Saccomanno's fixative, remains the method of choice. The apparent limit of 80-90% correct classification of carcinoma in situ by M A C s is possibly a result of the biological nature of non-small cell lung cancer. Not all of these lesions may release factors which cause MACs. Payne 96 MACs in Sputum: Mayo Clinic Samples Seventy-three sputum samples from the Mayo Clinic portion of the NCI study on early detection of lung cancer were analyzed for MACs to determine if the sensitivity of the sputum test could be increased. Forty sputum slides from nine patients who developed lung cancer during the trial and thirty-three sputum slides from eleven subjects who did not develop lung cancer during five to eight years of follow-up were studied (Table 8). Approximately 1500 objects were collected per slide. Nuclei used for analysis were selected by two methods as outlined previously. In the first method, two cytotechnologist reviewed the images of all objects and only those which both technologists agreed were normal epithelial nuclei and which were in the GO/Glpeak of the IOD plot were accepted for M A C analysis. The second method was built from the first; nuclear features which best discriminated the nuclei selected by the first method from all other objects in the same IOD range were used to select nuclei for M A C analysis. An average of 400 nuclei per slide were selected by both methods. For analysis, a limit of 100 epithelial nuclei per slide was used so that slides for which, for example, 1000 epithelial nuclei were collected would not overwhelm slides with only 300 epithelial nuclei. Forward step-wise linear discriminant function analysis was used to determine which features best separated the normal nuclei from cancer patients from normal nuclei from non-cancer patients. The results are shown in Tables 10 and 11. The M A C function was able to correctly identify -80% of the slides as from cancer or non-cancer patients when the test set was based upon the set of nuclei edited by humans. For nuclei selected by nuclear features without human editing ('machine' selected), -77% of the slides were correctly identified. The False Positive and False Negative rate can be demonstrated in a plot of the operating characteristics of the M A C test. Payne 97 Figure 17 shows examples of Operating Characteristics curves (also known as a Receiver Operating Characteristics or ROC curve) (158). This chart can be used to monitor the quality of a process. In this particular case we are interested in the False Positive and False Negative performance of M A C s in the sputum test. Figure 17: ROC curve examples. The straight diagonal line is equivalent to a test with no ability to discriminate. A plot which follows the abscissa and ordinate would be a test with perfect ability to discriminate; 100% sensitivity and 100% specificity. The curved line is an example of a hypothetical test which has neither 100% sensitivity nor specificity. With this plot it is possible to choose an optimized operating point for a particular test. If the population of interest is high risk, having a low false negative may be important. If one develops a highly sensitive test that has low specificity, that test may be appropriate for one disease False Negative Rate False Positive Rate 98 Payne but not for another. As an example, if the treatment for that disease is to administer low levels of vitamin C, the low specificity is not a concern, however i f the treatment is to administer high dose radiation, the low specificity is of great concern. This demonstrates that a screening program must be tailored in a disease-specific manner. Figure 18: ROC Curve for MACs in Mayo Clinic Sputum 100 100 Percent False Positive The ROC curve for M A C s in Mayo sputum shows that i f a low false positive rate of about 3% was chosen, the false negative rate would be about 80% (Figure 18). This translates into 20% of people with no other signs or symptoms of lung cancer would be correctly identified as harboring a lung cancer. Based upon 160,000 new cases of lung cancer and a present conventional sputum sensitivity of 20%, 128,000 new cases of lung cancer remain occult to conventional sputum and x-ray in the United States (159). If M A C s are 20% sensitive to lung cancers that are occult to conventional methods, then 20%o of the 128,000 occult cases (25,600) should be detected. If a higher sensitivity Payne 99 (lower specificity) were chosen for M A C detection, then at some point the large number of bronchoscopies, or other follow-up procedures, would likely become cost prohibitive. One possible operating point is at approximately 80% sensitivity and 80% specificity. The reduction in specificity as compared to the previously mentioned operating point may be handled by a second screening by methods such as bronchoscopy. Increasing the frequency of the screening by conventional methods from annual to once every four months does not offer any mortality advantage (170). The NCI studies on lung cancer screening were designed for demonstrating a 50% reduction in lung cancer mortality with an 80% power to evaluate any advantages to adding conventional sputum to standard chest roentgenography (161, 162). Any reductions in lung cancer mortality less than this would have required a sample size of 60,000 and a follow-up of 10 years (134, 162). Flehinger points out that i f detection were perfect with no change in cure, mortality may be reduced by 32%, and that only a 10% reduction would amount to over 14,000 lives potentially saved simply because of the huge number of lung cancers. The NCI study showed that the sensitivity of sputum cytology in the late 1970's was insufficient to demonstrate the projected reduction in lung cancer mortality. Although not all early lesions are cured, improvements in therapy are likely to be most effective on early lesions rather than advanced stage cancer, hence any means of identifying early lesions should provide a survival advantage as better treatments evolve. The detection of M A C s in sputum of cancer patients who have no other indications of cancer is a significant step in improving detection. Mulshine, Tockman and Smart proposed that any screening program must be a reliable method of detection which