Computerized analysis of cellular features and biomarkers for cytologic diagnosis of early lung cancer

OBJECTIVE: To present a set of novel computerized analysis algorithms to construct a computer-aided cytologic diagnosis (CACD) system to differentiate lung cancer biomarkers and identify cancer cells in the tissue-based specimen images. STUDY DESIGN: Molecular methods, including application of cancer-specific markers, may prove to be complementary to cytology diagnosis, especially when they are combined with CACD system for biomarker assessment. We trained a novel CACD system to recognize expression of the cancer biomarkers histone H2AX in lung cancer cells and then tested the accuracy of this system to distinguish resected lung cancer from preneoplastic and normal tissues. The major characteristics of CACD algorithms is to adapt detection parameters according to cellular image contents. Our newly developed wavelet transform is able to adaptively select different resolution and orientation features based on image content requirements. RESULTS: Visual, statistical and quantitative results as CACD performance evaluation are presented in this paper. CONCLUSION: The presented algorithms and CACD system for cellular feature enhancement, segmentation and classification are very important in distinguishing benign and malignant lesions.     

Computerized image analysis of Ki-67 in ductal breast carcinoma in situ.

OBJECTIVE: To develop and determine the staining protocols and computerized image analysis methods that are the most effective combination for performing quantitative analysis of Ki-67. STUDY DESIGN: We compared conventional bright-field light microscopy and refractive optical enhancement methods in combination with various immunodetection and filter enhancement methods, including immunogold in combination with epipolarization refractive optics and enzymatic conversion of chromogenic substrates in combination with optical filter enhancement. Initial Ki-67 tests were performed on lymph node tissues and cultured human breast cells and then applied to 200 ductal carcinoma in situ (DCIS) samples. DCIS acini were digitally acquired, and a region of interest was manually outlined in each one with a digital stylus to include only the cellular component; then the Ki-67 staining index was quantified by segmentation analysis. RESULTS: Although combining epipolarization analysis with immunohistogold staining was the most sensitive detection method, nonspecific binding was too high. The streptavidin-horseradish-peroxidase enzymatic conversion of 3,3'-diaminobenzidine (DAB) in combination with optical enhancement filters was the most effective method tested. Ki-67 stain was associated with dense fibrillar structures of the nucleoli in the less intensely staining nuclei and was most intense in paired nuclei. CONCLUSION: The method of measuring Ki-67 expression by DAB staining combined with optical enhancement filters and quantification via computer-assisted image analysis techniques produced objective and reproducible results. As such, this method can offer (1) less intraobserver and interobserver variability, (2) a digital archival record, and (3) a baseline for digital exchange of information between studies.     

Value of morphometric nuclear image analysis using the Feulgen reaction in renal cell carcinoma

OBJECTIVE: To evaluate retrospectively the ability of morphometric nuclear image analysis to predict survival in patients with renal cell carcinoma. STUDY DESIGN: The subjects were 40 patients with previously untreated renal cell carcinoma. Pathologic stage was determined using Robson's stage system. Nuclear grade was assigned according to the criteria of Fuhrman et al. We used the Feulgen staining technique, which has been widely used for the histochemical assessment of nuclear DNA content. A minimum of 300 nuclei were analyzed from each subject. Five variables in morphometric nuclear image analysis were measured: nuclear area, nuclear perimeter, nuclear ellipticity, nuclear regularity and DNA content. Cox's proportional hazard model was applied to identify prognostic usefulness with respect to survival time. RESULTS: All nuclear morphometric variables but nuclear regularity correlated with tumor grade. According to univariate survival analyses, Robson stage and nuclear ellipticity revealed a prognosis on survival with statistical significance. After adjustments for age and sex, nuclear ellipticity remained the only significant prognostic factor related to survival (P < .01). The survival rates were relatively high for patients with nuclear ellipticity > 773 as compared to those with nuclear ellipticity < 773 (P < .05). CONCLUSION: These findings indicate that morphometric nuclear image analysis using the Feulgen reaction is a reliable and efficient technique and that nuclear ellipticity is the most discriminating morphometric variable for predicting the prognosis of renal cell carcinoma patients.     

