A prognostic-classification system based on a probabilistic NN for predicting urine bladder cancer recurrence

Abstract

In this paper our purpose was to design a prognostic-classification system, based on a probabilistic neural network (PNN), for predicting urine bladder cancer recurrence. Ninety-two patients with bladder cancer were diagnosed and followed up. Images from each patient tissue sample were digitized and an adequate number of nuclei per case were segmented for the generation of morphological and textural nuclear features. Automatic urine bladder tumor characterization as a potential to recur or not was performed utilizing a PNN. An exhaustive search based on classifier performance indicated the best feature combination that produced the minimum classification error. The classification performance of the PNN was optimized employing a 4-dimensional feature vector that comprised one texture feature and three descriptors of nucleus size distribution. The classification accuracy for the group of cases with recurrence was 72.3% (35/47) and 71.1% (32/45) accuracy for the group of cases with no recurrence. The proposed prognostic-system could prove of value in rendering the diagnostic nuclear information a marker of disease recurrence.