considers both sensitivity and specificity, and once detected, there must be an effective treatment (163). Additionally it must be acceptable to the population being screened. This usually means a test must be; inexpensive, 100 Payne non-invasive, low risk as compared to the disease, have little psychological or sociological cost, and take into account cost of further testing, misclassification and mistreatment. Conventional sputum cytology has not thrived as a screening tool because it is not sufficiently sensitive. If the lesion detected is early stage, successful treatment by surgery is available, but i f the lesion detected is late stage, there is currently no effective treatment. By improving the sensitivity of the sputum test for earlier lesions while maintaining specificity, survival should improve since effective treatment is available for early lesions. The work on Malignancy Associated Changes in bronchial biopsy and brushing specimens demonstrated that MACs were detectable in carcinoma in-situ and dysplasias suggesting that MACs in sputum may be useful in detecting candidates for photodynamic therapy or chemoprevention trials. M A C expression disappeared after tumor resection, suggesting that quantitative microscopy may be useful as a simple, non-invasive means of treatment follow-up for recurrence or complete resection. This point is addressed in the next section. Payne 101 MAC in Sputum: Tokyo Specimens Two points from the previous section that required further investigation were; i f M A C s can be detected in sputum samples of non-small cell cancer patient other than squamous cell carcinoma, and i f MACs are reduced after surgical resection on a larger set of sputum. In collaboration with Dr. Toichiro Katsumi these questions were addressed by investigating two groups of patients with a variety of primary non-small cell lung cancers, it was shown that patients with a good prognosis (no recurrence nor metastasis for at least 30 months) can be differentiated from patients with a poor prognosis with over 80% accuracy. Furthermore, the sputum from patients with a good prognosis showed a statistically significant reduction in MACs in their sputum samples after surgical resection, while the poor prognosis did not show this reduction. It is apparent from the earlier bronchial biopsy data and this sputum data that M A C s are not specific only to squamous cell carcinoma but can be detected in the most common non-small cell carcinomas. This means that M A C s are useful in at least 80% of histologic types of lung Carcinoma. The significant reduction in M A C value after surgical resection in patients who do not develop a recurrence or metastasis for 30-60 months suggests strongly that M A C s are caused by a soluble factor released by the tumour. This particular point is being addressed by Dr. Sharon Sun who has developed an in vitro M A C model (unpublished data). Dr. Sun has shown in co-cultures of normal human bronchial cells and lung tumour cells, that MAC-like changes can be induced in the normal cells. This effect has been shown to be media-independent. Payne 102 M A C Mechanism A possible cause of MACs was believed to be that of 'field cancerization' as first proposed by Slaughter (16). This theory suggested that when an epithelium is exposed to a carcinogen, all regions of that epithelium have a higher probability of developing a cancer. There is molecular evidence supporting this model (192). When a tissue is exposed to a carcinogen, the cancer usually arises at one or a few points rather than at all locations at the same time. The other regions of the tissue may show some intermediate amount of damage which will be repaired without progressing to cancer. Other areas wil l however, have persistent damage. As an example, the lungs can be entirely exposed to cigarette smoke, yet when cancer occurs it is found in one (or few) location, while other areas of the lung show dysplasia. Recent work on bronchial biopsies of 'pre-malignant' lesions has demonstrated that loss of heterozygosity (LOH) of the short arm of chromosome 3 occurs in 76% of regions of hyperplasia in lung cancer patients (150). In collaboration with Dr. Luc Thiberville we have demonstrated that such L O H also occurs on 5q and 9p (151). The percentage of lesions which demonstrated one or more deletions increased with the grade of dysplasia. We have also demonstrated that biopsies with L O H at those sites can be differentiated from biopsies without L O H with 75% accuracy using only one nuclear feature, nuclear area, and 85% accuracy using two features (189). It was thus reasonable to consider that M A C s are a reflection of unknown, non-specific chromosomal damage. Payne 103 The M A C data from the bronchial biopsy study however conflicts with this because M A C changes are reduced to normal levels after tumour resection. This observation suggests that MACs may be caused by a factor released by the tumour. Preliminary work in cell cultures also tends to support the soluble factor hypothesis (164). The truth may in fact be some combination. With the M A C data in bronchial biopsies it was observed that patients with dysplasia may or may not exhibit M A C s in their normal epithelium. It is possible that either some damage must occur in the normal area (field effect) for a message from the dysplasia to be received, or some specific damage must occur in the dysplasia before the message is released, or that similar tumour types release different substances. There are some strengths and possible utilities that arise from the results presented in this thesis. Collection of sputum is a fairly simple procedure which is not difficult or harmful to a patient. The method used for assessing M A C s is based upon a quantitative method which should improve the reproducibility of results as compared to conventional methods. The sensitivity and specificity of such a test can be adjusted by choosing a threshold level for M A C positivity as determined by the ROC curve. When compared to cervical cytology which is targeted at a large population with a low cancer population, a much smaller target population can be chosen such as, smokers over 55 years of age with 30 pack-years of smoking. It may be possible to