- Email: [email protected]
Morphometric differentiation between responsive tumor cells and mesothelial hyperplasia in second-look operations for ovarian cancer
Morphometric Differentiation Between Responsive Tumor Cells and Mesothelial Hyperplasia in Second-Look Operations for Ovarian Cancer L, DELIGDISCH, MD, H. KERNER, MD, C. J. COHEN, MD, D. DARGENT, MD, AND J. GIL, MD We developed a procedure based on computerized image analysis to establish objective criteria for the differential diagnosis between mesothelial hyperplasia and cancer in peritoneal tissue samples obtained at second-look operations for ovarian cancer. The tumor tissue after chemotherapy was classified as “nonresponsive” if it was found by histologic criteria to be roughly similar to the tumor before chemotherapy and as “responsive” if it was found to be different (small clusters of bland-looking cells with no mitotic activity). Eighty-five samples of tissue had been classified previously by a pathologist into one of the four following groups: ovarian tumor prior to chemotherapy,“responsive” tumor, “nonresponsive” tumor, or mesothelialhyperplasia Cell profiles of the tissue samples were studied by computerized image analysis using 21 morphometric descriptors derived from the nuumaltracingsof tumor nuclei, including nuclear perimeter, nuclear area, maximal chord, circularity factor, and standard deviations of these descriptors. Size distribution curves of nuclear areas and maximal chords were included in the analysis. A multivariate discriminant analysis confirmed the separation into the four diagnostic groups, accomplished with consideration of the physical descriptors alone, except for some overlapping between groups 1 and 3. The separationbetween carcinoma and mesothelial hyperplasia was clear in all cases. HUMPATHOL24: 143-147. Copyright 0 1993 by W.B. Saunders Company
On second-look operations for ovarian carcinoma, tumor tissue is often found to be histologically different from the original tumor. Clusters of “bland-appearing” malignant cells resembling mesothelial hyperplasia may create a diagnostic dilemma. Current literature emphasizes the use of image analysis for densitometric applications, such as DNAbased cell cycle studies or quantitation of oncoprotein concentration. This method of analysis is applied to prognostic rather than diagnostic studies. For this study we used a morphometric type of computerized image analysis based on extraction of physical characteristics of nuclear cell profiles (such as profile area, perimeter length, and shape) as viewed in hematoxylin-eosinstained slides. By means of this technique we established objective criteria for the differential diagnosis between From the Departments of Pathology and Obstetrics-Gynecology and Reproductive Science, Mount Sinai School of Medicine, New York, NY; the Department of Pathology, Rambam Hospital, Haifa, Israel; and the Department of Obstetrics-Gynecology, Hopital Edouard Herriot, Lyon, France. Accepted for publication May 5, 1992. Kq words: morphometry, second-look operation, ovarian cancer, chemotherapy, mesothelial hyperplasia. Address correspondence and reprint requests to L. Deligdisch, MD, Department of Pathology, Box 1194, Mt Sinai Medical Center, One Gustave L. Levy Place, New York, NY 10029-6574. Copyright 0 1993 by W.B. Saunders Company 0046~8177/93/2402-0005$5.00/O
143
mesothelial hyperplasia and responding and nonresponding ovarian cancer tissue and we developed a database against which difficult cases can be tested by the means of classification statistics.
