Different U-Net Variants for Segmentation of Histological Breast Images: An Analytical Comparison

Abstract

The diagnosis and treatment of Breast Cancer disorders depend on information from Breast image segmentation, which is a crucial task in medical image analysis. Convolutional neural networks (CNNs) have demonstrated outstanding performance in a number of medical picture segmentation tasks, including Breast image segmentation, in recent years. In this study, using a publicly available dataset, we assess the histopathological Breast image segmentation performance of three CNN models, specifically U-Net, U- Net++, and U-Net3++.The U-Net++ and U-Net3++ models are improved variants of the well-known U-Net model that were created to address the short comings of the original architecture. Despite U-Net3++ surpassing the other two models in terms of dice coefficient and surface distance, the experiments demonstrate that all three models obtained good accuracy. According to our findings, U-Net3++ is a promising Breast image segmentation model with the potential to increase the precision of Breast Cancer disease detection and therapy.