The effectiveness of artificial neural networks in evaluating treatment plans for patients requiring external beam radiotherapy

This study was designed to determine the ability of a neural network to use data summarized in artificial generated dose volume histograms (DVH) for the rectum to evaluate and compare different patient treatment plans for the treatment of localized prostate cancer. One radiotherapist evaluated 250 artificially generated DVHs representing the distribution of a dose of ration throughout the rectum during external radiotherapy for patients with prostatic adenocarcinoma (PAC). The data were also analyzed using the Lyman NTCP-model for assessing complication probabilities. A neural network consisting of 10 input nodes and one output node was trained to categorize the plans according to the radiotherapist's score. The volume in each isodose was used as input, and the risk for a severe complication was presented as output. The classifications made by the neural network matched those determined by the radiotherapist and the NTCP-model. All three techniques showed a high correlation between each other. Artificially generated dose volume histograms (DVH) for the rectum can be used for training a neural network for scoring rectal DVHs in treatment plans for localized prostate cancer.