Research On Medical Image Segmentation Algorithm Based On Deep Learning

In the process of clinical treatment,the precise segmentation of organs or lesions in medical images can significantly improve the efficiency of doctors.With the development of artificial intelligence technology and deep learning,image segmentation based on deep learning has been widely used in various industries,including medical image field.How-ever,there are still some challenging problems in medical image segmentation,such as the variable target size,the low contrast between the target and its surrounding area,and the presence of numerous difficult samples and samples with complex shapes.In this paper,we take polyp segmentation and retinal vessel segmentation as exam-ples to investigate the above problems and give the following basic ideas.Firstly,we fully exploit the characteristics of medical images in different segmentation tasks and design corresponding methods to improve the segmentation capability of the model.Further-more,advanced feature extractors are used to extract more robust and powerful represen-tational information.In addition,attention mechanism and multi-scale design are used to help deep networks focus on complex and scale-variant regions.Specifically,the work in this paper is divided into the following points:(1)A polyp image segmentation network(FEGNet)based on edge supervision and feedback attention mechanism is proposed.The network combines the feedback mecha-nism and attention mechanism to improve the recognition ability of the model on complex samples.In addition,a low-cost edge extractor is designed to obtain clear polyp edges by shallow supervision of the network.(2)A retinal vessel segmentation network(Swin-ASNet)based on RGB adaptive selection and Swin Transformer is proposed.According to the fundus map character-istics,this network proposes an adaptive selection aggregation module based on multi-scale design and attention mechanism,which can utilize the useful information in different channels.In addition,a high-low interaction module is designed to inject the high-level semantic information into the low-level detail information to further obtain accurate seg-mentation results.(3)A medical image segmentation system based on the Flask framework is designed and implemented.This system can segment the medical images uploaded by doctors and visualize segmentation results,which can assist the doctor diagnosis.This system realize the engineering implementation of deep learning algorithms.