Research On Medical Image Segmentation Algorithm Based On Improved U-Net

With the continuous development of computer technology,artificial intelligence has been widely applied and developed in various fields.In the field of medicine,computer-aided diagnosis and treatment technology plays a key role,and medical image segmentation is one of the most important steps.The purpose of medical image segmentation is to segment the valuable parts in the image,thus providing assistance for doctors to analyze and diagnose medical images.However,medical images usually contain a large amount of noise and artifacts,and different tissue structures often have different sizes,shapes,and positions.Therefore,there are still many problems to be solved in the task of medical image segmentation.Traditional image segmentation algorithms cannot accurately segment medical images,but the development of deep learning technology provides new opportunities for medical image segmentation.This article proposes two different improvement methods based on the U-Net network to address the above problems and challenges.This model can effectively improve the segmentation accuracy of medical images.The main research work includes the following aspects:(1)This article proposes a CT image segmentation model based on U-Net to improve the accuracy of traditional networks in CT image segmentation.In this model,the authors combine residual structures and attention modules to replace the original convolution operation,avoiding the problem of network degradation and enabling the model to focus on important feature information.In the skip-connection part,an Attention Gate attention gating mechanism is also introduced to effectively reduce noise interference,suppress irrelevant regions in the input image,and enhance salient features that are useful for specific tasks,thereby improving the segmentation accuracy of the model.In the experimental section,the authors tested the model on lung and liver datasets and compared it with other methods.The experimental results show that the proposed method achieves better segmentation results compared to other methods.This indicates that the model has high potential and practical value in medical image segmentation.(2)In the task of retinal vessel segmentation,traditional segmentation models have difficulty dealing with the different shapes,thicknesses,and sizes of retinal vessels,as well as the influence of lighting and noise.To address this issue,this chapter proposes a new retinal vessel segmentation network based on the U-Net structure.The network introduces a multiscale feature extraction module and a self-attention mechanism to better extract features and obtain global spatial information.The multi-scale feature extraction module replaces the original 3x3 convolution operation,allowing the network to obtain different-sized receptive fields through internal residual structures and extract multi-scale information from the input feature map.Additionally,the module introduces an SE channel attention mechanism,which can weight different channels of input information and suppress the influence of irrelevant information.Between the encoder and decoder,a self-attention mechanism is introduced to effectively supplement the original network’s shortcomings in obtaining global spatial information.Experimental results on the DRIVE,STARE,and CHASE＿DB1 datasets demonstrate that the proposed network outperforms other models in terms of experimental metrics and can achieve better segmentation results in retinal vessel segmentation tasks.