MATERIALS
AND METHODS
Fifty cells from 85 samples of tissue were studied by morphometry. The 85 samples had been classified as belonging to four diagnostic categories. Group 1 (35 samples) consisted of ovarian metastatic peritoneal tumor tissue before chemotherapy (Fig 1). Group 2 (14 samples) comprised tumor tissue obtained at second-look operations that was different from the original tumor before chemotherapy. The tissue consisted of microscopic clusters of tumor cells often arranged around psammoma bodies showing absence of mitotic activity and with a bland, relatively regular chromatin pattern; these clusters were considered to be tumor cells responsive to chemotherapy (Fig 2). Group 3 (24 specimens) was composed of tumor tissue obtained at second-look operations demonstrating a histologic pattern roughly similar to that before chemotherapy (Fig 3). Group 4 (12 specimens) comprised peritoneal tissue obtained from patients who underwent operation for benign conditions with hyperplastic mesothelial cells (Fig 4). The patients from groups 1, 2, and 3 were treated with standard first-time chemotherapy regimens based on cisplatin with the addition of either Adriamycin (Adria Laboratories, Columbus, OH) or cyclophosphamide. Morphometric work was done in a self-assembled interactive system that we have used previously.‘-3 Nuclei of interest were traced on a touch-sensitive screen superimposed over a video monitor screen displaying cell images originated in a camera mounted on a microscope. The following descriptors were extracted from the tracings: (1) length of the nuclear perimeter, (2) enclosed area of the nuclear profile, (3) maximal chord (diameter) of the profile, (4) so-called Ll, which is the imaginary perimeter of a perfect circle with the same area as the traced profile, and (5) circularity factor, a non-unique shape descriptor obtained by dividing Ll by the actual perimeter length. Finally, descriptors 6,7,8,9, and 10 were the standard deviations of the descriptors labeled 1 through 5. This was supplemented by selected bins of the size distribution curves of the maximal chord (descriptors 11, 12, 13, 14, and 15) from 3 to 10.5 pm, step size 1.5 pm, and also by selected bins of the size distribution of nuclear areas from 10 to 70 pm2, step size 10 pm2 (descriptors 16, 17, 18, 19, 20, and 21). In our previous experience, size distribution tables, which are comparable to fingerprints, have proven to be very powerful discriminators. The reason for limiting the analysis to the bins in which numeric values were present for all four groups was the operational difficulty encountered by the cluster analysis with bins of value “0.”
HUMAN PATHOLOGY
Volume 24, No. 2 (February
1993)
FIQURE 1. Peritoneal implant of ovarian carcinoma before chemotherapy. Note some extremely large nuclei and mitotic activity. (Hematoxylin-eosin stain; magniflcation X 100.)
These 21 descriptors are somewhat mutually dependent because they are all derived from the nuclear tracing; the multivariate discriminant program determined the homogeneity, internal consistency, and reality of the four groups on the basis of the 21 descriptors discussed above. We used the commercial discriminant program from the STAT/LIB by IMSL (Houston, TX). The theory of this approach, borrowed from the field of pattern analysis,4*5has been previously described as applied to our material.‘-s’6.7
RESULTS Table 1 shows the average values of the descriptors used (nuclear perimeter length, area, maximal chord,
and circularity) and Fig 5 shows the size distribution of the areas. Table 2 shows the classification results according to posterior probabilities between group 4 (mesothelial hyperplasia) and group 2 (responsive tumors). Correct classification was confirmed in all cases with high posterior probability. Groups 1 (untreated tumors) and 3 (nonresponsive tumors) showed great similarity with overlap of many individual descriptors, and a number of misclassifications occurred (Tables 1 and 2). As a whole, the differentiation was possible in most cases, but unconvincingly so due to low posterior probabilities. For instance, a
FIQURE 2. Cluster of “responsive” tumor cells adjacent to psammoma bodies, with relatively bland nuclei and no mitotic activity. (Hematoxylin-eosin stain; magnification X40.)
MORPHOMETRY
IN OVARIAN
CANCER
(Deligdlsch
et al)
FIOURE 3. Same patient as in Fig I, after chemotherapy. Residual “nonresponsive” tumor cells after chemotherapy, quite similar to those in Fig I, except for the largest nuclei. (Hematoxylin-eosin stain; magnification X 100.)
group 1 case received a probability value of .501 for inclusion in group 1 and a probability value of .499 for inclusion in group 3. The reason for this lack of clear separation is apparent when examining both the average values of the descriptors (Table 1) and the size distribution plots of the area (Fig 5). The similarity between groups 1 and 3 is evident. A remarkable difference, however, was the existence of many large nuclear profiles in the last bin of the untreated tumor. Those largest profiles had disappeared after treatment, even in otherwise nonresponsive tumors. In addition, the average
FIWRE 4. Mesothelial hyperplasia from benign peritoneal tissue. (Hematoxylin-eosin stain; magniflcation X40.)
nuclear area was 58.20 pm in group 1 compared with 53.24 pm in group 3. Of great importance was the clear separation between carcinoma and mesothelial hyperplasia, which was achieved in all cases. The size distribution characteristics of these two groups, as shown in Fig 5, differ markedly. Note in particular the early, narrow, and tall peak (bin 10 to 20 pm) in mesothelial hyperplasia, which in our previous experience is characteristic of benign cells. The size distribution of malignant cells shows a shift of the highest bin to the right along with less tall and broader
Volume 24, No. 2 (February
HUMAN PATHOLOGY
TABLE 1.
