THE CONTRIBUTION OF IMAGE CYTOMETRY AND ARTIFICIAL INTELLIGENCE-RELATED METHODS OF NUMERICAL DATA-ANALYSIS FOR ADIPOSE TUMOR HISTOPATHOLOGIC CLASSIFICATION

Thirty-five lipomatous tumors were quantitatively described using 47 v ariables generated by means of computer-assisted microscope analysis. Of these 47 quantitative variables, 27 were computed on Feulgen-staine d specimens (25 on cytologic and 2 on histologic samples) and, of the remaining 20, 8 related to vimentin and S-100 protein immunostaining p atterns and the other 12 to the glycohistochemical staining patterns o f peanut agglutinin, succinylated wheat germ agglutinin, and concavali n A agglutinin. The 35 lipomatous tumors included 6 atypical lipomas a nd 8 well differentiated, 5 dedifferentiated, 6 myxoid, and 10 pleomor phic liposarcomas. The actual diagnostic value contributed by each of the 47 variables with respect to the 5 lipomatous tumor groups was det ermined by means of the decision tree technique, an artificial intelli gence-related algorithm that forms part of the supervised learning alg orithms. Of the 47 quantitative variables, the decision tree technique retained 8: ie, 2 tissue architecture-, 2 DNA ploidy level-, 2 morpho nuclear-, 1 lectin histochemical-, and 1 vimentin immunostain-related variables. The decision tree technique made use of these 8 variables t o set up logical rules that make it possible to identify atypical lipo mas from well differentiated liposarcomas, on the one hand, and dediff erentiated liposarcomas from those that are well differentiated and pl eomorphic, on the other. Thus, the combination of an artificial intell igence algorithm analyzing quantitative variables generated by means o f the computer-assisted microscope analysis of cytologic and histologi c samples from lipomatous tumors can be considered an expert system co ntributing significant diagnostic information to conventional diagnosi s.