May 1, 1990 - Fifty-eight Feulgen-stained imprint smears from freshly resected colorectal tissue were analyzed by means of a SAMBA 200 cell image ...
Characterization of Human Colorectal Mucosa, Polyps, and Cancers by Means of Computerized Morphonuclear Image Analyses Alain Verhest, MD,” Robert Kiss, PhD,t Dominique d’Olne, MD,$ Denis Larsimont, MD,$ Isabelle Salmon, MD,” Yvan de Launoit, MSc,t Catherine Fourneau, MD,$ Jean-Lambert Pasteels, MD,? and Jean-Claude Pector, MDS
Fifty-eight Feulgen-stained imprint smears from freshly resected colorectal tissue were analyzed by means of a SAMBA 200 cell image processor to establish a quantitative grading of their evolution from normal to malignant mucosa on the basis of 15 morphonuclear parameters relative to morphometry (nuclear size), densitometry (DNA content), and texture (chromatin pattern). The colorectal samples belonged to six groups: normal mucosa from patients without (Group 1)or with (Group 2) colorectal cancer, adenomas (Group 3), and cancers corresponding to the three stages of the Dukes’ classification (Groups 4 to 6).Results indicated that analysis of the morphonuclear parameters computed on 250 to 300 nuclei/cases objectively and quantitatively showed the adenoma-cancer sequence. This need for only a small number of nuclei to assess a highly efficient analysis created a preoperative investigative tool using cytologic smears during endoscopy. The authors also made preliminary data banks for objective and reproducible grading of unknown cases by comparisons (discriminant analyses) with their contents. This approach must be validated prospectively on a large series of cases to furnish a system for colorectal malignancy diagnosis. Cancer 65:2047-2054,1990.
C
carcinoma is one of the most prevalent cancers in the western world.’ Banner et al.’ state that although numerous prognostic factors have been established for colon they are reported inconsistently4 and often inaccurately with considerable variability among observer^.^ Accordingly, Schwartz et ~ l .in, a~ special report on pathology and probabilities, found that pathologists typically give their opinions in a qualitatively, i.e., using ambiguous language, when anaOLORECTAL
From the *Department of Pathology, Hospital Erasme, ?Laboratory of Histology, Faculty of Medicine, Free University of Brussels, SDepartment of Pathology, Institute J. Bordet and Hospital St. Pierre, and SDeptartment of Surgery, Institute J. Bordet, Brussels, Belgium. Supported by the “Fonds de la Recherche Scientifique Mtdicale” (FRSM, Belgium), the “Association Sportive Contre le Cancer” (ASCC, Belgium), and the “Fondation Anticancer R.O.S.E.”. Address for reprints: Robert Kiss, PhD, FNRS, Laboratorie DHistologie, Faculte de Medecine, Rue Evers 2, B-1000, Brussels, Belgium. Accepted for publication October 1, 1989.
2047
lyzing biopsy specimens. Furthermore, most of the currently useful morphologic parameters are obtained from examination of resected specimens and are thus not available before major surgical intervention. New image analyzing systems are easy to manipulate and are specifically designed for automatic, objective, and reproducible quantification of images of biopsy specimens and fine-needle aspiration samples.’ We illustrate to what extent such cell-image processors can be helpful for pathologists and clinicians. We analyzed normal, adenomatous, and malignant colorectal mucosae from 58 patients by means of a SAMBA 200 system (TITN, Grenoble, France) whose software provides quantitative morphologic information on nuclear size, DNA content, and chromatin We report the establishment of mathematical data banks for the detection of “abnormal or precancerous” cells in the so-called unaffected mucosa and adenomas and for the grading of colorectal carcinomas.
’
CANCER May I 1990
2048 Materials and Methods
Human Tissue Samples The colorectal mucosa studied were divided into six groups. The first one (Group 1 ) corresponds to normal colorectal tissue which was endoscopically or surgically removed from 12 patients who had undergone rectoscopy as a screening procedure for nonmalignant diseases such as hemorroids or diverticulosis. All other patients underwent partial colectomy for cancer. Group 2 corresponds to the so-called unaffected mucosa taken proximal to the surgical specimen; this mucosa appeared normal after gross and microscopic pathologic examinations of the 46 surgically resected malignant specimens. Group 3 corresponds to five adenomatous polyps found in surgically resected malignant tissues. Finally, Groups 4 to 6 correspond to the 46 cancers that were classified according to the Dukes' classification.*There were 10 Dukes' A (Group 4), 15 Dukes' B (Group 5), and 21 Dukes' C (Group 6) cancers.
