A Novel Multi-View Fusion Model for Early Diagnosis of CEM Images

In this section, we will describe the model we propose. Currently, all CEM images and FFDM multi-view models use three fusion strategies: (1) view-wise, only fuse the information from homogeneous views of contralateral breasts; (2) breast-wise, only fuse the information from inhomogeneous views of the ipsilateral breast; (3) joint, fuse all views' information. Recent studies [] have confirmed that the third fusion strategy achieves the best results, and it also aligns with the actual diagnostic practices of experienced doctors. Our model is proposed under the third strategy and addresses the practical medical issues in early diagnosis of CEM images.