Research On Medical Image Super Resolution Reconstruction Based On Convolutional Neural Network

The quality of medical images has an important impact on the analysis and diagnosis of the patient's condition.However,due to factors such as equipment and environmental conditions,it is sometimes impossible to obtain clear medical images,which will affect the correct analysis and understanding of the disease.Image super-resolution reconstruction refers to the process of reconstructing high-resolution images using one or more low-resolution images.Applying super-resolution reconstruction technology to the field of medical images to improve the definition of medical images and improve the quality of medical images is of great significance for the diagnosis and treatment of the disease.With the rapid development of deep learning,it has achieved great success in computer vision such as image classification and face recognition.In recent years,compared with traditional algorithms,the application of deep learning to image super-resolution reconstruction has achieved good results.In order to improve the reconstruction quality and reconstruction speed of medical images,this paper focuses on the problems existing in the current super-resolution reconstruction algorithms based on convolutional neural networks.The main research work includes:(1)Medical image super resolution reconstruction based on deep residual network: aiming at the problems of the current network model based on convolutional neural network super-resolution reconstruction algorithm,such as simple network model,insufficient feature extraction,and blurred image texture,a deep residual network model is used to fully improve the clarity of medical images.Firstly,the depth of the convolutional neural network is deepened to fully extract the detailed features of the image,and then the global residual network structure is used to accelerate the deep network convergence and avoid the gradient disappearing.Finally,the parameters are reduced by selecting appropriate medical image data sets and performing mixed training.The experimental results show that the algorithm has achieved good results in both subjective visual and objective evaluation indicators,and it has certain significance for doctors to improve the quality of diagnosis.(2)Fast medical image super resolution reconstruction based on sub-pixel convolution: aiming at the problem that the complexity of the network structure increases and the speed is slowed down,a fast network model based on sub-pixel convolution is proposed.Firstly,the original low-resolution image without image preprocessing step is directly used as the input of the network to extract features,greatly reduce the complexity of the model,then add local residual units in the network to speed up the model convergence,while using a more efficient Adam optimization method,and finally use the sub-pixel convolution layer at the end of the network to rearrange the pixels to obtain high resolution images.Analysis of experimental results show that the proposed algorithm can not only improve the sharpness of medical images and ensure the quality of image reconstruction,but also improve the speed of image reconstruction.