Averages
Nuclear perimeter length brn) Nuclear area profile km*) Maximal diameter of nuclear profile km) Circularity of profile
of Main
Physical
Descriptors
1993)
of Nuclear
Profiles
Group 1
Group 2
Group 3
Group 4
27.97 t 4.34 58.20 + 18.30 10.37 f 1.69 1.06
22.40 t 2.19 35.26 f 7.41 8.98 f 0.81 1.09
27.20 t 2.98 53.24 +- 11.12 10.19 f 1.30 1.08
17.93 t 2.26 21.64 f 5.7 7.00 + 0.98 1.11
Note: Note the similarity between group 1 (untreated tumor) and group 3 (nonresponsive tumor) and the clear differences between group 2 (responsive carcinoma) and group 4 (mesothelial hyperplasia).
dieting the response to chemotherapy for ovarian cancer. The present study focused on the measurements of nuclear profiles in cases of treated ovarian cancer as compared with cases of nontreated ovarian cancer and with mesothelial hyperplasia. The averages values of groups 1 and 3 were close to those previously published for metastatic ovarian carcinoma.3 The average values of group 4 ranged between those of normal and hyperplastic mesothelial cells in our previous publication3 and came closest to, although definitely discriminated from (by the multivariate analysis), group 2 samples that represented the “responsive” carcinomas. The histopathologic description of the latter by conventional means is that of “shrinking” tumor cells, found randomly in small aggregates throughout the peritoneal cavity, with no mitotic activity, and a rather regular chromatin pattern, often seen in the proximity of psammoma bodies. Pathologists are often faced with the diagnostic dilemma of whether these cells represent “responsive” tumor cells or mesothelial hyperplastic cells. The morphometric analysis enables us to discriminate between the two. The fact that nontreated and treated but nonresponsive tumor samples often show overlapping of nuclear profiles is not surprising because even if chemotherapy manages to destroy tumor cells and to reduce the volume of the cancer mass (cytoreduction), new clones of cancer cells may proliferate and display features similar to the nontreated cells. It appears, however, that the largest nuclei were found in the nontreated cancer specimens, thus suggesting that the most anaplastic cells may be more sensitive to chemotherapy. The “responsive” tumor cells often are seen in patients with longer survival and disease-free intervals.
peaks. In the past we consistently have found this to be characteristic of epithelial malignancies. DISCUSSION Recent literature has emphasized the quantitative approach to the classification and prognostic value of ovarian cancer. ‘-ii In our previous studies we have used methods measuring the variable descriptors of nuclear size and shape, subjecting them to multivariate discriminant analysis. l-3,6 This is an application of a statistical technique widely used in the fiqd of pattern recognition and in our case serves to determine the internal consistency of the groups chosen by a pathologist.4 A high posterior probability in a large series means that the computer would have classified a case identically with the pathologist. Once this has been achieved, computerized classification of unknowns becomes possible by methods (k-nearest neighbor or neural networks) that compare the new case with the database and decide what group is most similar to the unknown. In previous publications1*3 we have established a database for primary and metastatic ovarian carcinoma and for mesothelial cell hyperplasia. Some investigators” have studied the value of morphometric studies pre-
0.45 : 0.40 Z 0.35 5 0.30 2 0.25 d 0 0.20 “0 0.15 I ~
TABLE 2. %
IO 20
30
40
50
60
70
80
90
Classification
91 291
Results Classification by Posterior Probabilities
MICRONS2 Classification by Pathologist
FIQURE 5. Second-look ovarian carcinoma. Size distribution of nuclear proflles. The frequency of tumor cells (abscissa) has been normalized to add up to 1 in each group. Ordinate is the size of the proftles in square microns. Note the similarity between nonresponsive and untreated carcinoma. Of interestis the rise of the last bin (nuclear area > than 91 pm*), the so-called “malignant tail,” in untreated carcinomas. In addltlon, note the great differences between mesothelial hyperplasia (a tall, early peak at 11 to 20 pm*) and responsive carcinoma. The latter, although characteristically malignant, also differs from the untreated and unresponsive groups.
