Cardiac Image Segmentation Based on Improved U-Net
Guang Xiao Qiao, Ji Hong Song · 2022 International Conference on Image Processing, Computer Vision and Machine Learning (ICICML) · 2022
Due to the complexity of the whole heart tissue, low segmentation accuracy, easy adhesion of tissue and incomplete segmentation in CT images, this paper proposes a method to improve the whole heart segmentation of U-Net. According to the structural morphological characteristics of the human heart, this paper introduces the ASPP module into the coding layer of the U-Net network, and adds the attention mechanism to the jump layer connection of the U-Net network. In this paper, the improved whole heart segmentation algorithm is compared with other popular segmentation algorithms using the public dataset MMWHS. Experimental results show that the segmentation accuracy of the proposed algorithm in this paper reaches 89.74%, which improves the segmentation accuracy of the whole heart structure.