Attention-Driven Multi-Sequence Mri Feature Learning and Fusion for Breast Cancer Diagnosis

The heterogeneity and complexity of breast lesions, along with the vast amount of imaging data provided by each sequence Magnetic resonance imaging (MRI), make breast cancer diagnosis highly challenging. Many studies use deep learning methods for multi-sequence MRI joint analysis to improve diagnostic performance. However, existing approaches often ignore the requirements for distinct feature learning processes for different sequence MRIs. Additionally, they have not fully leveraged or discriminated relevant information during modeling, limiting the effectiveness of the model. Inspired by the diagnostic workflow of radiologists, we propose an Attention-driven Feature Learning and Fusion (AFLF) framework to classify benign and malignant breast lesions using multi-sequence MRI. Our framework employs an attention-based encoding network to learn attention-aware representations for each sequence MRI, and an adaptive attention learning fusion module to interact and integrate these representations, thereby enabling efficient breast cancer diagnosis. Experimental results demonstrate that AFLF achieves superior performance compared to state-of-the-art algorithms.