Slide Preparation, Cell Nuclei Monitoring, and Parameters Analyzed Each colorectal sample was treated as previously described,' e.g., squashed onto a microscopic slide, fixed in ethanol-formal-acetic acid (EFA), and stained according to the Feulgen reaction, which allows selective and quantitative (stoichiometric) staining of DNA. Cell-image analyses were done with a SAMBA 200 microscopic image processor whose hardware and software were described by Brugal et al.' Fifteen morphonuclear parameters were computed per nucleus, and 250 to 300 nuclei were selected per case by the same pathologist according to the procedure described previo~sly.~ The morphonuclear parameters used belong to three categories: (1) one morphometric parameter, i.e., the nucleus area (SURF); (2) five densitometric parameters, i.e., integrated optical density (IOD) measuring the amount of absorbent material (DNA), mean optical density (MOD) measuring the concentration of absorbent material, and nuclear optical density variance (VOD) and the Kurtosis (K) and Skewness (SK) indices mathematically developed on the IOD parameter and globally describing chromatin distribution; and (3) nine textural parameters quantitatively describing chromatin patterns, i.e., to short-run (SRL) and long-run (LRL) length measuring the frequency of small and large dense chromatin clots, respectively, together with their relative distribution (RLD) and percentage (RLP); gray level distribution (GLD) measuring the uniformity of optical density distribution; the local mean (LM), the energy (E), and the coefficient variance (CV) of the co-occurrence matrix measuring the overall
Vol. 65
chromatin condensation level; and contrast (C) measuring the number of boundaries between nuclear regions of different extinction values. The proliferation index (PI) and the ploidy balance (PB) was calculated according to the definition of Opfermann et al." The PB index was calculated as the percentage of euploid cells (2c, 4c,8c) minus the percentage of aneuploid cells (3c, 5c, 6c, 7c, >8c). This difference can vary from 100%(all cells are euploid) to - 100%(all cells are aneuploid). The PI index represents the percentage of cells outside the major peak (regardless of its ploidy level) and the related peaks (peaks standing at 50% of the modal value and/or double the modal value)."
+
Statistical Analyses Results are given as mean +- standard error of the mean (SEM). We compared results with the Fisher F test (oneway variance analysis) and the conventional stepwise linear discriminant analysis (DISCRI software; TITN). Paired data were analyzed with nonparametric statistics, i.e., the Spearman and Kendall rank correlation techniques. Results
Correlation Between Histopathology and Morphonuclear Parameters Relative to Chromatin Pattern and Nuclear Size Figure 1 illustrates the relationship between histopathology (normal mucosa, unaffected mucosa, polyps, and
'0°
-Too
F
I
P P I
I1
I11
Iv
HIsrneAmoux1ct.L
V
VI
CROUPS
FIG. 1. Relationship between histopathology and morphonuclear parameters related to hyperchromatism and assessed by means of the SAMBA 200 cell image processor using 250 to 300 nuclei in Feulgenstained imprint smears from freshly resected colorectal tissues. Groups I and 11, respectively, represent normal mucosa from patients without (I) or with (11) colorectal cancer; Group 111 relates to adenomatous polyps and Groups IV to VI correspond to the cancers classified according to the three stages of the Dukes' classification (IV = A, V = B, VI = C). 0 = short-run length emphasis (SRL); 0 = long-run length emphasis (LRL); A = local mean (LM). The parameter mean value (+SEM) recorded in Groups I1 to VI is compared (one-way ANOVA) with that of Group I. *P < 0.05. **P < 0.01. ***P< 0.001.
