Breast Ultrasound Image Segmentation Based on Attention U-Net

Breast cancer remains a significant health challenge, with early diagnosis critical for improving patient outcomes. This study explores the application of the Attention U-Net model for breast ultrasound image segmentation, aiming to enhance diagnostic accuracy in breast cancer detection. Experimental results demonstrated the model's accuracy stabilizing at approximately 0.95, with predicted masks showing close alignment to expert-annotated ground truths. Additionally, Grad-CAM visualizations illustrated the model's capability to concentrate on critical regions, enhancing interpretability. Despite its computational demands, the Attention U-Net model offers significant potential for medical applications, providing a robust framework for improving clinical diagnosis and treatment planning in breast cancer care.