Nuclear image morphometry and cytologic grade of breast carcinoma

OBJECTIVE: To correlate visual cytologic grade with automated nuclear morphometry of carcinoma of the breast. STUDY DESIGN: We randomly selected 24 histologically proven infiltrating ductal carcinomas of the breast and 10 benign breast lesions (fibroadenoma). Hematoxylin-eosin-stained fine needle aspiration cytology (FNAC) smears were selected for both cytologic grade and automated image morphometry. The same hematoxylin-eosin-stained FNAC smears were studied for area, convex area, standard deviation of nuclear area, diameter, perimeter and convex perimeters of nucleus. At least 100 cells from each case were measured with an image cytometer. RESULTS: Mean nuclear area, standard deviation of nuclear area, nuclear diameter, convex area, convex perimeter and perimeter were significantly increased from benign versus grade 1 carcinomas and grade 1 versus grade 2 and 3 carcinomas (one way ANOVA test). However, there was no significant difference in grade 2 versus grade 3 carcinomas. CONCLUSION: Automated image cytometry rapidly and successfully measures various nuclear parameters. The measurement of various nuclear parameters would be helpful in future applications of automated diagnosis and grading of breast carcinomas from cytologic material.     

Computerized morphometric analysis of the chick embryo duodenum organogenesis

A computerized morphometric analysis was performed on histological sections of the proximal and distal limbs of the chick embryo duodenum, from the 8th to the 15th day of incubation, with the aim of evaluating the quantitative aspects of organogenesis. A semiautomatic digital system was used. The areas of total section, of the lumen, of the wall and of its components (subserous stratum, muscle layer, lamina propria, epithelium) and the thickness of the epithelium and muscle layer were measured; the mean +/- S.E.M. of the values obtained was calculated. The percentage of shrinkage, due to the histological procedures, was calculated for each day. The values obtained were added to the measurements of each area and thickness thus giving the true value of each parameter measured. The differences between the mean values of the areas of the proximal and the distal tract were statistically evaluated. Exponential curves and r coefficient were determined to evaluate the general growing pattern of the mean area of the components of the duodenal wall, as a function of age. The main results are the following: the areas of the wall components of the proximal limb are always greater than the corresponding areas of the distal limb of the duodenum. There are some differences between the proximal and the distal limbs of the duodenum in the growing pattern of the intestinal wall components. The anlage of the duodenum also has a different developmental pattern compared with that of the chick embryo ileum.     

Prognostic evaluation of breast cancer in cytologic specimens

OBJECTIVE: To evaluate prognostic factors in breast cancer using cytologic samples and to determine the correlation between those factors and ploidy. STUDY DESIGN: Two hundred sixteen fine needle aspirates from patients with primary breast cancer were analyzed for expression of estrogen receptors (ERs), progesterone receptors (PRs), Ki-67 antigen, expression of p53 tumor suppressor gene and overexpression of c-erbB-2 using a standard immunochemical method. Not all subjects had all biomarker information because of the study design (c-erbB2 added later). The specimens were analyzed also for ploidy. We used the SAMBA 4000 image analysis system for quantification of the percent of cells stained positively by the different immunocytochemical stains andfor ploidy. RESULTS: A significant correlation wasfound between ER and PR and between Ki-67 and positive p53. Steroid receptor content was not significantly related to p53, Ki-67 or c-erbB2. No correlation was found between c-erbB2 and the other biomarkers. Ploidy had a significant correlation with all the biomarkers used. CONCLUSION: A reliable and rapid evaluation of markers for breast cancer can be achieved by measuring cells stained positively by immunocytochemical stains, as well as ploidy, by means of an image analysis system. ER, PR Ki-67, p53 and c-erbB2 had a significant correlation with ploidy and overall prognostic value in breast cancer.     

