Quality of Treatment Planning Evaluation for Head and Neck Cancer Using Artificial Neural Networks Intelligence System

摘要

Three types of artificial neural networks (ANNs) are instructed by three different training algorithms to effectively evaluate the quality of the Head and Neck cancer (HN) treatment plans. One hundred sets of HN treatment plans are collected to be the input data of the neural networks. Three ANNs including Elman (ANN-_E), feed-forward (ANN-_(FF)), and pattern recognition (ANN-_(PR)) were trained by using three different models, i.e., leave-one-out (Train-_(loo)), random selection (Train-_(random)), and user defined (Train-_(user)) method. The conformal index (CI) and homogeneity index (HI) are used to be the feature values and to train the neurons. The networks with higher accuracy are ANN-_(-PR-(loo)) (93.65±3.60)%, ANN_(FF-(loo)) (88.05±5.84)%, and ANN-(E-(loo)) (87.55±5.86)%, respectively. The ROC curves show that the ANN-_(-PR(loo)) approach has the highest sensitivity, which is 99%. It can be concluded that ANN-_(PR-(loo)) is a better choice for evaluating the quality of treatment plans for HN, this method reduces the amount of trail-and-error during the iterative process of generating inverse treatment plans.