further define the high risk group by assessing lung function. It has been demonstrated that patients with an FEVj /FVC of less than 70% are more likely to harbour a lung cancer than patients with a normal ratio (190). Payne 104 Significance of MACs as a Diagnostic Tool Other investigators continue to explore alternative methods for the early detection of lung cancer. Many of these are applied to sputum samples which are easy to obtain and can be portioned out for multiple studies. Mao examined sputum from fifteen patients of the Johns Hopkins portion of the NCI screening trial who developed adenocarcinoma or large cell carcinoma of the lung. Ten of those fifteen patient's tumors contained either a p53 or ras gene mutation (67% sensitivity) (191). They had no false positives (100% specificity). Using a PCR based assay the sputum of eight of the ten patients with mutations showed identical mutations one to thirteen months prior to clinical diagnosis. The sputum samples examined by Mao were cytologically 'negative' for cancer. Unfortunately their definition of 'normal' was sputum which contained at least 'moderately' atypical cells in the sputum. The advantage of M A C s is that atypical cells need not be present. Mao's assay was not able to detect mutations greater than 13 months prior to clinical diagnosis. We have demonstrated that M A C s are present in at least some patients three years before clinical diagnosis, although at the threshold chosen a single M A C positive result would have to be verified to reduce the chances of a false positive result. Since their method requires the presence of abnormal cells while the M A C assay does not, it is likely that M A C s would be more useful in a general screening program. Tockman at the Johns Hopkins University has developed a monoclonal antibody assay for sputum samples (192). Using murine monoclonal antibodies to a SCLC Payne 105 glycolipid antigen and a N S C L C protein antigen Tockman investigated i f these antibodies would improve the sensitivity and specificity of sputum. From the Johns Hopkins portion of the NCI Lung Study sixty-nine sputum with moderate or greater atypia were chosen; twenty-six sputum from twenty-two patients who progressed to cancer and forty-three sputum from forty patients who did not. Fourteen of the twenty-two patients who eventually developed cancer stained positively with at least one antibodies (64% sensitivity) while five of the patients who did not develop cancer stained positively (88% specificity). On close examination of their sample set another pattern emerges. The Johns Hopkins portion of the NCI trial enrolled 5,226 male smokers for cytology screening. Of those, 626 (12%) had moderate atypia or greater on one or more of their specimens. It is from these 626 participants that specimens were selected. Of this group, 577 had marked atypia once or less, and 49 had marked atypia two of more times in their sputum. Of the 577 cases, 40 (7%) actually developed lung cancers representing all the major types; squamous (twelve), adeno- (seven), small cell (nine), large cell (eight), and others or mixed (four). Of the 49 with two or more marked atypias, 46 developed lung cancers including squamous (forty-one), adeno (three), and large cell (two). If these values are used in the sensitivity and specificity equations, a sensitivity of 94%, and a specificity of 99% is attained i f the marker used is 'Two or more marked atypias'. The problems with applying a 'two or more marked atypias' as a marker are, first, the reproducibility, since this atypia is determined subjectively and prone to inter-observer variation, and secondly, as with Mao's study, highly atypical cell must be present in the specimen. Payne 106 In conclusion, the hypothesis that the sensitivity of sputum cytology for detecting early lung cancers can be improved by exploiting the phenomenon of malignancy associated changes has been addressed. The sensitivity of conventional sputum one year before clinical detection in the Mayo Clinic set of slides was 0%. The sensitivity of quantitative M A C detection in sputum depends on the operating point chosen on an ROC curve. Even choosing a relatively low sensitivity with high specificity is a significant improvement over conventional cytology. In the Tokyo set of sputum samples, it was possible to correctly classify recurrence or second primary lung tumours with greater than 80% overall accuracy as much as three years in advance. It would be interesting to pursue these observations on a larger scale and in a prospective manner to confirm these findings. Payne 107 Appendix A Form used in Inter-Pathologist Diagnostic Agreement for Bronchial Biopsies. DYSPLASIA BRITISH COLUMBIA CANCER AGENCYA/.G.H./CANCER IMAGING PATHOLOGIST: DIAGNOSIS SLIDE NEGATIVE MILD MODERATE SEVERE/ MARKED C.I.S. CARCINOMA 0219 4020 1570 0194 1569 0218 4069 0205 4035 1568 0211 1551 1567 4049 1559 0198 0213 1565 DYSPLASIA STUDY: PAGE 1 Payne 108 Appendix B Selected Nuclear Features A l l nuclear features are described in detail in references 186, 187. This appendix includes examples of the major categories of features used in this thesis. 1. Size and shape features These are the easiest to visualize and include features such as size and radius. Example: Nuclear area. The sum of all connected pixels forming an object. 2. Photometric features These values report optical density levels. Example: Integrated Optical Density (IOD). The sum of all optical density values of the object. E OD (x,y), where OD = -log 1 0 (I/I0) and I 0 is the intensity of the incident light and I is the intesity of the transmitted light, For n pixels with position (x,y) in the object. 3. Discrete texture features These features are based upon the segmentation of the nucleus into discrete regions of low, medium, and high density. Example: Low D N A Area = (Area iow)/(Area t o t a L), where Area l o w is the area of all pixels classified as having a low IOD. 4. Run Length Features Run length texture features describe chromatin distribution in terms of the length of lines comprised of consecutive pixels with the same gray level value. Payne 109 5. Fractal texture features These features represent measurements of the area of a three dimensional surfae created on the basis of the optical density of the D N A in the nucleus. This can be likened to a mountain range; the greater the range in values, and the greater the number of peaks and valleys, the greater the fractal area. 6. Markovian texture features. These features characterize the gray level variation between adjacent pixels in the object image. The Markovian features used in this work use the probability distributions of adjacent gray level sums and differences. 