Deep Learning Models for Breast Cancer Segmentation using Mammography images: A comparative study

Abstract

Breast cancer is one of the most decisive cancers worldwide. Automated segmentation of breast tumors from medical images is crucial for early detection and effective treatment planning. This article proposes a comparative study of three deep learning techniques for breast cancer segmentation: UNet++, PSPNet, and TransUNet. Experimental results on regions of interest (ROIs) of CBIS-DDSM and InBreast datasets demonstrate that UNet++ outperforms PSPNet and TransUNet, achieving superior performance with a Dice coefficient of 0.7980 and IoU of 0.8342 on the InBreast dataset, and a Dice coefficient of 0.8654 and IoU of 0.8116 on the CBIS-DDSM dataset.