Group Group Group Group
1: 2: 3: 4:
untreated tumors (35) responsive tumors (14) unresponsive tumors (24) mesothelial hyperplasia (12)
1
2
3
4
27 0 5 0
0 14 0 0
8 0 19 0
0 0 0 12
Note: Note the misclassifications between groups 1 and 3 reflecting the morphologic similarity between tumor cells of both groups. No misclassifications occurred between any other diagnostic categories, in particular not between groups 2 and 4.
146
MORPHOMETRY
IN OVARIAN
Correlation between morphometric data and clinical follow-up of the individual patients is the subject of ongoing studies. The morphometric study of the four above-described groups proved useful for the differential diagnosis of mesothelial hyperplasia and carcinoma, and offered an insight into the morphology of chemotherapy-treated ovarian cancer cells.
CANCER
(Deligdisch et al)
5. Bishop YMM, Fienberg SE, Holland PW: Discrete Multivariate Analysis: Theory and Practice. Cambridge, MA, Massachusetts Institute of Technology, 1975 6. Gil J: Editorial: Image analysis in anatomic pathology. What are the issues? HUM PATHOL 20:203-204, 1989 7. Hytiroglou P, Harpaz N, Heller DS, et al: Differential diagnosis of borderline and invasive serous cystadenocarcinomas of the ovary by computerized interactive morphometric analysis of nuclear features. Cancer 69:988-992, 1992 8. Baak JP: Possibilities and progress of quantitative pathological examination of ovarian tumors. Eur J Obstet Gynecol Reprod Biol 29:183-189, 1988 9. Schipper NW, Smeulders AW, Baak JP: Evaluation of automated estimation of epithelial volume and its prognostic value in ovarian tumors. Lab Invest 61:228-234, 1989 10. Miller BE, Lavia LA, Horbelt DV: The prognostic value of image analysis in ovarian cancer. Cancer 67: 13 18-l 32 1, 199 1 11. Haapasolo H, Collan Y, Montironi R, et al: Consistency of quantitative methods in ovarian tumor histopathology. Int J Gynecol Path01 9:203-216, 1990 12. Weger AR, Ludescher C, Mikuz G, et al: The value of morphometry to predict chemotherapy response in advanced ovarian cancer. Path01 Res Pratt 185:676-670, 1989
REFERENCES 1. Deligdisch L, Gil J: Characterization of ovarian dysplasia by interactive morphometry. Cancer 63:136-143, 1989 2. Gil J, Deligdisch L: Interactive morphometric procedures and statistical analysis in the diagnosis of ovarian dysplasia. Path01 Res Pratt 185:680-685, 1989 3. Deligdisch L, Heller D, Gil J: Interactive morphometry of normal and hyperplastic peritoneal mesothelial cells and dysplastic and malignant ovarian cells. HUM PATHOL 21:218-222, 1990 4. Duda RO, Mart PE: Pattern Classification and Scene Analysis. New York, Wiley-Interscience, 1973
147
On second-look operations for ovarian carcinoma, tumor tissue is often found to be histologically different from the original tumor. Clusters of “bland-appearing” malignant cells resembling mesothelial hyperplasia may create a diagnostic dilemma. Current literature emphasizes the use of image analysis for densitometric applications, such as DNAbased cell cycle studies or quantitation of oncoprotein concentration. This method of analysis is applied to prognostic rather than diagnostic studies. For this study we used a morphometric type of computerized image analysis based on extraction of physical characteristics of nuclear cell profiles (such as profile area, perimeter length, and shape) as viewed in hematoxylin-eosinstained slides. By means of this technique we established objective criteria for the differential diagnosis between From the Departments of Pathology and Obstetrics-Gynecology and Reproductive Science, Mount Sinai School of Medicine, New York, NY; the Department of Pathology, Rambam Hospital, Haifa, Israel; and the Department of Obstetrics-Gynecology, Hopital Edouard Herriot, Lyon, France. Accepted for publication May 5, 1992. Kq words: morphometry, second-look operation, ovarian cancer, chemotherapy, mesothelial hyperplasia. Address correspondence and reprint requests to L. Deligdisch, MD, Department of Pathology, Box 1194, Mt Sinai Medical Center, One Gustave L. Levy Place, New York, NY 10029-6574. Copyright 0 1993 by W.B. Saunders Company 0046~8177/93/2402-0005$5.00/O
143
mesothelial hyperplasia and responding and nonresponding ovarian cancer tissue and we developed a database against which difficult cases can be tested by the means of classification statistics.
