3D MR Image Super Resolution Reconstruction Algorithm Based On Deep Learning

Magnetic resonance imaging(Magnetic Resonance Imaging,MRI)technology is a non-invasive,high-tissue soft contrast and multi-parameter imaging method.Compared with other imaging technologies such as X-ray and ultrasound,it has unique advantages,so it quickly becomes one of the most important clinical diagnostic tools.During image post-processing and pathological evaluation of body organs,high-resolution MR images are essential.However,the resolution of magnetic resonance images is limited by various factors such as hardware conditions,patient comfort,signal-to-noise ratio,and scan time.An efficient and cost-effective solution is to apply single image super resolution(SISR)technology to magnetic resonance images.Single image super-resolution refers to the use of computer technology to restore high-resolution images from a single low-resolution image without changing hardware conditions.At present,with the extensive application of deep learning technology in the field of computer vision,the image super-resolution reconstruction technology based on convolutional neural network(CNN)has achieved remarkable results in 3D-MR image super-resolution reconstruction.This paper focuses on the 3D-MR image superresolution reconstruction algorithm based on deep learning,the main work is as follows:(1)In view of the fact that the existing deep learning-based 3D-MR image superresolution reconstruction algorithm uses a large number of 3D convolutions,which leads to the problem of excessive network parameters and high calculation costs,this paper proposes a separable 3D convolution for replacement 3D convolution.Separable 3D convolution reduces the number of parameters and calculation cost through spatial decomposition of convolution,and uses residual learning to solve the training difficulty caused by the deepening of the network after the solution of the convolution integral.In this paper,the Separable 3D convolution is used to replace the 3D convolution in the existing algorithm ReCNN,and the S3D-ReCNN algorithm is proposed.Experimental results show that our proposed 3D separable convolution can greatly reduce the number of network parameters and computational cost,while the performance only slightly decreases.(2)Based on separable 3D convolution,this paper designs and implements a 3DMR image super-resolution reconstruction algorithm based on S3D-RBDN network.The S3D-RBDN network makes full use of the hierarchical features in the original lowresolution image through residual learning and dense connection,which provides more useful information for image reconstruction,which can improve the quality of the reconstructed image.Experimental results show that the proposed S3D-RBDN has certain advantages over other existing algorithms in reconstruction performance.