Studies On The Reduction Of Motion Artifacts Of MRI Images

Magnetic resonance imaging(MRI)is susceptible to image artifacts caused by voluntary and involuntary motion of the patient during MRI scan.Motion artifacts will degrade the quality of the MRI images and even make them non-diagnostic.Alleviate the motion artifacts in MRI images can promote the efficiency of MRI systems and help to extend the clinical use of MRI.This paper will explore different algorithms to reduce motion artifacts in MRI.Firstly,inspired by the pseudo-random sampling in the compressed sensing,we suggested that for better removing of motion artifacts,we need to sample the K space data in a pseudo-random order.Then we tried to use genetic algorithm to pick out K space lines that are least influenced by motion,and used these lines to reconstruct the image with compressed sensing.To evaluate the image quality during the optimization process,a new reference-less image quality assessment criteria is proposed.It was proved by experiments that the proposed index has a linear correlation with the full-reference image quality assessment indexes.Since the genetic algorithm is very time-consuming,we proposed a new scheme that can be used readily in a clinical environment.K space lines were analyzed in the same order as they are pseudo-randomly sampled to find the moment when the motion occurred.The the K space data sampled before this point were used to reconstruct the image.Finally,we tried to use Convolutional Neural Network(CNN)to remove motion artifacts in MR images.Simulated images with motion artifacts are generated from artifact-free images,and images with and without motion artifacts were used in pairs to train CNN model.The trained model can then be used to obtain artifacts-free images from images with artifacts.The CNN used in this paper is based on U-net.Experimental results demonstrated that CNN can be used to reduce motion artifacts and improve image qualities.