MATERIALS
AND METHODS
Fifty cells from 85 samples of tissue were studied by morphometry. The 85 samples had been classified as belonging to four diagnostic categories. Group 1 (35 samples) consisted of ovarian metastatic peritoneal tumor tissue before chemotherapy (Fig 1). Group 2 (14 samples) comprised tumor tissue obtained at second-look operations that was different from the original tumor before chemotherapy. The tissue consisted of microscopic clusters of tumor cells often arranged around psammoma bodies showing absence of mitotic activity and with a bland, relatively regular chromatin pattern; these clusters were considered to be tumor cells responsive to chemotherapy (Fig 2). Group 3 (24 specimens) was composed of tumor tissue obtained at second-look operations demonstrating a histologic pattern roughly similar to that before chemotherapy (Fig 3). Group 4 (12 specimens) comprised peritoneal tissue obtained from patients who underwent operation for benign conditions with hyperplastic mesothelial cells (Fig 4). The patients from groups 1, 2, and 3 were treated with standard first-time chemotherapy regimens based on cisplatin with the addition of either Adriamycin (Adria Laboratories, Columbus, OH) or cyclophosphamide. Morphometric work was done in a self-assembled interactive system that we have used previously.‘-3 Nuclei of interest were traced on a touch-sensitive screen superimposed over a video monitor screen displaying cell images originated in a camera mounted on a microscope. The following descriptors were extracted from the tracings: (1) length of the nuclear perimeter, (2) enclosed area of the nuclear profile, (3) maximal chord (diameter) of the profile, (4) so-called Ll, which is the imaginary perimeter of a perfect circle with the same area as the traced profile, and (5) circularity factor, a non-unique shape descriptor obtained by dividing Ll by the actual perimeter length. Finally, descriptors 6,7,8,9, and 10 were the standard deviations of the descriptors labeled 1 through 5. This was supplemented by selected bins of the size distribution curves of the maximal chord (descriptors 11, 12, 13, 14, and 15) from 3 to 10.5 pm, step size 1.5 pm, and also by selected bins of the size distribution of nuclear areas from 10 to 70 pm2, step size 10 pm2 (descriptors 16, 17, 18, 19, 20, and 21). In our previous experience, size distribution tables, which are comparable to fingerprints, have proven to be very powerful discriminators. The reason for limiting the analysis to the bins in which numeric values were present for all four groups was the operational difficulty encountered by the cluster analysis with bins of value “0.”
HUMAN PATHOLOGY
Volume 24, No. 2 (February
1993)
FIQURE 1. Peritoneal implant of ovarian carcinoma before chemotherapy. Note some extremely large nuclei and mitotic activity. (Hematoxylin-eosin stain; magniflcation X 100.)
These 21 descriptors are somewhat mutually dependent because they are all derived from the nuclear tracing; the multivariate discriminant program determined the homogeneity, internal consistency, and reality of the four groups on the basis of the 21 descriptors discussed above. We used the commercial discriminant program from the STAT/LIB by IMSL (Houston, TX). The theory of this approach, borrowed from the field of pattern analysis,4*5has been previously described as applied to our material.‘-s’6.7
RESULTS Table 1 shows the average values of the descriptors used (nuclear perimeter length, area, maximal chord,
and circularity) and Fig 5 shows the size distribution of the areas. Table 2 shows the classification results according to posterior probabilities between group 4 (mesothelial hyperplasia) and group 2 (responsive tumors). Correct classification was confirmed in all cases with high posterior probability. Groups 1 (untreated tumors) and 3 (nonresponsive tumors) showed great similarity with overlap of many individual descriptors, and a number of misclassifications occurred (Tables 1 and 2). As a whole, the differentiation was possible in most cases, but unconvincingly so due to low posterior probabilities. For instance, a
FIQURE 2. Cluster of “responsive” tumor cells adjacent to psammoma bodies, with relatively bland nuclei and no mitotic activity. (Hematoxylin-eosin stain; magnification X40.)