CELLIMAGEANALYSISOF COLORECTAL CANCERS *
No. 9
40000
1 -
I
-
- 0 I
I1
111
v
IV
Vl
HISTOPATWOWCICN. GXOUPS
FIG. 2. Relationship between histopathology and three morphonuclear parameters, i.e., nuclear area (SURF = O ) , contrast (C = 0),and gray level distribution (GLD = A). I = normal mucosa from patients without colorectal cancer; I1 = normal mucosa from patients with colorectal cancer: I11 = adenomatous polyps; IV = Dukes’ A cancer: V = Dukes’ B cancer; VI = Dukes’ C cancer. The computerized parameters were assessed by means of the SAMBA 200 system using 250 to 300 nuclei in Feulgen-stained imprint smears obtained from freshly resected colorectal tissues. The parameter mean value (+SEM) recorded in Groups I1 to VI is compared (one-way ANOVA) with that ofGroup I. *P< 0.05. **P < 0.0 1. ***P < 0.00 1.
cancers belonging to the Dukes’ A to C classes) and three morphonuclear parameters related to SRL and LRL and to LM of the gray level. We observed a significant positive correlation ( P < 0.00 1 by Spearman-Kendall tests) for these three indices versus the histopathology defined as normal (Groups 1 and 2), adenomatous (Group 3), and malignant (Groups 4 to 6). Indeed SRL (by assessing the frequency of small dense chromatin clots), LRL (by assessing the frequency of large dense chromatin clots), and LM (by assessing the overall chromatin condensation
FIGS. 3A-3F. DNA histograms assessed in three separate cases of colorectal cancers (B, D, and F), and for each one there is also the DNA histogram of the corresponding unaffected mucosa as a control of the method. A represents the unaffected mucosa of case B, C that of case D, and E that of case F. We used the integrated optical density (IOD) value for assessingthe DNA content on each Feulgen-stained nucleus by means of the SAMBA 200 cell image processor. These IOD assessments make it possible to calculate both proliferation index (PI) and ploidy balance (PB), the exact values of which are given in Table 1.
Verhest et al.
2049
level) significantly increased from normal to malignant colorectal tissues. In contrast, Groups 1 and 2 mucosa appeared similar with regard to these three parameters. The same observation was made regarding the three classes of cancers (Groups 4 to 6). Figure 2 corroborates these results. Measuring as they do the uniformity of optical density distribution, SURF and GLD increased dramatically from Groups 1 and 2 mucosa to Groups 4 to 6 cancerous mucosa; polyp values were intermediate. The GLD index was significantly lower in Group 1 compared with Group 2. Measuring as it does the number of boundaries between nuclear regions of different extinction values, C significantly decreased from the normal or unaffected mucosa to the cancerous mucosa.
Correlation Between Histopathology and Morphonuclear Parameters Relative to DNA Content Figure 3 illustrates DNA histograms assessed in three different cases of colorectal cancers (B, D, and F). For each we also show the DNA histogram of the corresponding unaffected mucosa as a control of the method. Thus, A represents the unaffected mucosa of case B, C that of case D, and E that of case F. We used the IOD value assessed on 50 lymphocytes of each case as an internal diploid standard; the value of 2000 arbitrary IOD units corresponds to a 2c quantity of DNA. Table 1 reports the values PI, PB, and the percentage of polyploid nuclei (P), it..,the nuclei having more than 4c DNA content in each of these six samples. Our results indicate that case B corresponded to a diploid cancer (with a high PB value) and a relatively high proportion of proliferating cells (PI index)
,o 20 1o
IS 10
IO
,O
!o I’
:,
0
2000
1000
6000
BOO0
I0000
0
2000
1000
(OW
8000
CANCERMay I 1990
2050
Vol. 65
TABLE1. Proliferation Index (PI), Ploidy Balance (PB), and Percentage of Polyploid Cells (P) in the 3 Cancers and Their Respective Unaffected Mucosa Case
PI
PB
P
A B C D E F
3.8% 10.7% 1.3% 4.1% 0.9% 10.3%
+89.4% +54.3% +94.4% +58.8% +94.0% -40.6%
0% 3.7% 1.2% 10.1% 0.6% 11.2%
