Computer-assisted pathology of intraepithelial adenocarcinoma and related lesions: 3-D distribution, structural...

To discriminate among intraepithelial neoplasms, we have been relying on tissue microscopy, but pathologists' subjectivity sometimes impairs diagnosis. Even an individual pathologist is sometimes unable to reproduce exactly his or her own previous diagnosis. Are various atypical lesions classifiable in a reproducible way, and if they are, how? The reliability of a diagnosis will be strengthened if we can define the "natural" categories inherent in cells and tissues. Morphometry and statistical analysis using a computer can provide answers. Atypia, a morphological feature of carcinoma, is essentially multivariate. Quantification of a tissue feature requires reducing it to a set of ten or more quantities, including size, shape and position of the nucleus, nucleolus, and the cell itself. The grade of aberration from the norm can be assessed only by a synthetic approach, using a computer for multivariate cluster analysis. This classification has been attempted in adenocarcinoma and related lesions of the lung and pancreas. The categories thus established are reproducible, because the lesions fall into divisions according to their forms. We can also examine the organ distribution of intraepithelial neoplasms by three dimensional (3-D) computer-assisted mapping. To reach a higher level of reliability, as many meaningful features as possible should be taken into account. Particularly, we emphasize the significance of architectural pattern as a biomarker for intraepithelial glandular neoplasms. Computer-aided 3-D structural analysis visualizes the basic skeleton of these neoplasms around which the cells adhere. Instead of the dichotomous tree pattern of normal glands, the tumors basically harbor a 3-D network, tubular or porous, which increasingly deviates from the norm along with the transition from adenoma to well to moderately to poorly differentiated adenocarcinoma. This structural aberration, if recognizable on 2-D sectional images, will serve as a surrogate endpoint biomarker for glandular tumors.