Advancing Breast Cancer Diagnosis: Attention-Enhanced U-Net for Breast Cancer Segmentation

Breast cancer segmentation is a pivotal aspect of the diagnostic and treatment process for effective breast cancer management. This paper introduces a novel approach to breast cancer segmentation by proposing an attention-enhanced U-Net architecture. The primary objective is to elevate the precision and efficiency of the segmentation process. Within our proposed architecture, attention blocks are seamlessly integrated into the U-Net framework, empowering the model to discern and prioritize critical spatial regions characterized by high saliency. Our results indicate a substantial enhancement in tumor segmentation accuracy when compared to models lacking these attention-focused layers. Specifically, our approach achieves an impressive Dice similarity coefficient of 90.5%, as demonstrated on a dataset comprising 510 images. Notably, incorporating saliency-driven attention mechanisms holds significant potential for augmenting accuracy and resilience in analyzing medical images related to various organs. This is achieved by providing a mechanism to assimilate specialized knowledge tailored to specific tasks within deep learning frameworks. Additionally, our comparative analysis against YOLOv5, YOLOv8, and Mask R-CNN highlights the efficacy of our attention-enhanced U-Net in breast cancer segmentation tasks. This innovative model contributes to advancing breast cancer diagnosis and showcases promise in broader medical imaging applications, fostering a more nuanced and specialized approach within deep learning paradigms.