A neural network system for the diagnosis of breast cancer

A medical decision support system based on neural networks has been developed to be used as an aiding tool for the diagnosis of breast cancer. The system receives a digital mammogram as input and gives one of three possible answers as output: suspicious of malignant breast cancer, suspicious of benign breast cancer, and without suspicion of bi east cancer. The system consists of an input layer made of a set of identical single-layer neural networks with localized receptive fields, without overlapping, in the mammographic image. This input layer is connected to a two-layer perceptron trained with the backpropagation algorithm. The network used in the input layer was trained previously via the competitive learning algorithm with regions taken from several mammograms to become a feature extractor. The system's performance was evaluated in terms of its ability of discriminating malignant from non-malignant cases. The area under ROC curve was 0.84.