Enhanced Semantic Segmentation for Triple- Negative Breast Cancer Detection Using Modified U-Net Architecture

Triple-negative breast cancer (TNBC) is an exceptionally aggressive type of breast cancer defined by the absence of estrogen receptors, progesterone receptors, and HER2 expression. Timely detection and precise diagnosis are vital for effective treatment and enhanced patient outcomes. This research introduces a new strategy for improving the semantic segmentation of TNBC cells in histological images by utilizing a modified U-Net architecture. Accurately segmenting TNBC cells in histological images poses significant challenges due to the complex and heterogeneous characteristics of breast tissue. Current techniques often fall short in capturing detailed nuances and separating cancerous cells from non-cancerous tissue. To tackle these issues, we propose a modified U-Net architecture that uses "same" padding throughout the entire network. This design choice ensures that the output segmentation mask matches the dimensions of the input image, enabling more accurate localization of cancer cells. Furthermore, the U-Net's natural ability to capture both contextual information and intricate details makes it particularly suitable for this objective. Our proposed modified U-Net architecture outperformed baseline methods, achieving superior segmentation accuracy and better capturing the intricate features of TNBC cells. The implementation of "same" padding significantly contributed to improving the localization accuracy of the segmentation masks.