A Study of Hepatocellular Carcinoma Using Morphometric and Densitometric Image Analysis

Abstract

Hepatocellular carcinoma is often difficult to diagnose in cytologic material and biopsy specimens. To demonstrate the utility of image analysis in discriminating benign and malignant hepatocytes, 42 malignant cell groups were compared with 26 benign cell groups with a wide range of nuclear morphology in hematoxylin and eosin-stained histologic sections from 42 patients with hepatocellular carcinoma. Nuclear measurements were performed with a relatively inexpensive microcomputer-based image analysis system using a highly flexible imaging software package. Twenty-two nuclear morphometric and densitometric parameters were evaluated. The best single discriminator of benign and malignant cells was the nuclear major axis. Classification of the test samples using optimized linear discriminant functions achieved the following positive predictive values (PV+) and negative predictive values (PV-) for hepatocellular carcinoma: 95.0% PV+ and 85.7% PV- for the major axis; 90.5% PV+ and 84.6% PV- for five densitometric parameters; 100% PV+ and 86.7% PV- for three morphometric parameters; and 95.5% PV+ and 100% PV- for nine combined morphometric/densitometric parameters. These results demonstrate that multivariate linear discriminant functions of nuclear features measured by image analysis can be used to classify benign and malignant hepatocytes accurately.