Enhancing Breast Cancer Segmentation with Attention Mechanisms in U-Net Architectures

Abstract

Breast cancer is still a major global health concern, as seen by the high incidence and death rates among women, which underscore its grave implications. Since treating this ailment necessitates quick identification and therapy planning, accurate segmentation is essential to medical imaging. The purpose of this work is to expand the choices for diagnosis and treatment by examining the effectiveness of Attention U-Net topologies in enhancing the accuracy of breast cancer segmentation. With a Dice coefficient of 97.99%, an Intersection over Union (IoU) of 93.41%, and an accuracy of 95.91% on the test set, the Attention U-Net outperforms both the standard U-Net and the Residual U-Net models when compared. These findings highlight the need of using deep learning frameworks to increase breast cancer segmentation accuracy and, as a result, provide improved patient care alternatives. As deep learning technologies evolve, the integration of Attention U-Net topologies may aid in breast cancer detection and therapy, resulting in better patient outcomes and aiding global efforts against the illness.