Artificial neural networks for early breast carcinoma detection

This paper shows a model designed for an efficient detection of patients with breast carcinoma (BC), using an artificial neural network (NN). A multilayer NN is used as a convenient modeling tool for the identification of a chosen set of risk factors and symptoms which characterizes the patients suffering from BC. We used a properly selected representative learning set of patterns referring to 60 patients for the BC identification process (the NN learning task). The other set of patterns referring to 140 patients was used as a test set for verification of the proposed model. The instructed NN successfully identified 198 test patterns, which means that identification accuracy was greater than 98%. Previously set hypothesis, which claimed that hyperspace of weight matrix enables an efficient separation of healthy group of patients from group suffering from BC, proved to be correct. Expert knowledge was used for proper selection of corresponding sets of risk factors and symptoms on one side and for selection of a representative set of learning patterns on the other side.