Automated Tumor Segmentation in Breast-Conserving Surgery Using Deep Learning on Breast Tomosynthesis.

Abstract

Breast cancer is one of the leading causes of cancer-related deaths among women worldwide, with approximately 2.3 million diagnoses and 685,000 deaths in 2020. Early-stage breast cancer is often managed through breast-conserving surgery (BCS) combined with radiation therapy, which aims to preserve the breast's appearance while reducing recurrence risks. This study aimed to enhance intraoperative tumor segmentation using digital breast tomosynthesis (DBT) during BCS. A deep learning model, specifically an improved U-Net architecture incorporating a convolutional block attention module (CBAM), was utilized to delineate tumor margins with high precision. The system was evaluated on 51 patient cases by comparing automated segmentation with manually delineated contours and pathological assessments. Results showed that the proposed method achieved promising accuracy, with Intersection over Union (IoU) and Dice coefficients of 0.866 and 0.928, respectively, demonstrating its potential to improve intraoperative margin assessment and surgical outcomes.