with only a few polyploid cells. Case D also corresponded to a diploid tumor with a spectacular blockade of cells in the G2 phase, but Case F was an aneuploid cancer. Figure 3 shows that two of the three so-called unaffected mucosas contained a significant proportion of polyploid cells, a feature that we observed in 31 of the 46 cases studied here. As shown in Figure 4, a significant correlation (P < 0.00 1 by Spearman-Kendall tests) exists between the DNA content, assessed by both PI and PB, and the histopathology of the tissue. The PB of Groups 3 to 6 were dramatically lower compared with Groups 1 or 2. This mean PB value was around +80% in Groups 1 and 2 and between +20% and -10% in Groups 3 to 6. Polyps and the cancers had a relatively equal proportion of euploid (2c, 4c, and 8c quantity of DNA) and aneuploid (2.5c, 3c, 3.5c, 5c, 6c, 7c, and >8c quantity of DNA) cells, but the normal and unaffected mucosa had mostly euploid cells. The PI were much higher in polyps and cancers
4
-15 -120 -100
-80
-60
-40
-20
0
+20
4 0 +60
4 0
FIG. 5. Example of the setting up of preliminary data banks corresponding to the six colorectal histopathologies studied, i.e.,normal mucosa from patients without (I) or with (11) colorectal cancers, adenomatous polyps (111), and cancers according to the three stages of the Dukes’ classification (IV = A, V = B, VI = C ) .In each 300 nuclei of each group, described by the 15 computerized morphonuclear parameters and therefore located in a 15dimensional space, underwent principal components analysis and were projected onto the two-dimensional factorial plane in accordance with the canonic transformation of data. The coordinates represent complex multifactorial functions featuring the nuclei. The center ( 0 )of each cluster represents the 95% confidence interval on the mean position in the factorial plane, and the large ellipsis represents the 95% confidence interval of the factorial cell distribution and shows a weak and nonsignificant overlap between normal (I + II), adenomatous (111), and malignant (IV + V + VI) nuclei. A discriminant analysis showed that this weak overlap corresponded to normal nuclei still present in the 111 to VI data banks.
than in normal or unaffected mucosa. With regard to the cancers, no significant differences appeared between the three stages of the Dukes’ classification versus the PI or PB. 2 75
5
Selection of Reference Normal, Adenomatous, and
; Malignant Cells by Mathematical Analyses 50
5
-w
25
e I
I1
1x1
IV
V
VI
HISTOPAnlOLOGlCAl. CROUPS
Frc. 4. Relationship between histopathology and cell cycle kinetic parameters, i.e.,proliferation index ( 0 )and ploidy balance (0).I = normal mucosa from patients without colorectal cancer; I1 = normal mucosa from patients with colorectal cancer: 111 = adenomatous polyps; IV = Dukes’ A cancer: V = Dukes’ B cancer; VI = Dukes’ C cancer. The computerized parameters were assessed by means of the SAMBA 200 system on 250 to 300 nuclei in Feulgen-stained imprint smears obtained from freshly resected colorectal tissues. The parameter mean value (+SEM) recorded in Groups I1 to VI is compared (one-way ANOVA) with that of Group I *P i0.05. **P i 0.01. ***P < 0.001.
Figure 5 illustrates the six preliminary data banks in accordance with the procedure described in detail by Larsimont et aI.,9 corresponding to normal and unaffected mucosa, polyps, and Dukes’ A, B, and C cancers. This figure illustrates the 95% confidence interval delineated by the ellipsis around the cell population “mean profile” located in the 15-dimensional space and related to the 15 morphonuclear parameters. It is bidimensionally projected onto the factorial plane using a canonic treatment of the data. Our mathematical analysis indicates that normal and unaffected mucosa are similar, as are the three types of cancer according to the Dukes’ classification. Indeed, multiparametric analysis located the nuclei from the normal and unaffected mucosa in the same part of the fac-
No. 9
CELLIMAGEANALYSIS OF COLORECTAL CANCERS
We described a computer-based expert system for diagnosing colorectal neoplasms. As stated by Weber et al., expert systems represent computerized ways to apply previously accumulated data or information to the analysis
''
Figure 7 represents six specimens obtained from normal (A), unaffected (B), polyp (C), Dukes' A (D), Dukes' B (E), and Dukes' C cancer (F) smears. These six cases were
i
+9+ 3 f".
.
++&-:. _.. /. .
