Research On Super-Resolution Reconstruction Algorithms For Medical Images Based On Deep Learning

With the widespread application of medical imaging technology,medical images have become an important diagnostic aid in clinical practice.High-resolution medical images can provide more accurate and complete pathological information,helping doctors determine the location,extent,and nature of diseases,which is of great significance in clinical diagnosis and treatment.Super-resolution reconstruction technology can enhance the clarity of medical images without increasing costs,thereby improving the accuracy and effectiveness of diagnosis.Compared with traditional reconstruction algorithms,deep learning-based image super-resolution reconstruction can enable the model to achieve higher reconstruction accuracy and generalization ability.This article aims to improve the quality of medical image reconstruction by studying and analyzing the problems in current deep learning-based medical image super-resolution reconstruction algorithms,using CT and MRI images as the research objects,and proposing optimizations.The main tasks are as follows:(1)To address the issues of blurry image details and insufficient utilization of global information in existing medical image super-resolution reconstruction,we propose a medical image super-resolution reconstruction algorithm based on dilated separable convolution and an improved hybrid attention mechanism.This method combines the ideas of depth wise separable convolution and dilated convolution,using different receptive fields to extract features at different scales to enhance feature representation capability.We introduce an edge-channel attention mechanism to fuse edge information while extracting high-frequency features,thereby improving the reconstruction accuracy of the model.Considering the special characteristics of medical images,we use a hybrid L1 loss and perceptual loss function as the overall loss function to make the reconstructed images more visually pleasing to human perception.We conduct comparative experiments on the dataset,and the results show that the proposed model outperforms the compared algorithms in terms of PSNR and SSIM indicators and significantly improves the visual clarity of the images.(2)To address the problem that most current image super-resolution reconstruction algorithms fail to make full use of shallow image features and tend to ignore the global structure,a medical image super-resolution reconstruction algorithm based on two-way gated feature fusion is proposed.The method uses the residual structure as a whole,and uses a basic block structure consisting of residual dense blocks and multi-scale channel attention modules to extract image features at different stages while making the network more focused on effective information.The two-channel gated feature fusion module is used to interact information between the feature maps at each stage and determine the fusion weight of different feature maps to effectively extract and utilize the features at each level of the image to achieve better reconstruction results.Using pixel attention to weight different features at the pixel level,the medical images are finally reconstructed using subpixel convolution.The experimental results show that the proposed model performs well in terms of evaluation metrics,and the visual effect also enhances the effect and texture features of the image and completely restores the overall image structure.