Research On Medical Image Segmentation Algorithm Based On Residual Network And Attention Mechanism

In recent years,medical image segmentation technology has played an increasingly important role in medical research.With the help of medical image segmentation,abnormal parts in various medical images,such as histiocyte images and organ CT images,can be accurately identified and segmented.This is essential for medical students and researchers to analyze diseased parts of patients and formulate customized treatment options.In the past few years,deep learning-based semantic segmentation algorithms,such as Unet,have made significant progress in medical image processing.But the traditional convolution operation-based algorithms have limitations in realizing global semantic information interaction and high-precision output needed for accurate medical segmentation tasks.To address this problem,we propose two improved medical image segmentation network models that combine the advantages of residual networks and self-attention mechanisms.These models have been tested to enhance the accuracy of segmentation tasks in medical images.This article proposes two new deep learning-based medical image segmentation network models that leverage residual networks and self-attention mechanisms to achieve greater accuracy than existing methods.The first model,called RS-Net,processes cellular images by using an improved residual network that extracts more high-precision semantic features and solves the gradient problem in the training process.The decoder introduces attention modules to reduce semantic differences.RS-Net outperformed other deep learning-based algorithms on two medical cell datasets in terms of evaluation metrics including Dice,Io U,and HD.The second model,called ST-Net,focuses on retinal image segmentation using a neural network that leverages both deconvolution modules and Vision Transformer to reduce computational burden and obtain remote dependencies between feature maps of different scales respectively.Comparative experiments on the DRIVE dataset have demonstrated that ST-Net can handle low scale and complex image features more effectively,and performs better than previous methods on Dice,Io U,and HD evaluation metrics.Finally,modular ablation experiments were conducted on RS-Net and ST-Net,including channel number ablation experiments on ST-Net.These experiments further validate the feasibility and efficiency of both models for various medical image segmentation tasks.