MORPHOMETRY
IN OVARIAN
CANCER
(Deligdlsch
et al)
FIOURE 3. Same patient as in Fig I, after chemotherapy. Residual “nonresponsive” tumor cells after chemotherapy, quite similar to those in Fig I, except for the largest nuclei. (Hematoxylin-eosin stain; magnification X 100.)
group 1 case received a probability value of .501 for inclusion in group 1 and a probability value of .499 for inclusion in group 3. The reason for this lack of clear separation is apparent when examining both the average values of the descriptors (Table 1) and the size distribution plots of the area (Fig 5). The similarity between groups 1 and 3 is evident. A remarkable difference, however, was the existence of many large nuclear profiles in the last bin of the untreated tumor. Those largest profiles had disappeared after treatment, even in otherwise nonresponsive tumors. In addition, the average
FIWRE 4. Mesothelial hyperplasia from benign peritoneal tissue. (Hematoxylin-eosin stain; magniflcation X40.)
nuclear area was 58.20 pm in group 1 compared with 53.24 pm in group 3. Of great importance was the clear separation between carcinoma and mesothelial hyperplasia, which was achieved in all cases. The size distribution characteristics of these two groups, as shown in Fig 5, differ markedly. Note in particular the early, narrow, and tall peak (bin 10 to 20 pm) in mesothelial hyperplasia, which in our previous experience is characteristic of benign cells. The size distribution of malignant cells shows a shift of the highest bin to the right along with less tall and broader
Volume 24, No. 2 (February
HUMAN PATHOLOGY
TABLE 1.
Averages
Nuclear perimeter length brn) Nuclear area profile km*) Maximal diameter of nuclear profile km) Circularity of profile
of Main
Physical
Descriptors
1993)
of Nuclear
Profiles
Group 1
Group 2
Group 3
Group 4
27.97 t 4.34 58.20 + 18.30 10.37 f 1.69 1.06
22.40 t 2.19 35.26 f 7.41 8.98 f 0.81 1.09
27.20 t 2.98 53.24 +- 11.12 10.19 f 1.30 1.08
17.93 t 2.26 21.64 f 5.7 7.00 + 0.98 1.11
Note: Note the similarity between group 1 (untreated tumor) and group 3 (nonresponsive tumor) and the clear differences between group 2 (responsive carcinoma) and group 4 (mesothelial hyperplasia).
dieting the response to chemotherapy for ovarian cancer. The present study focused on the measurements of nuclear profiles in cases of treated ovarian cancer as compared with cases of nontreated ovarian cancer and with mesothelial hyperplasia. The averages values of groups 1 and 3 were close to those previously published for metastatic ovarian carcinoma.3 The average values of group 4 ranged between those of normal and hyperplastic mesothelial cells in our previous publication3 and came closest to, although definitely discriminated from (by the multivariate analysis), group 2 samples that represented the “responsive” carcinomas. The histopathologic description of the latter by conventional means is that of “shrinking” tumor cells, found randomly in small aggregates throughout the peritoneal cavity, with no mitotic activity, and a rather regular chromatin pattern, often seen in the proximity of psammoma bodies. Pathologists are often faced with the diagnostic dilemma of whether these cells represent “responsive” tumor cells or mesothelial hyperplastic cells. The morphometric analysis enables us to discriminate between the two. The fact that nontreated and treated but nonresponsive tumor samples often show overlapping of nuclear profiles is not surprising because even if chemotherapy manages to destroy tumor cells and to reduce the volume of the cancer mass (cytoreduction), new clones of cancer cells may proliferate and display features similar to the nontreated cells. It appears, however, that the largest nuclei were found in the nontreated cancer specimens, thus suggesting that the most anaplastic cells may be more sensitive to chemotherapy. The “responsive” tumor cells often are seen in patients with longer survival and disease-free intervals.