-3
205 1
Discussion
Example ofthe Usefulness of Objective Data Banks for the Grading ofColorecta1 Tissues
FIGS.6A-6D. The setting up of three preliminary malignant data banks corresponding to three types of colorectal cancers in accordance with cut-off values related to proliferation index and ploidy balance. The first data bank represents the cancers with PI < 5% and PB > 50% (chart A), the second one the cancers with 5% < PI < 10%and 0% < PB < 50%(chart B) and, lastly, the third one the cancers with PI t 10%and PB < 0%(chart C). These three types of cancers might represent three stages of aggression. Chart D represents the discrimination obtained among these three types of cancers after discriminant analysis and the canonic transformation of data depending on the method described in the legend to Figure 5 (m = first data bank 0 = second data bank, A = third data bank).
Verhest et al.
analyzed by means of the six preliminary data banks shown in Figure 5. For this purpose, we tested 300 cell nuclei selected at random in each of the six cases against 300 cell nuclei selected at random in each of the six data banks. This was done by means of the DISCRI software of the SAMBA 200 system which revealed that the normal and unaffected data banks did not contain typical cancer cell nuclei, but the three cancer data banks contained 1% to 5% of typical normal cell nuclei (data not shown), as is suggested by the slight overlap in the ellipses in Figure 5. The results show that the first case, representing a normal mucosa, has an 86% fit with the normal data bank and an 8% one with the unaffected bank; the other 6% fits in with the other four data banks. The second case, representing an unaffected mucosa, shows a 94% fit with the first two data banks and a 6% one with the other four. The three cancers had 93%, 83%, and 51% fits, respectively, with the three cancer data banks. Finally the polyp had a 78% fit with the two normal data banks, a 2 1% one with the three cancer banks, and a 1% one with its own bank.
torial plane, those from the polyps in another, and those from the three cancer types in a third. In other words, multiparametric analysis based on the 15 morphonuclear parameters was unable to discriminate between the normal and the unaffected mucosa or between the three cancer types. In contrast, with regard to colorectal cancers, Figure 6 indicates that it is possible to create three distinct data banks using PI and PB. Chart A (Fig. 6) represents the data bank for the cancers with a PI index lower than 5% and a PB value higher than 50%, independent of their Dukes' stage. Chart B (Fig. 6) represents the bank characterized by 5% < PI < 10%and 0% < PB < 50%, while chart C (Fig. 6) represents the bank characterized by PI > 10% and PB < 0%. In other words, data banks A and C (chart D of Fig. 6) might represent less and more aggressive colorectal cancers, respectively, with the B bank representing the intermediate state. Case D in Figure 3 represents a cancer fitting in with the data bank A in Figure 6, while Case B in Figure 3 fits in with data bank B, and Case C in Figure 3 matches data bank C in Figure 6 (chart D).
+IS
-
*
\
-9 -
,
,
.
,
.
.
+2
-4
I
+8
.
.
+I4
.
.
.
_
+20
+26
+I0
+ +I4
'"1
. ,
.
...
-6
-2
+2
+6
-6
-2
+?
-
+A
+I0
-
+I4
CANCERMay 1 1990
2052
7jF””f ........ ........ sol;
....... ....... ....... ........
....... ......................... I.
................. ................. ......................... I
11
IT1
I!’
HI STOPAIl10L0(: 1 CAL
V MIA
0
V l
BANKS
FIGS. 7A-7F. Illustration of the usefulness of the preliminary data banks to score “unknown” colorectalJpathologies objectively. (A-F) 6 cases that were diagnosed as normal mucosa from patients without (A) and with (B) colorectal cancer, adenomatous polyp (C), Dukes’ A cancer (D), Dukes’ B cancer (E), and Dukes’ C cancer (F). On the basis of the DlSCRl software of the SAMBA 200 system, it appeared that when “tested” against the six data banks (abscissa), case A had most cells displaying morphonuclear features belonging to the first data posbank (normal mucosa of patient without colorectal cancer). Case B had an equal proportion of cell nuclei reclassified into the first two data banks. Case C, i.e., the polyp, contained a large pattern of nuclei types reclassified into the six data banks; this was also the case for the three cancers. Prospective evaluations are now under investigation to determine the “alarm” cut-off value for diagnosing a premalignant or malignant tissue in a colorectal biopsed or resected sample.
