Quantitative morphology of human cirrhotic livers. Part II: The statistically adequate morphological classification of liver cirrhosis. Multivariate analysis from quantified data of form.

Abstract

In a previous report, we developed methods by which to quantify various patterns of cirrhosis. A set of parameters were used: (i) the mean nodular radius; (ii) the coarseness; (iii) the mean septal thickness; and (iv) the degree of nodular separation. This was applied to 70 cirrhotic livers in an attempt to establish a reproducible classification, and the data were subjected to four-dimensional cluster analysis (Ward method) using a mainframe computer. Five clusters appeared: cluster A, fine nodules, thin septa; cluster B, coarse nodules, relatively thin septa; cluster C, fine nodules, thick septa; cluster D, extremely coarse nodules; and cluster E, coarse nodules, thin septa and incomplete nodulation. Of these, cluster A was considered to correspond to the nutritional type of Gall, cluster B to posthepatitic type (Nagayo's Type B), cluster C to postnecrotic type (Nagayo's Type A) and cluster E to the subtype of posthepatitic cirrhosis with incomplete nodulation (Miyake's Type BI). Cluster D comprised cases with extremely coarsened nodules. The reproducibility of this clustering was fully ensured by linear discriminant analysis. Another discriminant analysis, the canonical, allowed us to visualize the separation of clusters in a two-dimensional (2-D) scatter diagram. We thus managed to establish a most adequate classification from a geometric and statistical point of view.