Neural Network Based Microwave Tumour Detection Using Breast Pairs

This study investigated the quality of tumour detection using radar microwave signals from paired breasts (simulating bilateral breast pairs) as input to a neural network classifier. The detection quality was compared to that of a classifier that used single breasts as test samples. The samples consisted of sinograms of experimental microwave scans of MRI-derived breast phantoms. The results demonstrate an improvement in detection quality of 3% in terms of the receiver operating characteristic - area under the curve (ROC AUC) and 5% in sensitivity for a classifier that used paired breasts. Notably, the classifier exhibited higher detection quality for smaller and denser breasts.