Decision tree induction: a useful tool for assisted diagnosis and prognosis in tumor pathology

The aim of the present work is to show that decision tree induction algorithms are a useful tool for extracting reliable information from data series, with the objective of assisting pathologists in identifying specific diagnostic and prognostic markers in various types of tumor pathologies. In terms of accuracy, we show that the decision tree technique exceeds other more sophisticated techniques, such as multilayer neural networks. Furthermore, because of the case with which decision tree results can be interpreted (logical classification rules), new methodologies can be readily developed to further assist in analyzing complex data that mix heterogeneous features. In this paper, we illustrate such capabilities in the context of different complex diagnostic and/or prognostic problems in tumor pathology relating to bladder, astrocytomas, and adipose tissues.