An improved segmentation algorithm of CT image based on U-Net network and attention mechanism

Abstract

Medical image segmentation is one of the important steps in clinical diagnosis, and accurate segmentation of lesions is of great significance to clinical treatment. Therefore, a CT image segmentation algorithm based on depth learning is proposed to solve the problems of poor robustness, weak anti noise ability and low segmentation accuracy of existing image segmentation algorithms.. Firstly, we improve the u-net network structure, increase the batch standardization layer to improve the robustness of the network model, and introduce the attention mechanism to focus on specific things according to the needs, improve the recognition ability of the model. Then, the improved U-Net network structure is applied to CT image segmentation, and the cross entropy loss function is used to reduce the possibility of insufficient segmentation and segmentation leakage, and improve the accuracy of image segmentation. Finally, on the basis of data preprocessing, the segmentation network is trained to get the image segmentation model based on deep learning, and the prediction is made on the test set to get the segmentation results. The experimental results show that compared with other algorithms, the proposed method achieves 0.9594 in the common evaluation standard Dice coefficient, and has strong robustness. It can accurately segment the lung organs in CT images, which is helpful for doctors to obtain pathological information and assist in the diagnosis of lung diseases.