Journals Proceedings

Abstract

We present the results of an experimentation with dynamic features for breast cancer detection in Contrast-Enhanced Magnetic Resonance of the female breast. In order to understand how good the various features are we built a dataset from real dataset with proven histological diagnosis. We compared human-readable features, commonly used in the clinical practice, with a non-linear artificial neural network trained with a double k-fold cross validation. The results show that the ANN reaches very good results when two specific dynamic features are used. The particular validation procedure used in this experimentation allows us to better understand the discriminative power of the various approaches and move toward a better classifier that might be used in the clinical environment. Breast cancer, in fact, is the second form of diagnosed cancer among women living in the western countries and any improvement in its diagnosis would lead to a lower mortality rate.