Sparse Representation-Based Image Reconstruction for Faster MRI Processing

This work is a contribution to a new approach for enhancing MRI reconstruction by the use of image processing by sparse representation reconstruction technique. Some of the popular MRI techniques scan a limited field and takes more time, hence hampering their clinical applicability. This study indicates that using compressive sensing and deep learning methodologies makes it possible to cut down the time of scanning tremendously while at the same time maintaining quality images. WMR uses sparse representation and, alongside with wavelet transform and Orthogonal Matching Pursuit algorithm for image reconstruction from undersampled k-space data. For that purpose, a convolutional neural network is used to improve the reconstruction; the network fixes the missing information in the undersampled data. The proposed methodology is tested on the NYU FastMRI dataset and the implementation is carried out in Python and PyTorch. MSE, PSNR, SSIM, and RMSE measurement indices show the fact that the proposed approach achieves significant improvements in succeeding rate while maintaining the diagnostic quality. The findings therefore support the fact that sparse representation with DL could indeed cut half the MRI acquisition time thus offering more clinical value. This approach has the likelihood of the transforming MRI working methods through improving patient comfort, cutting costs, and increasing efficiency of diagnostic imaging in medical MRI.