of subsequent situations. Banner et al.’ recently wrote that there was a need for prognostic and therapeutic indicators in colon carcinoma that are accurate, objective, reproducible, available before surgery, and capable of acclimating themselves to continually changing malignant systems. These authors report on the basis of both clinical and experimental data that colorectal tumors evolve over time, with a waxing and waning of their cell populations. I Current prognostic indicators obtained at only one point in time may be insufficient for the ongoing management
Vol. 65
of cancer patients. I Furthermore, conventional prognostic parameters for colorectal cancers are usually determined after resection and thus have limited predictive value. We used the SAMBA 200 cell image processor7 to describe in a quantitative, objective, and reproducible fashion the morphonuclear and cell kinetic characteristics of normal, adenomatous, and malignant colorectal mucosa. We then set up preliminary data banks to educate our expert system to grade objectively the aggressiveness of unknown colorectal adenomatous, premalignant, and malignant lesions. We use the term “preliminary” data banks because proper clinically available data banks must integrate at least several hundred cases, an objective that we are now pursuing. We used traditional parameters, i.e., PI and PB, to validate this method versus other previous ones like tritiated thymidine (3H-dThd), autoradiography, and flow cytometer analysis (FCA). Our results clearly indicate that the PI dramatically increased from normal to malignant mucosa concomitant with a significant decrease in PB values, which corresponded to an increase in the proportion of aneuploid cells. These findings are corroborated by others who show that FCA parameters correlate with disease-free intervals and with survival rate.’,’2Such studies show that aneuploidy and poor survival were correlated in c ~ l o r e c t a l ’or ~ ’gastric15 ~~ cancers, although correlations with individual conventional prognostic factors were inconsistent. I 6 With regard to data obtained with 3H-dThd autoradiography, Bleiberg and coworker^'^ show that the S-phase duration in colorectal tumors was significantly longer than in unaffected mucosa and polyps, on the one hand, and significantly longer in the unaffected mucosa of cancer patients than in the mucosa of patients without gastrointestinal pathology,” on the other. These authors hypothesize that an early malignant step involves the whole colorectal mucosa but results in a malignant growth in only one privileged locus. l 8 When doing multiparametric analyses we obtained only one significant difference between the unaffected mucosa of cancer patients and those of patients without gastrointestinal growth. This difference was revealed by the GLD which measures the uniformity of optical density distribution. However, this morphonuclear difference has not yet led to any significant discrimination between these two types of mucosa (Fig. 5). Unaffected mucosa from patients with colorectal cancers was first described by Filipe,” and Mughal and Filipe2’ proposed a “crescendo gradient of change” from normal mucosa through transitional mucosa to adenoma and carcinoma. For Hamilton and transitional mucosa is not a primary premalignant change but a secondary reactive phenomenon that does not show the same morphometric characteristics of regenerative colonic epithelium as those de-
No. 9
CELLIMAGEANALYSISOF COLORECTAL CANCERS *
Verhest et af.
