Research On Super-resolution Algorithm Of Magnetic Resonance Image Based On Deep Learning

Magnetic resonance imaging(MRI)is a non-invasive in-vivo imaging technique with advantages of high contrast and radiation-free.Clear images can provide great help for doctors to make a correct diagnosis.But the factors such as patient motion and equipment noise can lead to a decrease image quality.Super-resolution reconstruction technology can improve the image quality without changing the hardware equipment,which has the advantages of low cost and significant effect,and has important application and research value in the field of medical image reconstruction.In this paper,the work was carried out for the shortcomings of existing super-resolution reconstruction algorithms as follows:1)To handle the unclear and untrue reconstructed image texture,the texture migration mechanism was introduced into the image super-resolution algorithm,formed a referencebased Super-Resolution.The VGG19 network is used for feature extraction,and feature exchange is performed based on the feature similarity;the residual network is used to transfer the similar texture from the reference image to the low-resolution image.The experimental results showed that the closer the texture of the low-resolution image and the reference image,the higher the quality of the reconstructed image;even when there is a gap between the texture of the low-resolution image and the reference image,the quality of the reconstructed image is still better than traditional algorithms.2)To handle the inadequate feature extraction in traditional algorithms,in this paper,feedback mechanism was applied to image super-resolution algorithm,formed an image super-resolution algorithm based on the feedback mechanism.The information refinement module constructed by dense jump connections feeds back the high-level information of the image,refines the low-level information of the image,and enables the network to better extract high-level features.Through the fusion of high-level features and low-level features,high-quality images are reconstructed.The experimental results show that,compared with the traditional algorithm,the image reconstructed by the image super-resolution reconstruction algorithm based on the feedback mechanism has more detailed information similar to the original high-resolution image,and does not increase the error information.Through research,when the reference image with similar texture to the low-resolution image can be obtained,the super-resolution algorithm based on the reference image could obtain the reconstructed image with more real texture details;In the case that the reference image with similar texture failed to be obtained,it was available to use the image superresolution algorithm based on feedback mechanism to reconstruct the high-quality image.