Breast Cancer Image Semantic Segmentation with Attention U-Net

Abstract

Semantic segmentation is to segment objects in an image into meaningful units. Among them, the basic idea of U-Net is to use low-dimensional as well as high-dimensional information to extract image features and enable accurate location identification. In this paper, we present a new model that combines Attention Gates with U-Net and evaluate the results through semantic segmentation with breast cancer datasets. To this end, this study proposes and tests a methodology for breast cancer image segmentation based on Attention U-Net. In conclusion, when comparing the performance with the existing U-Net, It can be seen that IoU is 0.069 higher than the existing U-Net. Thus, the proposed model enables better image semantic segmentation.