110 Payne REFERENCES 1. National Cancer Institute of Canada. Canadian Cancer Statistics 1995, Toronto, Canada, 1995. 2. American Cancer Society. Cancer Facts and Figures-1995, Atlanta, GA, 1995. 3. Du Y X , Wu JM. International symposium on lifestyle factors and human lung cancer, Guangzhou, China, 12-16 December, 1994. Lung Cancer 1996;14(1):S1-S2. 4. Stanley K , Stjernsward J. Lung cancer—a worldwide health problem. Chest 1989;96(1):1S-5S. 5. Evans W K . Rationale for the treatment of non-small cell lung cancer. Lung Cancer 1993;9(2):S5-S14. 6. Boring CC, Squires TS, Tong T. Cancer Statistics. 1991;41:19-36. 7. Kosary C L , Ries L A G , Miller B A , Hankey BF, Harras A , Edwards B K (eds). SEER Cancer Statistics Review, 1973-1992: Tables and Graphs, National Cancer Institute. NLHPub. No. 96-2789. Bethesda, M D , 1995. 8. Ettinger DS. New drugs for treating small cell lung cancer. Lung Cancer 1995;12(3):S53-S61 9. Johnson DH. Future directions in the management of small cell lung cancer. Lung Cancer 1995;12(3):S71-S75 10. Kreyberg L . Histological typing of lung tumours, 2nd ed. Geneva, World Health Organization, 1981 11. International Union Against Cancer (TJICC) International Union Against Cancer, Geneva, 1987;69-73 12. Woolner L B , Fontana RWS, Sanderson DR. Mayo lung project evaluation of lung cancer screening through December 1979. Mayo Clin Proc 1981;56:544-555 13. Vincent RG, Pickren JW, Lane WW, Bross I, Hiroshi T, Houten L, Gutierrez A C , Rzepka T. The changing histopathology of lung cancer: a review of 1682 cases. Cancer 1977;39:1647-1655 14. Melamed MR, Flehinger BJ, Zaman M B . Screening for early lung cancer—results of the Memorial Sloan-Kettering study in New York. Chest 1984;86:44-53. Payne 111 15. Carter D, Eggleston J. Tumors of the lower respiratory tract. Washington, D.C., AFIP, 1980. 16. Slaughter DP, Southwick HW, Smejkal W. "Field cancerization" in oral stratified squamous epithelium. Cancer 1953;6:963-968. 17. Wang T, Zhou B, Shi J. Lung cancer in nonsmoking Chinese women: a case-control study. Lung Cancer 1996;14(l):s93-s98. 18. Ffittelman W N . Advances in the detection of preneoplasia and early carcinoma:opening remarks to mini-symposia Lung Cancer 1994; 11(2): 13.). 19. Harting FH, Hesse W. Der Lungenkrebs, die Bergkrankheit in der Schneeberger Gruben. Vgschr Med Gerich Off Sanit 1879;30:296-309 20. Saupe E. Carcinoma of the lung in arsenic miners: two cases. Arch Gewerbepathol 1930;1:582 21. Lynch K M , Smith WA. Pulmonary asbestosis: carcinoma of lung in asbestosilicosis. A m J Ca 1935;24:56-64 22. Mancuso TF. Mortality study of beryllium industry workers' occupational lung cancer. Env Res 1980;21:45-55 23. Pfeil E. Lung tumors and occupational disease in the German chromate-producing industry. Dtsch Med Wochenschr 1935;61:1197-1200 24. Doll R. The causes of death among gas-workers with special reference to cancer of the lung. Br J Ind Med 1952;9:180-185 25. Doll R, Morgan L G , Speizer FE. Cancers of the lung and nasal sinuses in nickel workers. Br J Ca 1970;24:623-632 26. Wagoner JK, Archer V E , Lundin FE, Holaday DA, Lloyd JW. Radiation as the cause of lung cancer among uranium miners. N Engl J Med 1965;273:181-188 27. Du Y , Cha Q, Chen X , Chen Y , Huang L , Feng Z, Wu X , Wu J. A n epidemiological study of risk factors for lung cancer in Guangzhou, China. Lung Cancer 1996;1:S9-S37 28. Frank A L . Epidemiology of lung cancer. In Biology of Lung Cancer. 6-15, 1985. 29. Adler I. Primary malignant growths of the lung and bronchi: A pathological and clinical study. Logmans Green & Co., New York, 1912. Payne 112 30. Glantz SA, Slade J, Bero L A , Hanauer P, Barnes DE. The cigarette papers. University of California Press, Berkeley, 1996. 31. Ochner A, DeBlakey M . Archives of Surgery 1941;42:209-258. 32. Wynder E L , Graham EA. Tobacco smoking as a possible etiologic factor in bronchogenic carcinoma: a study of six hundred and eighty-four proved cases. J A M A 1950;143(4):329-336. 33. Wynder E, Graham E, Croninger A. Experimental production of carcinoma with cigarette tar. Cancer Res 1953;13:855-864. 34. . Royal College of Physicians. Smoking and health. A report of the Royal College of Physicians on smoking in relation to cancer of the lung and other diseases. London: Pitman Medical Publishing Co., 1962. 35. USDHEW. Smoking and Health. Report of the Advisory Committee to the Surgeon General of the Public Health Service. US Department of Health, Education and Welfare. Public Health Service Publication No. 1103, 1964. 36. Frank A L . Epidemiology of Lung Cancer 37. Smoking prevalence and lung cancer death rates. In Smoking and Tobacco Control Mongraphs 1: Strategies to Control Tobacco Use in the United States. National Institutes of Health Publication No. 92-3316, 1991. 38. Feld R. Current perspectives in the treatment of non-small cell lung cancer. Lung Cancer 1993;9(2):S1-S4. 39. Mountain CF. A new international staging system for lung cancer. Chest 1986;89S:225S-233S. 40. Patel A M , Peters SG. Clinical manifestations of lung cancer. Mayo Clin Proc 1993;68:273-277. 41. Boucot KR, Horie U , Sokoloff MJ. Lung cancer detection by survey methods. A m J Public Health 1959;49:793-799. 42. Nash F A , Morgan JM, Tomkins JG. South London lung cancer study. Br Med J 1968;2:715-721. 43. Brett GZ. Bronchial carcinoma in men detected by selective and unselective minature radiography. A review of 228 cases. Tubercle 1959;40:192-195. Payne 113 44. Lilienfeld A M , Archer PG, Burnett CH. An evaluation of radiographic and cytologic screening for the early detection of lung cancer: A cooperative pilot study of the American Cancer Society and ther Veterans Administration. Cancer Res 1966;26:2083-2121. 