Organ Confinement of Prostate Cancer: Neural Networks Assisted Prediction

Abstract

In this paper the problem of prostatic capsule penetration prediction is approached using artificial neural networks (NN). The neural networks input variables were data carefully selected and preprocessed from the database records of 650 patients treated by radical prostatectomy for prostate cancer. Different NN architectures and algorithms have been tested. A comparison with the performance of the best and most widely used prediction statistical method, the logistic regresion, has been done. In all the cases the NN models performed better than the logistic regression. The model with the best accuracy/complexity ratio was a recurrent NN. The obtained performances of a mean square error of 0.00428 and a global prediction of 96.07% are better than the results of similar experiments available in literature. Yet, in our opinion, the limitations of predicting performance of NN are given by the reduced dimension of the database and by the data gathering methods, therefore could be further improved.