Weakly-Supervised Diagnosis and Detection of Breast Cancer Using Deep Multiple Instance Learning

The detection and classification of breast cancer lesions with computer-aided diagnosis systems has seen a huge boost in recent years due to deep learning. However, most works focus on 2D image modalities. Dealing with 3D MRI adds new challenges, such as data insufficiency and lack of local annotations. To handle these issues, this work proposes a new two-stage framework based on deep multiple instance learning, which requires only global labels (weak supervision) and provides: 1) classification of the whole volume and of each slice; and 2) 3D localization of lesions through heatmaps. Results show that the proposed approach achieves performances that are competitive with the state of the art, and a qualitative assessment of the heatmaps illustrates the effectiveness of this approach in finding the malignant lesion in the images.