45. Frost JK, Ball W C Jr, Levin M L . Early Lung cancer detection: Results of the initial (prevalence) radiologic and cytologic screening in the Johns Hopkins study. A m Rev RespirDis 1984;130:549-554. 46. Flehinger BJ, Melamed M D , Zaman M B . Early lung cancer detection:Results of the intial (prevalence) radiologic and cytologic screening in the Mayo Clinic study. A m Rev RespirDis 1984;130:565-570. 47. Fontana RS, Sanderson DR, Taylor WF. Early lung cancer detection: Results of the initial (prevalence) radiologic and cytologic screening in the Johns Hopkins study. A m Rev Respir Dis 1984;130:561-570. 48. Matsuda M , Susuki T, Sobue T, Horai T, Fukuoka M , Kusunoki Y , Kikui M . A n evaluation of sputum cytology in mass screening for lung cancer. G B K Fortbildung Aktuell. 1992;62:13-16. 49. Kubik A , Polak J. Lung cancer detection: results of a randomized prospective study in Czechoslovakia. Cancer 1986;57(12):2427-2437. 50. Allen W B , Whittlesey P. The results of the experimental use of sulfur dioxide in the production of material for cell studies in lung cancer. Ann Intern Med 1960;52:326-329. 51. Bayer E. Die Bedeutung der exfoliativzytologie fur die diagnostik maligner lungentumoren. Dissertation, Technical University of Munich, 1976. 52. Bedrossian CW, Rybka DL. Bronchial brushing during fiberoptic bronchoscopy for the cytodiagnosis of lung cancer: comparison with sputum and bronchial washings. ActaCytol 1976;20:446-453. 53. Bender B L , Cherock M A , Sotos SN. Effective use of bronchoscopy and sputa in the diagnosis of lung cancer. Diagn Cytopathol 1985;1:183-187. 54. Clee M D , Sinclair DJM. Assessment of factors influencing the results of sputum cytology of bronchial carcinoma. Thorax 1981;36:143-146. 55. Dahlgren SE, Lind B. Comparison between diagnostic results obtained by transthoracic needle biopsy and by sputum cytology. Acta Cytol 1972;16:53-58. Payne 114 56. Dudgeon LS, Wrigley CH. On the demonstration of particles of malignant growth in the sputum by means of the wet-film method. J Laryngol Otol 1935;50:752-763. 57. Duguid H L D , Huish DW. Clinical evaluation of cytodiagnosis in bronchial carcinoma. Br Med J 1963;2:287-291. 58. Ebner HJ, Schoen HR, Sandritter W. Wert der routinemussigen, auf drei verwertbare Einsendungen begrenzten sputumzytologie fur die diagnose des bronchialkarzinoms. Thoraxchirurgie 1965;15:337-343. 59. Erozan YS, Frost JK, Cytopathologic diagnosis of cancer in pulmonary material: A critical histopathologic correlation. Acta Cytol 1970;14:560-565. 60. Farber SM, Benioff M A , Frost JK, Rosenthal M , Tobias G. Cytologic studies of sputum and bronchial secretions in primary carcinoma of the lung. Dis Chest 1948;14:633-664. 61. Farber SM, McGrath A K , Benioff M A , Rosenthal M . Evaluation of cytologic diagnosis of lung cancer. 1950;JAMA 144:1-4. 62. Gagneten CB, Geller GE, Saenz M C . Diagnosis of bronchogenic carcinoma through the cytologic examination of sputum, with special reference to tumor typing. Acta Cytol 1976;20:530-536. 63. Gowar FJS: Carcinoma of the lung: The value of sputum examination in diagnosis. Br J Surg 1943;30:193-200. 64. Grunze H . Derzeitiger stand der zytodiagnostik bei erkrankungen des thorax. Dtsch Med Wochenschr 1966;91:1476-1483. 65. Hinson K F W , Kuper SWA. The diagnosis of lung cancer by examination of sputum. Thorax 1963;18:350-353. 66. Hjelt L . On cytologic diagnosis of pulmonary cancer. Acta Pathol Microbiol Scan (Suppl) 1952;93:128-134. 67. Holobro P, Dalquen P, Perruchoud A, Herzog H . Zytologische und histologische untersuchungsmethoden bei lungentumoren. Dtsch Med Wochenschr 1978; 103:17-20. 68. Hopker WW, Liil l ig H. Lungenkarcinom. Resektion, morphologie und prognose. Berlin, Springer-Verlag, 1987, pi7. 69. Hopmann G, Felten RW. Bronchuskrebs und zytologische diagnostik. Therapiewoche 1967;17:1229-1239. 115 Payne 70. Huuskonen MS, Taskinen E, Vaaranan V. Sputum cytology of asbesdos workers. Scand J Work Environ Health 1978;4:284-294. 71. Jay SJ, Wehr K, Nicholson DP, Smith A L . Diagnostic sensitivity and specificity of pulmonary cytology: comparison of techniques used in conjunction with flexible fiberoptic bronchoscopy. Acta Cyto 1980;24:304-312. 72. Jennings RC, Shaw K M . The value of cytological examination of the sputum in the diagnosis of carcinoma of the bronchus. Thorax 1953;8:288-294. 73. Kahlau G. Exfoliativcytologie der atemorange. Verh Dtsch Ges Inn Med 1973;56:204-211. 74. Kanhouwa SB, Matthew MJ. Reliability of cytologic typing of lung cancer. Acta Cytol 1976;20:229-232. 75. Kawachi H , Shimokata K. Factors affecting the rate of possitivity of sputum cytology in lung cancer. Jpn J Clin Oncol 1985;15:451-456. 76. Kern W H , Markovits B A , Lindesmith GG. The prognostic significance of sputum cytology. Acta Cytol 1968;12:430-432. 77. Kleeman W. Bedeutung der sputumzytologie fDr die diagnostik des bronchialkarzinoms. Arch Geschwulstforsch 1984;54:505-509. 78. Koss L G . Cytological diagnosis of lung cancer. Acta Pathol JPN 1958;8:99-111. 79. Koss L G , Melamed MR, Goodner JT. Pulmonary cytology: A brief survey of diagnostic results from July 1st 1952 until December 31st 1960. Acta Cytol 1964;8:104-113. 80. Laurie W, Szaloky L G . Sputum cytology in the diagnosis of bronchial carcinoma. Med J Aust 1971;30:247-251. 81. Liebow A A , Lindskog GF, Bloomer WE. Cytological studies of sputum and bronchial secretions in the diagnosis of cancer of the lung. Cancer 1948;1:223-233. 82. Lilienfeld A. An evaluation of radiologic and cytologic screening for the early detection of lung cancer. A cooperative pilot study of the American Cancer Society and the Veterans Administration. Cancer Res 1966;26:2083-2121. 83. Lukeman JM. Reliability of cytologic diagnosis in cancer of the lung. Cancer Chemother Rep (suppl) 1973;4:79-93. Payne 116 84. Meyer JA, Umiker WO. A review of problems relating to the "diagnostic triad" in lung cancer, bronchoscopy, scalene lymph node biopsy and cytopathology of bronchial secretions. S Clin North Am 1961;41:1233-1244. 85. Ng A B , Horak GC. Factors significant in the diagnostic accuracy of lung cytology in bronchial washing and sputum samples: II. Sputum samples. Acta Cytol 1983;27:397-402. 86. Oswald NC, Hinson K F W , Canti G, Miller A B . The diagnosis of primary lung cancer with special reference to sputum cytology. Thorax 1971;26:623-632. 