2053
pathologists to score objectively the degree of malignancy scribed by Allen et ~ 1Hamilton . ~ ~et ~ 1 . argue ~ ' that the of colorectal disease by using a cell-image analysis of measurement of nuclear DNA content in transitional Feulgen-stained nuclei. The monitoring of the morphomucosa might help to indicate hidden nuclear abnornuclear parameters computed on 250 to 300 nuclei/case malities which are not detected using morphometry. With regard to DNA content measurement, many a ~ t h o r s * ~ - ~ 'makes it possible to describe the evolution of colorectal malignancy objectively and quantitatively. This need for report, as do we, that this index may be a good indicator only a very small number of nuclei is promising in the of malignant and premalignant change and could be one creation of a preoperative investigative tool using fineof the first manifestations of premalignancy. needle aspiration cytology of any site along with echogRecent advances in cell-image analysis techniques reraphy or computed tomographic imaging support. We sulted in new approaches not only related to cell cycle also set up preliminary data banks that enable an objective kinetics (avoiding problems often encountered with 3Hand reproducible grading of unknown cases to be done. dThd a~toradiography),~~-~' but also to chromatin pattern Our approach must be validated prospectively on a large determinati~n.'*'~,~~-~~ We observed that nuclei from canseries of cases to create an expert system of colorectal cers were more hyperchromatic than those arising from malignancy diagnosis, a study that we are now undernormal mucosa, with polyp nuclei showing an intermetaking. diate pattern. It appeared that it was the condensation of the chromatin, that became more textured, which distinREFERENCES guished nonmalignant from malignant mucosa. We observed a concomitant increase in SRL and LRL, which 1. Banner BF, Tomas-De La Vega JE, Roseman DL, Coon JS. Should flow cytometric DNA analysis precede definitive surgery for colon carassess the frequency of small and large dense chromatin cinoma? Ann Surg 1985; 202:740-744. clots, respectively. In addition the mean value of C, which 2. Dukes CE. The classification of cancer of the rectum. J Pathol measures the number of boundaries between the nuclear Bacteriol 1932; 35:323-332. 3. Rosai J, ed. Ackerman's Surgical Pathology, ed. 6. St. Louis, MO: regions of different extinction values and increases with CV Mosby, 1981; 536-541. the number of dense and well-separated chromatin clots, 4. Roseman DL, Straus AK. Staging of carcinoma of the colon and was lower in malignant nuclei than in nonmalignant. In rectum. Surg Gynecol Obstet 1980; 151:93-95. 5. Blenkinsopp WK, Stewart-Brown S, Blesovsky L et ul. Histopaother words, the number of small and large dense chrothology reporting in large bowel disease. J Clin Puthol 1981; 34509matin clots increased so powerfully from nonmalignant 513. to malignant stage that they join together and finally pro6. Schwartz WB, Wolfe HJ, Pauker SG. Pathology and probabilities: A new approach to interpreting and reporting biopsies. N Engl J Med duce a decrease in the number of well-separated clots and, 198 1; 3051917-923. consequently, a decrease in C . 7. Brugal G, Garbay C, Giroud F, Adhel D. A double scanning miOur treatment of the data, by canonic projections onto crophotometer for image analysis: Hardware, software and biomedical applications. J Histochem Cytochem 1979; 27: 144- 152. the factorial plane followed by isolation of empiric clusters 8. Kiss R, de Launoit Y , Gras S, Thibaut S, Atassi G. A new assay on which discriminant analyses were done, allowed the to evaluate cell growth and drug sensitivity in culture using a cell image selection of homogeneous cell populations belonging to processor. Anticancer Res 1988; 8:765-774. 9. Larsimont D, Kiss R, d'Olne D et al. Correlation between nuclear each of the pathologic groups we described. This consecytomorphometric parameters and estrogen receptor levels in breast quently avoided the great variability occurring in indicancer. Cancer 1989; 63:2162-2168. vidual cases. Although we studied a small number of cases, 10. Opfermann M, Brugal G, Vassilakos P. Cytometry of breast carcinoma: Significanceof ploidy balance and proliferation index. Cytometry we found that the morphonuclear data processing showed 1987; 81217-224. a continuous evolution of nuclear features from normal 11. Weber JE, Barteels PH, Griswold W, Kuhn W, Paplanus SH, to malignant stages. Furthermore, we observed no differGraham AR. Colonic lesion: Expert system: Performance evaluation. Anal Quunt Cytol Histol 1988; 10:150-159. ences between cancers according to their Dukes' stages. 