Computerized analysis of chromosomal parameters in karyotype studies

Summary A study is presented of the possibilities and limitations of semi-automated karyotype analysis on the basis of chromosome length and centromere index. A number of computer programs have been developed for 1) quick and precise measurements of chromosome arm length with the help of a graphics tablet, 2) computing (relative) length and centromere index and statistical analyses of the data, and 3) representation of these chromosomal parameters in two-dimensional scattergrams. An ellipse representing 95% of the probability mass is drawn around the bivariate mean of each chromosome. The size and orientation of the axes are calculated from repeated measurements of the chromosomes of one metaphase plate. If there is a correlation between length and centromere index, which is often the case, the axes of the ellipse are tilted. Incorporation of such a covariance analysis proved to be of great importance for an accurate karyotype analysis. The Computer Aided Karyotyping package does not contain routines for an automated classification of the chromosomes. The main reason is that the variation in length and centromere index of a given chromosome in different cells is often much larger than the variation between nonhomologous chromosomes. In addition, it was our aim to develop universal karyotyping aids which can be used regardless of the species studied.     

Quantitative immunoprofiles of breast cancer performed by image analysis

OBJECTIVE: To determine the biopathologic profiles of breast cancer for greater knowledge of tumor natural history and clinical outcome. STUDY DESIGN: In 99 in situ (ISC) and 2718 infiltrating breast carcinomas (IC), biologic markers (estrogen receptor [ER], progesterone receptor [PR] proliferation index, cerbB-2/NEU, p53, bcl-2 and DNA ploidy) were evaluated with an image analysis system (CAS 200/486). In 105 mixed invasive cancers with size < or = 1 cm, a separate analysis of in situ (ISCm) and invasive component (ICm) was obtained. A clinical study of 836 invasive breast cancers was performed. RESULTS: Different biophenotypes were obtained: among ISCs, cribriform type exhibited biologic behavior similar to that of normal breast tissue (ER+, PR+, proliferation index [PI] low, NEU-, p53-, bcl-2+) the opposite profile was displayed by comedo type, and intermediate phenotypes were observed in noncomedo and lobular types. Comparing ISC and ISCm, PI and p53 expression had the highest levels in ISCm with respect to other groups. NEU overexpression exhibited a decreasing value from ICm to IC. Younger women (< or = 40 years) with IC demonstrated a worse biologic profile (high PI, p53+, ER- and size > 2 cm). In multivariate analysis, PI and NEU in node-negative patients, and NEU, PR and size in node-positive ones emerged as prognostic parameters. CONCLUSION: The results underline the importance of the quantitative biologic profile for defining tumor behavior and patient management.     

Prognostic markers in histologic and cytologic specimens of breast cancer

OBJECTIVE: To correlate histologic and cytologic specimens of breast cancer by the expression of prognostic factors, such as estrogen receptor (ER), progesterone receptor (PR) and c-erbB-2, with immunochemical staining. STUDY DESIGN: Cytologic and histologic specimens from 83 patients were analyzed for expression of ER and PR, and 30 cases were analyzed for overexpression of c-erbB-2 using a standard immunochemical method. The material used for immunocytochemical staining was taken from the needle and syringe after each aspiration and smear preparation. The material was washed into a small container with preservative solution. Immunohistochemical staining was performed on paraffin-embedded specimens. RESULTS: A significant association was found between the histologic specimens and cytologic specimens by means of the expression of immunochemical markers. The best correlation between cytologic and histologic specimens was found when using c-erbB-2. CONCLUSION: A reliable and rapid evaluation of markers for breast cancer can be achieved by immunocytochemical staining on cytologic material. A good association was found between histologic and cytologic specimens using immunostaining.     

Intraoperative cytologic diagnosis of sentinel node metastases in breast cancer

OBJECTIVE: To clarify the usefulness of imprint cytology for intraoperative investigations of sentinel lymph nodes in breast cancer, comparing the results with those of examinations using frozen and permanent sections. STUDY DESIGN: The material consisted of 303 sentinel lymph nodes from 124 cases of clinically node negative breast cancer. Touch imprint cytologic slides and frozen sections were obtained from the same cut surface of the sentinel nodes. Correlations with the final histopathologic results in paraffin sections were evaluated. RESULTS: The sensitivity, specificity and accuracy of imprint cytology were 70.3%, 99.6% and 96.0%, and those of frozen sections were 83.8%, 100%, 98.0%, respectively. The values were improved when the 2 methods were combined (89.2%, 99.6%, 98.3%), though the concordance between imprint cytology and frozen section was 91.9%. CONCLUSION: Both imprint cytology and frozen section are useful for evaluating sentinel lymph node status in breast cancer. However, the 2 techniques should be combined to improve the diagnostic sensitivity.     

