METHODOLOGICAL ASPECTS OF USING DECISION TREES TO CHARACTERIZE LEIOMYOMATOUS TUMORS

The aim of the present work is to present the potential uses of a clas sification technique labeled the ''decision tree'' for tumor character isation when faced with a large number of features, The decision tree technique enables multifeature logical classification rules to be prod uced by determining discriminatory values for each feature selected, I n this report, we propose a methodology that used decision trees to co mpare and evaluate the information contributed by different types of f eatures for tumor characterisation, This methodology is able to produc e a set of hypotheses related to a diagnosis and or prognosis problem, For example, hypotheses can be producted (on the basis of a set of de scriptive features) to explain why tumor cases belong to a given histo pathological group. To illustrate our purpose, this methodology was ap plied to the difficult problem of leiomyomatous tumour diagnosis, The aim was to illustrate what kind of diagnostic information can be extra cted from a sample data set including 23 smooth muscle tumors (14 beni gn leiomyomas and 9 malignant leiomyosarcomas) described by a large se t of computer-assisted, microscope-generated features, Three groups of features were used relating to: (1) ploidy level determination (10 fe atures), (2) quantitative chromatin pattern description (15 features), and (3) immunohistochemically related antigen specificities (6 featur es), All these features were quantified by digital cell image analysis , The results suggest that an objective distinction between leiomyomas and leiomyosarcomas can be established by means of simple logical rul es depending on only a few features among which the immunohistochemica lly revealed antigen expression of desmin plays a preponderant part, O ne of the combinations of features proposed by the methodology is inte resting for pathologists, because it includes two features describing the appearance of a nucleus in terms of chromatin distribution homogen eity and density, two features widely used by pathologists in tumor-gr ading systems. (C) 1996 Wiley-Liss, Inc.