87. Papanicolaou GN. Diagnostic value of exfoliative cells from cancerous tissue. J A M A 1946;131:372-378. 88. Parker RE, Reid JD. Five year survey of results of cytological examination for lung cancer. N Z Med J 1960;59:68-72. 89. Payne CR, Stovin PGI, Barker B, McVittie S, Stark JE. Diagnostic accuracy of cytology and biopsy in primary bronchial carcinoma. Thorax 1979;34:294-299. 90. Philips FR. The identification of carcinoma cells in the sputum. Br J Cancer 1954;8:67-96. 91. Pilotti S, Rilke F, Gribaude G, Ravasi GL. Sputum cytology for the diagnosis of carcinoma of the lung. Acta Cytol 1982;26:649-654. 92. Plamenac P, Nikulin A , Pikula B. Cytology of the respiratory tract in former smokers. Acta Cytol 1972;16:256-260. 93. Risse E K , van't Hof M A , Laurini RN, Vooijs GP. Sputum cytology by the Saccomanno method in diagnosing lung malignancy. Diagn Cytopathol 1985; 1:286-291. 94. Rome DS, Olson K B . Sputum specimens versus bronchial aspirates in diagnosis of bronchogenic cancer. J A M A 1957;164:167-169. 95. Russell WO, Neidhardt HW, Mountain CF, Griffith K M , Chang JP. Cytodiagnosis of lung cancer. Acta Cytol 1963;7:1-44. 96. Sabour MS, Osman L M , Glen J, Fahy T, Lamb P. Carcinoma of the lung: Review of 509 cases. Dis Chest 1962;41:530-546. 97. Saccomanno G, Saunders RP, Ellis H, Archer V E , Wood B G , Beckler PA. Concentration of carcinoma or atypical cells in sputum. Acta Cytol 1963;7:305-310. 117 Payne 98. Spjut HF, Fier DJ, Ackerman L V . Exfoliative cytology and pulmonary cancer. A histopathologic and cytopathologic correlation. J Thorac Surg 1955;30:90-107. 99. Truong L D , Underwood RD, Greenberg SD, McLarty JW. Diagnosis and typing of lung carcinomas by cytopathologic methods. A review of 108 cases. Acta Cytol 1985;29:379-384. 100. Umiker WO. Diagnosis of bronchogenic carcinoma: an evaluation of pulmonary cytology, bronchoscopy and scalene lymph node biopsy. Dis Chest 1960;37:82-90. 101. Wagner Rosa U , Prolla JC, da Silva Gastal E. Cytology in diagnosis of cancer affecting the lung. Chest 1973;63:203-307. 102. Wandall H H . A study of neoplastic cells in sputum as a contribution to the diagnosis of primary lung cancer. Acta Chir Scand (suppl) 1944;91:1-143. 103 .Booking A , Biesterfeld S, Chatelain R, Gien-Gerlach G, Esser E. Diagnosis of bronchial carcinoma on sections of paraffin-embedded sputum: sensitivity and specificity of an alternative to routine cytology. Acta Cytol 1992;36(l):37-47. 104.Papanicoloau G N , Traut HF. The diagnostic value of vaginal smears in carcinoma of the uterus. A m J Ob Gyn 1941;42:193-202. 105.Saccomanno G, Archer V , Auerbach O. Development of carcinoma of the lung as reflected in exfoliated cells. Cancer 1974;33:256-270. 106. Archer PG, Koprowska I, McDonald JR, Naylor B, Papanicolaou G N , Umiker WO. A study of variability in the interpretation of sputum cytology slides. Cancer Research 1966;6:2122-2144. 107. Browman GP, Arnold A, Levine M N , D'Souza T, Turner-Smith L M , Cayco R, Johnstone B, Birse R. Use of screening phase data to evaluate observer variation of sputum cytodiagnosis as an outcome measure in a chemoprevention trial. Cancer Research 1990;50:1216-1219. 108. Holiday DB, McLarty JW, Farley M L , Mabry L C , Cozens D, Roby T, Waldron E, Underwood RD, Anderson E, Culbreth W, Hieger LR. Sputum cytology within and across laboratories: A reliability study. Acta Cytol 1995;39(2): 195-206. 109. Band P, Feldstein M , Saccomanno G. Reversibility of bronchial marked atypia: implications for chemoprevention. Cancer Detection and Prevention: 1986:9; 157-160. Payne 118 I lO.Fidler H K , Boyes DA, Worth AJ. Cervical cancer detection in British Columbia. Br J Obstet Gynecol 1968;75:392-404. II l.Kottmeier FJL. Evolution et traitement des epitheliomas. Rev Franc Gynaecol Obstet 1961;56:821-825. 112. Burghardt E. Premalignant conditions of the cervix. Clin Obstet Gynecol 1976;3:257-295. 113. Gruner OC. Study of the changes met within the leukocytes in certain cases of malignant disease. Br J Surg 1916;3:506-522. 114. Nieburgs HE, Zak FG, Allen DC, Reisman H , Clardy T. Systemic cellular changes in material from human and animal tissues in presence of tumors. Transactions, 7th Ann Meet Inter-Soc Cytology Council pp 137. 1959. 115. Nieburgs HE, Goldberg AF, Bertini B, Silagi J, Pacheco B, Reisman H . Malignancy associated changes (MAC) in blood and bone marrow cells of patients with malignant tumors. Acta Cytol 1967;11:415-423. 116. Nieburgs HE, Parets A D , Perez V , Boudreau C. Cellular changes in liver tissue adjacent and distant to malignant tumors. Arch Pathol 1965;80:262. 117. Nieburgs HE. Recent progress in the interpretation of malignancy associated changes (MAC). Acta Cytol 1968;12:445-453. 118. Vetrani A , Marino M , Fulciniti F, Zabatta A , Palombini L , Boschi R, Marino R, Bianco P. Malignancy-associated cellular markers (MAC) in oral and bronchial epithelial cells in sputum specimens: possible morphologic marker for high-risk groups (asbestos-exposed workers). Cancer Detection and Prevention 1986;9:343-346. 119. van Oppen Toth B, Meisels A, St-Pierre S. Malignancy associated changes in sputum: a correlated study of 315 patients. Acta Cytol 1975;19(6):573-576. 120. Montag A G , Bartels PH, Dytch HE, Lerma-Puertas E, Michelassi F, Bibbo M . Karyometric features in nuclei near colonic adenocarcinoma: statistical analysis. Anal Quant Cytol Histol 1991;13:159-167. 121. MacAulay C, Lam S, Payne PW, LeRiche JC, Palcic B. Malignancy-associated changes in bronchial epithelial cells in biopsy specimens. Anal Quant Cytol Histol 1995;17(1):55-61. 122.Susnik B, Worth A, Palcic B, Poulin N , LeRiche J. Differences in quantitative nuclear features between ductal carcinoma in situ (DCIS) with and without 119 Payne accompanying invasive carcinoma in the surrounding breast. Anal Cell Pathol 1995;8:39-52. 123. Burger G, Jutting U , Rodenacker K. Changes in the benign cell population in cases of cervical cancer and its precursors. Anal Quant Cytol 1981;3:261-271. 124. Finch RR. A classification of nuclear aberration in relation to malignancy associated changes (MAC). Acta Cytol 1971;15:553-558. 125. Finch RR, von Haam E. Malignancy associated changes in buccal smears. Acta Cytol 1971;15:46-49. 