| In recent years,with the wide application of medical imaging technology in the examination of diseases and other aspects,a large number of medical images have been produced,and the analysis of medical images to obtain useful information needs to consume a lot of human,material and financial resources.In order to solve the problem that the analysis of medical image occupies a lot of resources,excellent and robust medical image analysis method is needed,and medical image segmentation is one of the main problems that medical image analysis needs to solve.As deep learning shines in various fields,it is also applied in the field of medical image segmentation.More and more medical image segmentation methods based on deep learning have been proposed to meet the needs of medical image segmentation in the medical system.While deep learning technology has achieved great success,it also has shortcomings.When extracting medical image features,the image segmentation method based on deep learning only pays attention to local information,without considering the global information of the image.To solve this problem,the main contributions of this paper are as follows:（1）Currently,image segmentation methods based on deep learning have achieved good results.However,most of these models do not take into account anisotropy and asymmetric features,which play an important role in describing the target profile.In order to solve this problem,a new loss function is proposed in this paper to be applied to the deep learning model,which is conducive to combining the features of image segmentation contour edges,so as to improve the stability of segmentation process and reduce the probability of outliers in segmentation results.This loss function can be used for 3D deep learning network models to calculate the edge contour of a single object in the image.By embedding the proposed loss function into the deep learning model based on convolutional deep network（CNN）,the end-to-end image segmentation of medical images can be carried out.The1 loss function and the proposed loss function were used for ablation experiments,and three data sets were used to verify the effectiveness of the loss function.Compared with recently designed methods to reduce boundary errors,the SOTA results of the loss function presented here are presented.（2）The loss function proposed above combines the edge features of image segmentation contour,but it can only calculate a single type of segmentation target in3D network model.In the field of medical image segmentation,there are a large number of excellent 2D image segmentation algorithms.In order to be used in 2D network model,the loss function is further refined in this paper,so that it can be applied to the segmentation calculation of multi-class region in 2D network.Then,this paper explores the effectiveness of the method combining the segmentation algorithm model and geometric model based on Transformer.In this paper,the 2D network model Swin-Unet based on Transformer is combined with the optimized loss function in this paper and verified on multiple medical data sets.The experimental results show that it is effective to embed the information of image edge contour into the loss function.（80）loss function is used to compare the loss function presented in this paper. |
