Differentiation of cells in abnormal human liver tissue by computer image processing: a preliminary investigation to...

The initial development of algorithms for the analysis of multicolor multicellular images of liver tissue sections is described. Using a three-color image recorded by an automated light microscope, a sequence of cellular logic operators was applied to the task of differentiating between hepatocyte and leukocyte nuclei in a section taking advantage of both spatial and chromatic features, namely, that hepatocyte nuclei usually appear larger than those of leukocytes and, in red illumination, are likely to be more transparent. Optimally, the developed algorithms properly differentiated 90% of the hepatocyte nuclie and 85% of the leukocyte nuclie; this performance level appears to equal that of a microscopist. The microscopist, however, rarely analyses all cells; thus, computer differentiation of cells eventually may provide an important new dimension to clinical histopathology.