Leveraging Attention Mechanisms for Breast Cancer Diagnosis

Recent advancements in deep learning have significantly impacted breast cancer diagnosis, enhancing the accuracy and efficiency of detection and classification. In this study, we applied various attention mechanisms in different configurations to evaluate their effects on our model’s performance, specifically in terms of F1 score and accuracy. We utilized ConvNeXt for initial feature extraction, followed by the implementation of attention mechanisms such as SRM (Spatial Relation Module), GCT (Global Contextual Transformer), LCT (Local Contextual Transformer), and Triplet Attention. These attention mechanisms were arranged both in parallel and series configurations to explore their impact on feature extraction. Our experiments demonstrated that the choice and arrangement of attention mechanisms significantly influenced the model’s performance, with the Triplet LCT in the parallel configuration achieving the highest accuracy and F1 score.