A probabilistic neural network for assessment of the vesicoureteral reflux's diagnostic factors validity

Abstract

This study examines Probabilistic Neural Network (PNNs) models in terms of their classification efficiency in the Vesicoureteral Reflux (VUR) disease. PNNs were developed for the estimation of VUR risk factor. The obtained results lead to the conclusion that in this case the PNNs can be potentially used towards VUR risk prediction. There is a redundancy in the diagnostic factors, so pruned PNN was used in order to evaluate the contribution of each one. Moreover, the Receiver Operating Characteristic (ROC) analysis was used in order to select the most significant factors for the estimation of VUR risk. The results of the pruned PNN model were found in accordance with the ROC analysis. The obtained results may support that a number of the diagnostic factors that are recorded in patient's history may be omitted with no compromise to the fidelity of clinical evaluation. © 2010 Springer-Verlag Berlin Heidelberg.