The use of artificial-intelligence neural networks in the evaluation of treatment plans for external-beam...

The purpose of our study was to evaluate whether a neural network is capable of evaluating different treatment plans for external radiotherapy produced by a 3D treatment planning system and presented as dose-volume histograms (DVHs). Three radiotherapists evaluated 27 treatment plans for the external radiotherapy of prostatic adenocarcinoma. DVHs for the dose delivered to the rectum and the bladder were presented. A commercially available neural network with 5x10 input nodes and two output nodes was modified to categorize the plans according to the score of the radiotherapists. The DVHs of the treatment plans were used as the inputs and accepted or not accepted were presented as the outputs. A comparison was made with different models for assessing complication probabilities. The neural network was able to accept or not accept the treatment plans according to the scoring made by the radiotherapists. If the radiotherapists disagreed, the network also expressed the span of opinions. Neural networks can be adapted to evaluate 3D dose-planning treatment plans presented as DVHs. It should be noted that the relation between the amount of data and the size of the neural network in this study was not optimal.