Computerized morphometric analysis of pathological prion protein deposition in scrapie-infected hamster brain.

Transmissible spongiform encephalopathies (TSEs or prion diseases) are characterized by a constellation of typical though variable pathological changes in the brain. Deposition of disease-associated abnormal prion protein (PrP(Sc)) is the pathological feature of TSEs most consistent and accessible for quantification. However, the evaluation of PrP(Sc) deposits detected by immunohistochemical techniques has been traditionally based on arbitrarily assigned semiquantitative scores. This approach is limited by its subjectivity and bias, yielding considerable variability. In this study, we used MetaMorph 6.1 image analysis software for quantitative analysis of immunostained PrP(Sc) deposits in the CNS of hamsters infected with the 263K strain of scrapie agent. Computerized morphometric analysis (CMA) allowed unambiguous detection of even minimal amounts of immunostained PrP(Sc). CMA values for intensity of staining and area stained correlated well with semiquantitative scores, providing reproducible quantitative data and objective criteria for analyzing PrP(Sc) deposition. CMA provides a simple and reliable method for improved and consistent diagnosis of TSEs that may also be used to quantify other immunostained biomarkers.     

Accuracy of blind cytologic diagnosis of diploid breast cancer

Fifteen fibroadenomas and 20 adenocarcinomas with identical diploid DNA histogram patterns were blindly diagnosed in fine needle aspirates by 22 cytopathologists. Despite the total lack of clinical information, each cytopathologist, on the average, managed to correctly diagnose 33 of the 35 cases. There were more false diagnoses among the fibroadenomas than among the cancers.     

CYBEST Model 4. Automated cytologic screening system for uterine cancer utilizing image analysis processing

CYBEST (Cyto-Biologic Electronic Screening System) utilizes image analysis technology for the automated prescreening of cervical cytology specimens. CYBEST Model 3, which includes a television scan system and automatic shading control, achieved our initial goal of rapid specimen processing (no more than three minutes to achieve a final specimen assessment). This paper describes CYBEST Model 4, developed in 1981; with the minicomputer of Model 3 replaced by a microcomputer, Model 4 is considerably smaller, about the size of a business desk. A new parameter, the intranuclear configuration (chromatin pattern), was added to the four parameters used in Model 3. The five parameters now used for the assessment and ranking of cytologic abnormalities are nuclear size, nuclear-cytoplasmic ratio, nuclear optical absorption, nuclear shape and intranuclear configuration. The other features of Model 4 are almost the same as those of Model 3. As an optional function, individual parameter measurement data, assessment of atypicality grade and cell images can be displayed on the CRT monitor by pointing to a cell with a light pen system. After completion of screening of a specimen, the ten cells judged to be most abnormal can be called automatically into the microscope optical field or the CRT monitor (in order ranging from the cell with the highest atypicality rank down) along with their associated data and the system's assessment of the specimen. By connection to a small business computer, all data can be transferred to a floppy disk for later retrieval.     

Three-dimensional image processing for morphometric analysis of epithelium sections.

The reproducible classification of poorly differentiated abnormal epithelium specimens is still a diagnostic problem. The computer-aided method described here improves the differentiation between benign and malignant epithelium specimens. Hematoxylin and eosin-stained sections of normal squamous epithelium, dysplasia, carcinoma in situ, and carcinoma were scanned in a TV microscope system and analyzed by means of image processing methods on a DEC 5000/200 workstation. From the 15-20 microns thick histological sections, 3-5 focus positions in steps of 1-4 microns were scanned. The segmentation of the cell nuclei was performed automatically by color analysis and geometric operations. For each nucleus the best focus level was selected and at this level the center of the cell was calculated. Graph theoretical methods were applied to analyze the morphometry of the epithelium specimens. The minimal spanning tree was computed in the three-dimensional (3D) space of the sections with the selected centers of the nuclei as vertices. The best feature found for discrimination of the specimens is the average length of all edges in a tree. In the two-dimensional (2D) analysis we had to accept an error probability of about 20% in differentiation of dysplasia and carcinoma. In contrast to this we differentiated normal squamous epithelium, dysplasia, and carcinoma with a correct classification rate of 100% in the 3D analysis.