SU-Next: An Image Segmentation Model of Breast Tumors Based on U-Next and Attention Mechanisms
Dong Zhu, Yongfang Wang, Mengzhu Yang, Guoqiang Li, Chengchao Wang, Hao Dong, Xiangxu Kong · Journal of Physics Conference Series · 2023
Abstract Breast cancer is one of the most common malignant tumors endangering women’s health today, and timely screening and targeted treatment of such tumors can effectively reduce the incidence of breast cancer. The commonly used method of breast tumors diagnosis is to detect the pathological image of breast tumor to find the spread of cancer cells and the difference in tissue morphology over time. There are a large number of complex problems such as interlaminar variation characteristics and structural and morphological diversity in the pathological sections of breast tumors, and the tumors must be accurately segmented before breast tumor detection, and the accuracy of breast tumor image segmentation directly affects the results of subsequent detection. Aiming at the problem that the U-Next network model is not clearly defined in the marginal region of breast tumors, we propose a SU-Next network model. The model adds an attention module to U-Next, which focuses on the feature values related to the current task and discards irrelevant features. By testing on the BUSI breast dataset, the effectiveness of this method in breast image segmentation was verified. Taking IoU and Dice as evaluation indicators, they reached 64.26% and 77.40%, respectively, and compared with U-Next, SU-Next improved by 1% and 1.19% respectively.