The combination of a decision tree technique with the computer-assisted microscope analysis of Feulgen-stained nuclei to assess aggressiveness in lipomatous and smooth muscle tumors.

The present study describes a computer-assisted methodology whose purpose is to reduce the degree of subjectivity in the diagnosis of soft tissue tumors. This methodology associates three complementary techniques, namely digital cell image analysis, the discretisation of numerical data and a Decision Tree technique (DT). The first technique relies on the use of the digital cell image analysis of Feulgen-stained nuclei, a technique which makes possible a quantitative and thus objective description of nuclei with the help of 24 numerical parameters (15 morphonuclear and 9 DNA content- (ploidy level and proliferation activity) related). The second technique transforms each numerical parameter into an ordinal one with a small number of values (2 to 4) so that only the relevant physical significance of the parameters is retained. The Decision Tree technique generates classification rules on the basis of the discretised parameters quoted above. This methodology was applied to 53 human soft tissue tumors which included 26 lipomatous tumors (13 malignant liposarcomas and 13 benign lipomas) and 27 smooth muscle tumors (11 malignant leiomyosarcomas and 16 benign leiomyomas). The results show that a distinction between benign (lipoma) and malignant (liposarcoma) lipomatous tumors can easily be made by means of simple logical rules depending on only four discretised cytological parameters (two ploidy- and two morphonuclear-related). In contrast, no stable or predictive characterisation can be obtained with respect to the difference between leiomyosarcomas and the leiomyomas. Hence, while lipomas and liposarcomas appeared to be two completely distinct biological entities, leiomyomas and leiomyosarcomas seem to involve a continuous biological process.