Quantitative morphology of human cirrhotic livers. Part I: Parameter expression of pattern as a basis for computerized classification.

Abstract

Although classification is a long-used method of histopathology, a reproducible one has yet to be created. We established a most adequate classification of cirrhosis from a geometric and statistical point of view, by reducing its form to a set of quantities and submitting the data to multivariate analysis. In this article, methods for quantification are described as a preliminary step for the statistical treatment that appears in another paper. The pattern was reduced to a set of four quantities: (i) the mean nodular radius; (ii) the coarseness; (iii) the mean septal thickness; and (iv) the degree of nodular separation. A model of dispersed spheres with various radii r was employed to assimilate cirrhosis; r was assumed to follow a logarithmic normal distribution. The parameters of this distribution were estimated stereologically from measurements on microscopic sections of chord lengths lambda generated from nodules by a test line. The coarseness was defined as the volume % of nodules larger than 1.5 mm in r. The mean septal thickness was determined stereologically on a plate model, into which the actual septa were transformed without changing their volume or surface density. The degree of nodular separation p theta was defined as a two-dimensional parameter, based on the curvature of nodulo-septal borders. It was demonstrated in several examples how accurately a set of these quantities describes various patterns of cirrhosis.