Online Dynamic Magnetic Resonance Imaging Based On Radial Sampling

Magnetic Resonance Imaging(MRI)has been widely used in clinical due to its non-radiation,high resolution,multi-orientation,multi-parameter,etc..However MRI,which takes a long time to scan the data,usually can not meet the needs of real-time online imaging in cardiac dynamic MRI(DMRI),which also restricts its development.So how to improve the speed of MRI,especially the DMRI,has became a hot issue in the field of MRI.Dynamic MRI has two imaging methods.one is offline reconstruction,another is online reconstruction.Offline reconstruction is often used in clinical application,but the online imaging is still in its infancy.Real-time online dynamic MRI requires high imaging speed,which is mainly affected by two aspects.one is speed of data acquisition,anther is reconstruction speed.It has been proposed that the current real-time online imaging using Cartesian sampling and frame predicting,could not completely meet the clinical needs of real-time online imaging,its imaging speed is still need to be improved.Compared to Cartesian sampling,non-Cartesian has many advantages,such as fast imaging,sensitive to the motion of object,etc..However,a major factor that restricts of Non-Cartesian sampling applying used in the clinical practice is the low speed of imaging reconstruction.Because the sampled data do not fall into the Cartesian grid points,which need gridding algorithm to uniform the data.So Fast Fourier Transform algorithm can not be used directly.This process is time consuming.So we can use Graphics Processing Units(GPU)Parallel computing to speed up the gridding algorithm.In this paper,we focus on dynamic MRI,combined with Radial sampling method to acquire data.This paper is based on improving the speed of gridding reconstruction with GPU acceleration,in order to achieve the goal of real-time online imaging.The primary work and innovation of this paper are as follows:1)Study the principle and implementation of online dynamic MRI.In order to acquire sparser data,Motion Estimation/Motion Compensation(ME/ME)method was proposed to predict current frame from previous reconstructed frames.An Overlapped Block Motion Compensation(OBMC)algorithm was used to eliminate motion artifacts.Compared with Motion predict online dynamic(MPOD)algorithm,ME/MC algorithm achieved the more better result.2)We use Radial sampling trajectory,one of Non-Cartesian sampling,which can reduce the time of data acquisition,which used two gradient in the direction of frequency encodingand phase encoding at the same time.But the sampled data can not be reconstructed directly and need to be uniformed.We used the gridding algorithm to interpolate the data into the uniform Cartesian points.The result shows that the speed of data acquisition and imaging reconstruction quality of Radial sampling are higher than that of Cartesian sampling.3)Radial sampling trajectory is non-uniform,need to be uniformed.We use the advantages of GPU parallel computing to improve the efficiency of the gridding algorithm and achieved good results.The results shows that GPU reconstruction speed is 12 times faster than that with CPU.We proposed feasible real-time online dynamic MRI method from MRI theory to Non-Cartesian sampling and reconstruction,which has played an active role to improve the real-time online imaging for non-Cartesian sampling in dynamic MRI.