peaks. In the past we consistently have found this to be characteristic of epithelial malignancies. DISCUSSION Recent literature has emphasized the quantitative approach to the classification and prognostic value of ovarian cancer. ‘-ii In our previous studies we have used methods measuring the variable descriptors of nuclear size and shape, subjecting them to multivariate discriminant analysis. l-3,6 This is an application of a statistical technique widely used in the fiqd of pattern recognition and in our case serves to determine the internal consistency of the groups chosen by a pathologist.4 A high posterior probability in a large series means that the computer would have classified a case identically with the pathologist. Once this has been achieved, computerized classification of unknowns becomes possible by methods (k-nearest neighbor or neural networks) that compare the new case with the database and decide what group is most similar to the unknown. In previous publications1*3 we have established a database for primary and metastatic ovarian carcinoma and for mesothelial cell hyperplasia. Some investigators” have studied the value of morphometric studies pre-
0.45 : 0.40 Z 0.35 5 0.30 2 0.25 d 0 0.20 “0 0.15 I ~
TABLE 2. %
IO 20
30
40
50
60
70
80
90
Classification
91 291
Results Classification by Posterior Probabilities
MICRONS2 Classification by Pathologist
FIQURE 5. Second-look ovarian carcinoma. Size distribution of nuclear proflles. The frequency of tumor cells (abscissa) has been normalized to add up to 1 in each group. Ordinate is the size of the proftles in square microns. Note the similarity between nonresponsive and untreated carcinoma. Of interestis the rise of the last bin (nuclear area > than 91 pm*), the so-called “malignant tail,” in untreated carcinomas. In addltlon, note the great differences between mesothelial hyperplasia (a tall, early peak at 11 to 20 pm*) and responsive carcinoma. The latter, although characteristically malignant, also differs from the untreated and unresponsive groups.
Group Group Group Group
1: 2: 3: 4:
untreated tumors (35) responsive tumors (14) unresponsive tumors (24) mesothelial hyperplasia (12)
1
2
3
4
27 0 5 0
0 14 0 0
8 0 19 0
0 0 0 12
Note: Note the misclassifications between groups 1 and 3 reflecting the morphologic similarity between tumor cells of both groups. No misclassifications occurred between any other diagnostic categories, in particular not between groups 2 and 4.
146
MORPHOMETRY
IN OVARIAN
Correlation between morphometric data and clinical follow-up of the individual patients is the subject of ongoing studies. The morphometric study of the four above-described groups proved useful for the differential diagnosis of mesothelial hyperplasia and carcinoma, and offered an insight into the morphology of chemotherapy-treated ovarian cancer cells.
CANCER
(Deligdisch et al)
5. Bishop YMM, Fienberg SE, Holland PW: Discrete Multivariate Analysis: Theory and Practice. Cambridge, MA, Massachusetts Institute of Technology, 1975 6. Gil J: Editorial: Image analysis in anatomic pathology. What are the issues? HUM PATHOL 20:203-204, 1989 7. Hytiroglou P, Harpaz N, Heller DS, et al: Differential diagnosis of borderline and invasive serous cystadenocarcinomas of the ovary by computerized interactive morphometric analysis of nuclear features. Cancer 69:988-992, 1992 8. Baak JP: Possibilities and progress of quantitative pathological examination of ovarian tumors. Eur J Obstet Gynecol Reprod Biol 29:183-189, 1988 9. Schipper NW, Smeulders AW, Baak JP: Evaluation of automated estimation of epithelial volume and its prognostic value in ovarian tumors. Lab Invest 61:228-234, 1989 10. Miller BE, Lavia LA, Horbelt DV: The prognostic value of image analysis in ovarian cancer. Cancer 67: 13 18-l 32 1, 199 1 11. Haapasolo H, Collan Y, Montironi R, et al: Consistency of quantitative methods in ovarian tumor histopathology. Int J Gynecol Path01 9:203-216, 1990 12. Weger AR, Ludescher C, Mikuz G, et al: The value of morphometry to predict chemotherapy response in advanced ovarian cancer. Path01 Res Pratt 185:676-670, 1989
REFERENCES 1. Deligdisch L, Gil J: Characterization of ovarian dysplasia by interactive morphometry. Cancer 63:136-143, 1989 2. Gil J, Deligdisch L: Interactive morphometric procedures and statistical analysis in the diagnosis of ovarian dysplasia. Path01 Res Pratt 185:680-685, 1989 3. Deligdisch L, Heller D, Gil J: Interactive morphometry of normal and hyperplastic peritoneal mesothelial cells and dysplastic and malignant ovarian cells. HUM PATHOL 21:218-222, 1990 4. Duda RO, Mart PE: Pattern Classification and Scene Analysis. New York, Wiley-Interscience, 1973
147