Four Dimentional Magnetic Resonance Image Reconstruction With Motion Correction

Compared with other medical imaging methods,four-dimensional magnetic resonance imaging,namely 4D MRI,has many advantages,such as no ionizing radiation and high soft tissue contrast.It can provide high-dimensional diagnostic information for clinical medicine,and has broad application prospects.But the relatively long scanning time limits its practical application.In addition,in the longterm scanning process,patients' autonomous or involuntary physiological movement will introduce artifacts in the reconstructed image,which will reduce the image quality and affect the diagnostic reliability.Therefore,speeding up 4D MRI and reducing the influence of motion on image is the key to broaden the application of4 D MRI.To reconstruct high-quality and high-resolution images under a high undersampling factor,this paper focuses on the four-dimensional magnetic resonance image reconstruction method combined with motion correctionFirstly,to accelerate imaging process,a four-dimensional magnetic resonance image reconstruction method based on prior constraint is studied.A four-dimensional magnetic resonance image reconstruction model based on higher degree total variation and local low rank constraint is proposed,and the model is solved by alternating iterative optimization algorithm based on maximization-minimization strategy.Experimental results show that the proposed method can reconstruct highquality image under a high undersampling factor,and the texture details are clear.Compared with a state-of-art method,the low rank constraint method and the combined total variation and low rank constraint method,it has better performance.Then,to solve the motion problem introduced in the imaging,the motion correction-oriented 3D MR image registration method is studied,including the diffeomorphic Log-Demons algorithm and an unsupervised learning method based on U-Net.The experimental results show that the two methods can achieve accurate registration.At the same time,it is found that the computing time of the diffeomorphic Log-Demons algorithm is relatively long,and the image registration results are greatly affected by the parameter settings,which will lead to different degrees of geometric distortion in the reconstructed image.The unsupervised learning method based on U-Net can achieve fast image registration,and the performance of the method is superior to the diffeomorphic Log-Demons algorithm based on Dice coefficient.Finally,to suppress the motion artifacts in the image,a four-dimensional magnetic resonance image reconstruction method based on motion estimation is studied.It includes the method of alternating image reconstruction and motion estimation in the form of multi-resolution,and the method of embedding motion estimation in image reconstruction model is proposed.In the simulation experiment,the joint reconstruction and motion estimation method uses the diffeomorphic LogDemons algorithm and the unsupervised learning method based on U-Net to estimate the motion field.Experimental results show that the images reconstructed by the proposed joint reconstruction and motion estimation method has higher quality than that reconstructed by the weighted spatiotemporal total variation reconstruction method.