چكيده به لاتين
Reconstruction of MR images is always a controversial inverse problem in medical imaging. Acceleration of MR scanning is of great importance for clinical, research and advanced applications, and one of the main efforts to achieve this is the use of compressed sensing (CS) theory. Compressed sensing is a promising approach to accelerate MRI. It aims to reconstruct the MR image using a small number of sampled data in k-space. Compressed sensing magnetic resonance imaging (CS-MRI) makes it possible to significantly shorten the data collection time. Nevertheless, existing CS-MRI approaches still have limitations such as loss of fine structure or high computational complexity. This research proposes a MR image reconstruction from sampled k-space data. In this research, we propose a new framework for incorporating sparse transformations in compact sensing MRI. One of the famous methods that has recently received attention is the fitting and running method with half-quadratic split splitting (PnP-HQS), in which the inverse problem for reconstructing MR images can be broken into two sub-problems and each one can be solved separately. solved The remarkable point in the reconstruction with this method is that one of the two subproblems can be replaced by a noise reducer, which, in addition to helping the SL0 algorithm to escape from local maxima to approach the regions of global maxima, increases the accuracy of the simple SL0 algorithm to the limit Grid-based algorithms increased. Also, in order to prevent permanent removal of features from the image that are important, the feature modifier is built in to prevent the removal of the main features, especially the edges of the image, which are important in medical imaging, by the noise remover. The proposed method consists of four steps: noise target, feature modifier, SL0 cost function optimization, and projection. By integrating BM3D, PnP-HQS, and CSMRI equipped with fixed transforms, we improve the SL0 algorithm for MRI reconstruction without introducing additional complexity. The aim of this project is to develop a simple SL0 algorithm for better image recovery. The simulation results show that the proposed method can improve image quality compared to IFR-Net.