126. Klawe H , Rowinski J. Malignancy associated changes (MAC) in cells of buccal smears detected by objective image analysis. Acta Cytol 1974;18:30-33. 127. Wied GL, Bartels PH, Bibbo M , Sychra JJ. Cytomorphometric markers for uterine cancer in intermediate cells. Analyt Quant Cytol 1980;2:257-263. 128. Zahniser DJ, Wong K L , Brenner JF, Ball HG, Garta GL, Hutchinson M L . Contextual analysis and intermediate cell markers enhance high resolution cell image analysis for automated cervical smear diagnosis. Cytometry 1991;12:10-14. 129. Junqueira L C , Carneiro J, Kelley RO. Basic Histology 8th ed. Appleton and Lange, Toronto, 1995. 130. Bettinger C, Zimmermann HW. New investigation on hematoxylin, hematein, and hematein-aluminum complexes. Histochem 1991;96:215-228. 131. Bales CE, Durfee GR. Cytologic Techniques. In: Koss L G , editor. Diagnostic Cytology and its Histopathologic Bases, 4th ed. Philadelphia: JB Lippincott Co.; 1992. 132. Keebler C M . Cytopreparatory Techniques. In: Bibbo M , editor. Comprehensive Cytopathology. Philadelphia: WB Saunders Co.; 1991. 133. Gi l l GW, Frost JK, Miller K A . A new formula for a half-oxidized hematoxylin solution that neither overstains nor requires differentiation. Acta Cytol 1974;18(4):300-311. 134. Flehinger BJ, Kimmel M , Polyak T, Melamed MR. Screening for Lung Cancer: The Mayo lung project revisited. Cancer 1993;72(5):1573-1580. 135.Soost H , Brem W, Hobel W, Otto K, Schwarz G. Comparison of Papnicolaou staining and Feulgen staining for automated prescreening. Anal Qunat Cytol 1983;5(l):61-65. : 120 Payne 136.Gurley A M , Hidvegi DF, Bacus JW, Bacus SS. Comparison of the Papanicolaou and Feulgen staining methods for D N A quantitation by image analysis. Cytometry 1990;11:468-474. 137.Schulte E K W Fink DK. Hematoxylin staining in quantitative D N A cytometry: an image analysis study. Anal Cell Pathol 1995;(9):257-268. 138. Feulgen R, Voit K. Uber den mechanismus der nuclealfarbung. II. Mittheilung. Uber das verhalten der kerne partiell hydrolysierter mikroskopscher praparate zur fuchsin-schwefligen saure nach voraufgegangener mit phenylhydrazin. Z Physiol Chem 1924:136:57-61. 139. Tezcan A, Garner D M , Lam P, Korbelik J, Palcic B. Analysis of thionin, gallocyanin and hematoxylin for automated quantitative image cytometry of cervical samples. 8th Annual Meeting, Clinical Applications of Cytometry, Charleston, South Carolina, September 15-18, 1993. 140. Kato H , Konaka C, Ono J, Takahashi M , Hayata Y . Cytology of the lung: techniques and interpretation. Tokyo:Igaku-Shoin, 1983. 141. Johnston WW. Ten years of respiratory cytopathology at D U M C . I. The cytopathologic diagnosis of lung cancer during the years 1970 to 1974 noting the significance of specimen number and type. Acta Cytol 1981;25:103-107. 142. Erozan Y S , Frost JK.Cytopathologic diagnosis of cancer in pulmonary material: a critical histopathological correlation. Acta Cytol 1970;14:560-565. 143. Pedersen B, Brans M , Holm K , Pedersen D, Lund C. The value of provoked expectoration in obtaining sputum samples for cytologic investigation: a prospective, conservative and controlled investigation of 134 patients. Acta Cytol 1985;29(5):750-752. 144.Saccomanno G, Sanders RP, Ellis H . Concentration of carcinoma or atypical cells in sputum. Beta Cytol 1963;7:305-310. 145. Perlman EJ, Erozan YS, Howdon A. The role of the Saccomano (sic) technique in sputum cytopathologic diagnosis of lung cancer. AJCP 1989;91(l):57-60. 146. Fleiss JL. Statistical methods for rates and proportions. 2nd ed, New York: Wiley, 1981. 147. B M D P Statistical Software Inc. BMDP. Berkley, University of California Press, 1988. Payne 121 148.Skov B G , Braendstrup O, Hirsch FR, Lauritzen AF, Nielsen HW, Skov T. Lung Cancer 1994;11:365-372. 149Bunn, Feld R, Borges M , Giner V, Ginsberg R, Harper P, Klastersky J, Lacquet L , Paesmans M , Payne D, Rosell R, Sause W, Sculier JP, Shaw E, Sorensen JB, Splinter T, Stahel R, Bunn P. Prognostic factors in non-small cell lung cancer. Lung Cancer 1994;11(3):S19-S23. 150. Hung J, Kishimoto Y , Sugio K, Virmani A , Mclntire DD, Minna JD, Gazdar A F . Allele-specific chromosome 3p deletions occur at an early stage in the pathogenesis of lung carcinoma. J A M A 1995;273(7):558-563. 151. Thiberville L , Payne P, Vielkinds J, LeRIche J, Horsman D, Nouvet G, Palcic B, Lam S. Evidence of cumulative gene losses with progression of premalignant epithelial lesions to carcinoma of the bronchus. Cancer Research 1995;55:5133-5139. 152.Schulte E K W , Wittekind DH. Standardization of the staining process as a prerequisite for successful quantitative microscopy and automated cytology screening. In Grohs H K , Husain O A N eds. Automated cervical cancer screening. Igaku-Shoin, New York, 1994. 153. Van Duijn P. A histochemical specific thionin-S02 reagent and its use in a bi-color method for deoxyribonucleic acid and periodic acid Schiff positive substances. J Histochem Cytochem 1956;4:55-63. 154. Giroud F. Standardization of specimen fixation in preparation for cytometric evaluation. In Grohs H K , Husain O A N eds. Automated cervical cancer screening. Igaku-Shoin, New York, 1991. 155. KoteInikov V M , Litinskaya L L . Comparative studies of the Feulgen hydrolysis for D N A . I. Influence of different fixatives and polyethylene glycols. Histochemistry 1981;71:145-153. 156. Makkus A C F , van't Hof-Grootenboer A E , Pahlplatz M M M , de Wilde P C M , Gemmink A J , Cuypers VMJI, Vooijs GP. Practical aspects of fixatives in high resolution nuclear image analysis. Cytometry 1994;15:302-310. 157.Swank PR, Greenberg SD, Hunter NR, Spjut HJ, Estrada R, Trahan EB, Montalvo J. Identification of features of metaplastic cells in sputum for the detection of squamous cell carcinoma of the lung. Diagn Cytopathol 1989;5:98-103. 158.Ott R L . A n introduction to statistical methods and data analysis. 4th ed. 1993, Duxbury Press, Belmont CA. Payne 122 159Karsell PR, McDougall JC. Diagnostic tests for lung cancer. Mayo Clin Proc. 1993; 68(3):288-296. 160. Melamed MR, Flehinger BJ, Zaman M B . Impact of early detection on the clinical course of lung cancer. Surgical Clinics of North America 1987; 67(5):909-924. 161. Melamed MR, Flehinger BJ. Screening for lung cancer. Letter to the editor. Ann Intern Med 1989; 111:764-765. 