A Unified deep learning approach for the segmentation of breast masses and calcifications

Abstract

Breast cancer is a leading cause of mortality among women and early detection using computer aided diagnosis (CAD) systems assist in improving the survival rates. Segmentation is a crucial step in CAD systems as it helps to accurately extract breast abnormalities. This study proposes a unified deep learning based approach that can accurately segment breast masses and calcifications in contrast to studies that propose approaches to segment only a single type of abnormality. The proposed framework uses a modified Hybrid Transformer U-Net (HTU-Net) for segmentation of masses and U-Net for segmentation of calcifications. Quantitative results show that the U-Net achieved a Dice Similarity Coefficient (DSC) of 0.8022 and a precision of 0.8958 for segmentation of breast calcifications. For segmentation of breast masses, the modified HTUNet achieved a DSC of 0.6064 and a precision of 0.5988. These results demonstrate the effectiveness of the proposed unified approach in segmentation of breast masses and calcifications, highlighting its potential in being integrated to CAD systems.