MRI Super-resolution Reconstruction Based On Frequency Separation

MRI images are widely used in clinical diagnosis,but obtaining high-resolution MRI images often requires longer scanning time or improved hardware performance,.Reconstruction of MRI images using image super-resolution reconstruction algorithms can greatly reduce the hardware requirements,improve the efficiency of patient treatment.However,the performance of traditional super-resolution reconstruction algorithms cannot meet the requirements of medical MRI images for accurate reconstruction results.The image super-resolution reconstruction algorithms based on deep learning can reconstruct more realistic images,but the existing super-resolution reconstruction algorithms have high model complexity,large number of parameters.And the structure of MRI images is complicated,there are still few algorithms applied to MRI image super-resolution reconstruction.Therefore,in this paper,we design a fast and accurate MRI image super-resolution reconstruction algorithm based on deep learning.This paper designed a MRI image super-resolution reconstruction algorithm based on the adaptive feature aggregation,this algorithm uses Dense Connect to achieve feature reuse and make full use of the features of all layers,Residual Connect to accelerate the convergence of the network.MRI images have complex tissue structure,and super-resolution reconstruction needs to utilize all level features.An adaptive feature fusion module based on the channel attention mechanism is designed so that the network can automatically select the features needed for reconstruction,and prevent the problem of inaccurate MRI images super-resolution reconstructed caused by the weakening or disappearance of feature information during the transmission process.The rationality of the model design and its superior performance are demonstrated through experiments.In order to ensure the speed and accuracy of super-resolution reconstruction,based on the above-designed algorithm,it is further optimized to design a frequency separation-based MRI image super-resolution reconstruction algorithm.From the perspective of frequency,this paper introduces a frequency separation module in the above algorithm,which makes the algorithm focus more on high-frequency information by continuously separating high-frequency and low-frequency information and doing different processing on them.Finally,the adaptive feature aggregation module is used to aggregate all the high-frequency features,low-frequency features,the output of high-frequency branching network and the output of low-frequency branching network to achieve the selection of different frequency features,which is used to reconstruct MRI images with rich details.The super-resolution reconstruction of MRI images at multiple scales demonstrates that the algorithm designed in this paper can achieve fast and accurate super-resolution reconstruction of MRI images.