162. Tockman MS, Mulshine JL, Smart CR. Screening for lung cancer. Letter to the editor. Ann Intern Med 1989; 111:765-766. 163. Mulshine JL, Tockman MS, Smart CR. Considerations in the development of lung cancer screening tools. 1989; J National Cancer Institute 81(12):900-906. 164. MacAulay C, Payne P, Wilton D, Doudkine A, Lam S, Palcic B. MACs : Field cancerization or signal response. IS A C Congress XVIII, Rimini, Italy, April 13-18, 1996. 165. Lam S, MacAulay C, Hung J, LeRiche J, Profio A E , Palcic B. Detection of dysplasia and carcinoma in-situ with a lung imaging fluorescence endoscope device. J Thorac Cardiovasc Surg 1993; 105(6): 1035-1040. 166. Ebeling K , Nischan P. Screening for lung cancer—results from a case-control study. Int J Cancer 1987:40; 141-144. 167. Early Lung Cancer Coopeative Study Group. Early lung cancer detection: Summary and conclusions. A m Rev Respir Dis 1984:130;565-570. 168. Eddy D M . Screening for lung cancer. Annals of Internal Medicine, 1989:lll(3);232-237. 169. Fry W A , Menck HR, Winchester DP. The national cancer data base report on lung cancer. Cancer, 1996:77(9); 1947-1955. 170. Fontana RS, Sanderson DR, Woolner L B , Taylor WF, Miller WE, Muhm JR. Lung cancer screening: the Mayo program. 1986; J Occup Med 28:746-750. 171. Strauss G M , Gleason RE, Sugarbaker DJ. Chest x-ray screening improves outcome in lung cancer. A reappraisal of randomized trials on lung cancer screening. 1995 Chest: 107(6);270s-279s. 172. Risse E K J , Vooijs GP, van't Hof M A . Diagnostic significance of "severe dysplasia" in sputum cytology. Acta Cytol 1988:32(5);629-634. Payne 123 173.Mikel U V , Fishbein W N , Bahr GF. Some practical considerations in quantitative absorbance microspectrophotometry: preparation techniques in D N A cytophotometry. Anal Quant Cytol Histol 1985:7(2);108-118. 174.Schulte E K W . Standardization of the Feulgen reaction for absorption D N A image cytometry: a review. Anal Cell Path 1991:3;167-182. 175.Taubert G, Krug H. Hydrolysis of T C A Feulgen reaction on lymphocytes, liver cells, and Ehrlich ascites tumour cells. Acta Histochem 1986:79;223-228. 176.Soost HJ, Brem W, Hobel W, Otto K , Schwarz G. Comparison of Papanicolaou staining and Feulgen staining for automated prescreening. Anal Quant Cytol 1983:5(l);61-65. 177. Jaggi B, Palcic B. The design and development of an optical scanner for cell biology. IEEE Proc Eng Med Biol 1985:2;980-985. 178. Poon SSS, Jaggi B, Palcic B. Cell recognition algorithms for the cell analyzer. IEEE Proc Eng Med Biol 1987:4:1454-1456. 179. Jaggi B , Deen MJ, Palcic B. Design of a solid state microscope. Opt Engineer 1989:28(6):675-682. 180. Bacus JW, Bacus JV. A method of correcting D N A ploidy measurements in tissue sections. Modern Pathology 1994:7(6);652-664. 181. Poulin N , Hayes M , Susnik B, Guillaud M , Palcic B . Heterogeneity of image cytometric features in ductal carcinoma in situ of the breast. Cytometry 1996:8s; 128. 182. Karabulut A , Reibel J, Therkildsen M H , Praetorius F, Nielsen HW, Dabelsteen E. Observer variability in the histologic assessment of oral premalignant lesions. J Oral Path Med 1995:24(5); 198-200. 183. Jensen P, Krogsgaard MR, Christiansen J, Braendstrup O, Johansen A, Olsen J. Observer variability in the assessment of type and dysplasia of colorectal adenomas, analyzed using kappa statistics. Diseases of the Colon & Rectum 1995:38(2);195-198. 184. Young N A , Naryshkin S, Atkinson BF, Ehya H , Gupta PK, Kline TS, Luff RD. Interobserver variability of cervical smears with squamous-cell abnormalities: a Philadelphia study. Diagnostic Cytopathology 1994:11(4);352-357. 185. Bengtsson E, Nordin B. Densitometry, morphometry, and textural analysis as tools in quantitative cytometry and automated cancer screening. In Grohs H K , Husain O A N eds. Automated cervical cancer screening. Igaku-Shoin, New York, 1994. 124 Payne 186. Doudkine A, MacAulay C, Poulin N , Palcic B. Nuclear texture measurements in image cytometry. Pathologica 1995:87;286-299. 187. MacAulay C. Development, implementation and evaluation of segmentation algorithms for the automated classification of cervical cells. PhD Thesis. 1989, University of British Columbia. 188. Lam S, MacAulay C, LeRiche JC, Ikeda N , Palcic B. Early localization of bronchogenic carcinoma. Diagnostic and Theraputic Endoscopy 1994:l(2);75-78. 189Payne P, Thiberville L , MacAulay C, Lam S, LeRiche JC, Palcic B. Quantitative microscopy of Feulgen stained bronchial biopsies detects loss of heterozygosity in pre-cancerous lesions. Cytometry 1996:8s;58. 190. Van den Eeden SK, Friedman GD. Forced expiratory volume (1 second) and lung cancer incidence and mortality. Epidemiology 1992:3(3);253-257. 191. Mao L , Hruban RH, Boyle JO, Tockman M , Sidransky D. Detection of oncogene mutations in sputum precedes diagnosis of lung cancer. Cancer Research 1994:54(7);1634-1637. 192. Tockman MS, Gupta PK, Myers JD, Frost JK, Baylin SB, Gold EB, Chase A M , Wilkinson PH, Mulshine JL. Sensitive and specific monoclonal antibody recognition of human lung cancer antigen on preserved sputum cells: a new approach to early lung cancer detection. J Clin One 1988:6(11); 1685-1693. 193. Miozzo M , Sozzi G, Dongi R, Pilotti S, Radice P, Pastorino G M , Pierotti M A . Genetice evidence of 'field cancerizaion' in multiple lung cancer patients. Second International Cancer Chemo Prevention Conference April 28-30, Berlin. 1993:pl04. 194. Atkin N B , Richards B M . Deoxyribonucleic acid in human tumours as measured by microspectrophotometry of Feulgen stain: a comparison of tumours arising at different sites. Br J Cancer 1956;10:769-786. 195. Kinderamnn D, Hilgers CH. Glare-correction in D N A image cytometry. Anal C l l Path 1994;6:165-180. 196. Bartels PH. Numerical evaluation of cytological data. VII Multivariate significance tests. Analytical and Quantitative Cytology. 1981:3(1); 1-8. 197. Neter J, Wasserman W, Whitmore GA. Applied Statistics. 3rd ed. 1988, Allyn and Bacon, Toronto. Payne 125 198.Statistica Vol III. Ch 2. Discriminant function analysis. pp3055-3099. StatSoft Inc., Tulsa OK, 1994. 126 Payne 

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:
https://iiif.library.ubc.ca/presentation/dsp.831.1-0088262/manifest

Comment

Related Items