The application of image analysis and neural network technology to the study of large-cell liver-cell dysplasia and hepatocellular carcinoma.

Liver cell dysplasia (LCD) is considered a preneoplastic lesion, whose characterization and differentiation from hepatocellular carcinoma (HCC) and from the reactive changes seen in cirrhosis has been controversial. We studied 12 cases of LCD (large cell type) with image analysis techniques (IA) and compared the findings with those of HCC (n = 40), and a spectrum of non-neoplastic hepatic lesions including normal liver and cirrhosis (n = 49). A minimum of 200 Feulgen-stained nuclei were measured from each lesion with the CAS 200 image analysis system. The data were collected with the aid of CellSheet software. Thirty-four variables were measured, including geometric, textural, and photometric nuclear features and DNA ploidy. The data were analyzed with multivariate statistics and a backpropagation neural network (NN). Stepwise statistical analysis selected 22 variables that were statistically significant in the three groups with P values <.05. Various NN architectures were developed using these variables. The best NN architecture included a sigmoidal transfer function, 14 input, 16 hidden, and 3 output neurons. It trained to completion after 8,887 runs using 90% of the lesions. This NN yielded a 100% cross-validation rate for unknown cases. These data support the concept of LCD (large cell type) as a lesion that can be objectively distinguished from HCC and non-neoplastic liver. Our study also demonstrates the potential usefulness of IA for the evaluation of difficult histopathological problems.

A rule-based expert system for the automatic classification of DNA "ploidy" histograms measured by the CAS 200 image analysis system.

DNA "ploidy" histogram interpretation is one of the most important sources of variation in DNA image cytometry and is influenced by multiple technical factors such as scaling, selection of peaks, and variable classification criteria. A rule-based expert system was developed to automate and eliminate subjectivity from this interpretative process. Ninety-eight Feulgen stained histologic sections from patients with breast, colon, and lung cancer were measured with the CAS 200 image analysis system (Becton Dickinson, Santa Clara, CA); they included diploid (n = 42), aneuploid (n = 46), tetraploid (n = 7), and multiploid (n = 3) examples. The data was converted from listmode format into ASCII with the aid of CELLSHEET software (JVC Imaging, Elmhurst, IL). Individual microphotometric nuclear measurements were sorted to one of 64 bins based on DNA index. The 64 bins were then divided into 5 semi-arbitrarily defined ranges: hypodiploid, diploid, aneuploid, tetraploid, and hypertetraploid. The nuclear percentages in each range were calculated with EXCEL 4.0 (Microsoft, Redmond, WA). The histograms were divided into 2 equal sets: training and testing. The data from the training set were used to develop 16 IF-THEN rules to classify the histograms into diploid, aneuploid, or tetraploid. A macro was programmed in EXCEL to automate all these operations. The rule-based expert system classified correctly 45/50 histograms of the training set. Two tetraploid histograms were classified as aneuploid. Three multiploid histograms were classified as tetraploid. All histograms in the testing set were correctly classified by the expert system. The potential role of rule-based expert system technology for the objective classification of DNA "ploidy" histograms measured by image cytometry is discussed.

Quality assurance issues in DNA image cytometry.

DNA image cytometry is currently used in many clinical laboratories as a prognostic tool for the study of patients with malignant neoplasms. However, quality assurance (QA) procedures for image cytometry (IC) have not been standardized. National proficiency testing programs for DNA IC are yet to be developed. We describe our QA program for DNA IC. Indicators of quality; thresholds for evaluation; methods for collection, organization, and evaluation of data; and actions recommended to deal with problems are described for each of the phases of DNA IC. They include QA procedures for sample acceptability; staining technique; instrumentation, including linearity and calibration verification; and correlation with other cytometric methods. The difficult problem of standardizing the postanalytical phase of QA is discussed. The need for a consensus conference to adopt standardized QA procedures for DNA image analysis is emphasized.

Lymphoid-rich effusions. Diagnosis by morphometry using the CAS 200 System.

Lymphoid-rich effusions frequently present diagnostic problems in clinical cytology. In the authors' previous studies, most lymphoid-rich effusions had been correctly classified as benign lymphocytosis or malignant lymphoma by an experimental computerized interactive morphometry system, in which randomly selected lymphoid nuclear profile images were measured in Papanicolaou fixed and stained cytospin smears. The present study used the CAS 200 System and criteria from the previously described rule-based expert system to classify similar preparations of 134 lymphoid rich pleural, peritoneal, and pericardial effusions (90 benign lymphocytoses, 36 malignant lymphomas, and 8 chronic lymphocytic leukemias). A total of 98.9% of the benign lymphocytoses and 88.9% of the malignant lymphomas were correctly classified (predictive values of correct diagnoses 95.7% and 97.3%, respectively). Chronic lymphocytic leukemias could not be distinguished from benign lymphocytoses by nuclear profile areas. Optical density histograms of benign, lymphomatous, and chronic lymphocytic leukemias effusions are described. Advantages and limitations of image analysis and immunocytochemistry are discussed.

Hepatocarcinogenesis is the sequel to hepatitis in Z#2 alpha 1-antitrypsin transgenic mice: histopathological and DNA ploidy studies.

Z mutant-associated alpha 1-antitrypsin deficiency in human beings leads to hepatitis and, in some cases, hepatocellular carcinoma. To begin to delineate the molecular basis for the development of hepatocellular carcinoma in alpha 1-antitrypsin deficiency, we previously developed transgenic mice using human alpha 1-antitrypsin M and Z genomic clones. High-copy Z lineage mice (12 gene copies/haploid mouse genome; "Z#2") had hepatocytes distended with human alpha 1-antitrypsin deficiency globules. Hepatitis was present, and the morphological changes mimicked those observed in human alpha 1-antitrypsin deficiency-related liver disease. The numbers of hepatocytes containing alpha 1-antitrypsin globules decreased with age, and alpha 1-antitrypsin-negative nodular aggregates of hepatocytes increased in number and size. Hepatocytic dysplasia occurred as early as 6 wk and was almost universally present at 1 yr. Nodules of dysplastic cells demonstrating aneuploidy were seen as early as 10 wks. These became persistent, proliferative lesions. Dysplasia and aneuploidy distinctly increased with time and advancing microscopic stage as lesions progressed to malignancy. Tumors were seen after 1 yr as adenomas, which are aneuploid and most likely well-differentiated hepatocellular carcinoma, and borderline malignant lesions; and, in 82% of Z#2 mice 16 to 20 mo old, as invasive hepatocellular carcinoma. These observations suggest but do not conclusively prove that hepatocellular carcinoma in alpha 1-antitrypsin deficiency and other hepatic disorders arises as a result of a common, endogenously stimulated pathway for hepatocellular carcinogenesis.