Weakly Supervised Breast Lesions Detection in Dynamic Contrast Enhancement Magnetic Resonance Imaging

Breast cancer detection is a key part of breast computer-aided detection system, which has its significance for assisting doctors to diagnose breast cancer. The objective is to develop a weakly supervised approach via deep learning for breast cancer detection in dynamic contrast enhancement magnetic resonance imaging, and to reduce the cost of manually marking the database and doctors' workload, improves breast cancer detection rates. In our ablation experiments, the proposed structure is effective. The performance of classification network showed 94.67% accuracy, the weakly supervised detection showed the IoU of 0.803.