Research On Magnetic Resonance Image Denoising And Reconstruction Based On Regularization Method

Magnetic Resonance Imaging(MRI)technology is playing an important role in the field of healthcare due to its noninvasive advantage and the abundant information it provides.However,the measurements acquired from MRI machine are often corrupted by noise and sampling mode.Recovering the original images from their degraded measurements belongs to the inverse problem,and it could be solved properly by applying regularization method.The regularization model consists of data consistency error and priori constraints that are influential on image restoration.In this thesis,we deeply analyzes and discuss the denoising and reconstruction process of magnetic resonance images based on the regularization method.The core goal of image denoising is to achieve a better recovery effect while the computational efficiency is guaranteed.In this thesis,we study several improved methods of Total Variation(TV)and analyze their merits by some experimental contrast result first,then we propose an evaluation approach named Fast Majorization-Minimization(FMM)to enhance the calculation efficiency of Higher Degree Total Variation(HDTV)denoising method.By introducing auxiliary variable to minimize the objective function,FMM can cut computing time by about 5-7 times.Since Compressive Sensing(CS)theory applied to MRI technology,the low imaging efficiency caused by the large sampling number had been solved.With a small amount of data,the original image can be reconstructed by solving a regularization model.This thesis propose a novel method that combined TV and HDTV constraints(C-HDTV)by linear weights to realize an adaptation to different sparse features,thus reconstruct images with higher quality in the case of undersampling mode.In the Simulation experiments of two-dimensional images,the C-HDTV method achieve better result than the others,so we further extend it to 3D image reconstruction and experimental comparison proves an excellent result,too.For dynamic magnetic resonance images,CS-MRI is more practical because the volume of data is even larger but with a high redundancy.While C-HDTV constraints is applied for the promotion of spatial domain sparsity,we furthermore add the low rank constraint to expedite sparsity in time domain.Experiments shows that this time-space domain regularization leads to a better recovery of dynamic images in high under-rate sampling mode.