12. Kokal W, Sheibani K, T e n J, Harada JR. Tumor DNA content In contrast, we propose a new classification to set up three in the prognosis of colorectal carcinoma. JAMA 1986; 255:3 123-3 127. groups of cancers in accordance with cut-off values related 13. Wolly RC, Shreiber K, Koss LG et al. DNA distribution in human colon carcinomas and its relationship to clinical behavior. J Nut1 Cuncer to both the PI and PB. We must now test such a cancer Inst 1982; 69:15-22. aggressiveness scheme. Our results further indicate that 14. Forsslund G, Cedermark B, Onman U et al. The significance of it is easy to score quantitatively the pathology of an unDNA distribution pattern in rectal carcinoma: A preliminary study. Dis known colorectal case by this method, i.e., linear discrimColon Rectum 1984; 21:579-584. 15. Korenaga D, Okamura T, Saito A, Baba H, Sugimachi K. DNA inant analysis between the unknown case and the six exploidy is closely linked to tumor invasion, lymph node metastasis, and perimental data banks. The fact that the unknown polyp prognosis in clinical gastric cancer. Cuncer 1988; 62:309-3 13. 16. Temple WJ, Sugarbaker EV, Thornthwaite JT et ul. Correlation smear is not recognized as a polyp by the system should of cell cycle analysis with Dukes staging in colon cancer patients. JSurg be explained by the weak number of polyps (five cases) Res 1980; 28:314-318. used for the creation of the polyp preliminary data bank. 17. Bleiberg H, Salhadin A, Galand P. Cell cycle parameters in human colon: Comparison between primary and recurrent adenocarcinomas, In conclusion, our study points to the possibility for
2054
CANCERMay 1 1990
benign polyps and adjacent unaffected mucosa. Cancer 1977; 39:11901194. 18. Bleiberg H, Buyse M, Galand P. Cell kinetic indicators of premalignant stages of colorectal cancer. Cancer 1985; 56: 124-129. 19. Filipe MI. Value of histochemical reactions for mucosubstances in the diagnosis of certain pathological conditions of the colon and rectum. Gut 1969; 10:577-586. 20. Mughal S, Filipe MI. Ultrastructural study of normal mucosaadenoma-cancer sequence in the development of familial polyposis coli. J Nut1 Cancer Inst 1978; 60753-768. 21. Hamilton PW, Watt PCH, Allen DC.A morphometric assessment of transitional mucow in the colon. Histopathology 1988; 1 3 5 19-530. 22. Allen DC,Hamilton PW, Watt PCH, Biggart JD. Morphometrical analysis in ulcerative colitis with dysplasia and carcinoma. Histopathology 1987; 11:913-926. 23. Barlogie B, Raberm N, Schumann J et al. Flow cytometry in clinical cancer research. Cancer Res 1983; 43:3982-3997. 24. Banner BF, Chacho MS, Roseman DL, Coon JS. Multiparameter flow cytometric analysis of colon polyps. Am J Clin Pathol 1987; 87: 3 13-318. 25. Melville DM, Jass JR, Shepherd NA. DNA aneuploidy as a marker of neoplastic change in ulcerative colitis. J Pathol 1987; 152:215. 26. Cuvelier CA, Morson BC, Roels HJ. The DNA content in cancer and dysplasia in chronic ulcerative colitis. Histopathology 1987; 1 1:927939.
Vol. 65
27. Lofberg R, Tribukait B, Ost A, Bromstrom 0, Reichard H. Flow cytometric DNA analysis in longstanding ulcerative colitis: A method of prediction of dysplasia and carcinoma development? Gut 1987; 28: 1100-1 106. 28. Van den Ingh HF, Griffioen G, Cornelisse CJ. Flow cytometric detection of aneuploidy in colorectal adenomas. Cancer Res 1985; 45: 3392-3397. 29. Maurer HR. Potential pitfalls of 3H-thymidine techniques to measure cell proliferation. Cell Tissue Kinet 1981; 14:1 1 1-120. 30. Jozan S, Gay G, Marques B, Mirauze A, David JF. Oestradiol is effective in stimulating 'H-thymidine incorporation but not proliferation of breast cancer cultured cells. Cell Tissue Kinet 1985; 18:457-464. 3 1. Allison DC, Ridolpho PF, Anderson S, Bose K. Variations in the 'H-thymidine labeling of S-phase cells in solid mouse tumors. Cancer Res 1985; 45:6010-6016. 32. Atkin NB, Kay R. Prognostic significance of modal DNA value and other factors in malignant tumors based on 1465 cases. Br J Cancer 1979; 40:210-221. 33. Koga Y, Sugimachi K, Kuwano H, Mori M, Matsufuji H. Cytophotometric DNA analysis of oesophageal dysplasia and carcinoma induced in rats by N-methyl-N-amylnitrosamine. Eur J Cancer Clin Oncol 1988; 24:643-65 1.
