Font Size: a A A

Fast High-Resolution Magnetic Resonance Imaging

Posted on:2020-02-23Degree:DoctorType:Dissertation
Country:ChinaCandidate:S JiaFull Text:PDF
GTID:1364330599977514Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
Magnetic Resonance Imaging(MRI)can provide substantial and excellent soft-tissue contrast.Thus,it is desired to develop advanced MR imaging techniques such as real-time cardiac cine imaging(RT-cine)and dark-blood vessel wall imaging(VWI).RT-cine can track cardiac motion in a heart-beat to assess cardiac function using high temporal resolution imaging.VWI can examine artery wall-related lesions such as atherosclerosis plaque directly using high spatial resolution imaging.Other imaging modalities may be infeasible or inefficient to achieve these clinical demands.However,the major bottleneck of MRI is its low scan speed which would become more severe in high-resolution imaging scenarios,since Fourier encoding based MRI requires amount of phase encoding steps for spatial localization.For RT-cine,the low scan speed will lead to an insufficient temporal resolution to resolve cardiac motion in real-time.For high-resolution VWI,the slow scan speed will lead to a long scan time which may be clinically unacceptable and make VWI more susceptibility to motion.Fast MRI techniques are critically necessary to achieve high-resolution imaging.The strategy of undersampling is one of the most important ways to accelerate MR imaging.Fast MRI techniques based on undersampling improve scan speed by reducing phase encoding steps.The unacquired data will be recovered by image reconstruction algorithms utilizing redundancy in MR image and raw data.Parallel Imaging(PI)and Compressed Sensing(CS)are two major approaches for fast MRI.The spatial encoding capability from coil sensitivity maps of phased-array coil is utilized in PI to reduce data sampling.Image reconstruction is accomplished by inversing the sensitivity encoding matrix.High acceleration will make the sensitivity encoding matrix to be ill-conditioned and heavily amplify noise in the reconstructed image.Image sparsity is employed in CS to construct a nonlinear optimization problem that will be solved iteratively for image reconstruction.At high acceleration rates,CS requires a stronger sparsity penalty to suppress noise and aliasing artifacts.Low-contrast image details mixed by noise may be blurred by this nonlinear filtering process.The choice and design of fast MRI techniques should be made and optimized according to the signal characteristics,available image redundancy and clinical requirements of the target applications.This thesis focuses on achieving high temporal-resolution RT-cine and high spatial-resolution VWI by overcoming the limitations of current approaches.The major contributions of this thesis include:1.RT cardiac cine requires high temporal resolution and low-latency reconstruction.This thesis proposes a new non-iterative PI algorithm named NL-VCC-TGRAPPA.The inversing condition of PI reconstruction is improved by aggregating the spatial encoding capability of background image phase into TGRAPPA in the manner of virtual conjugate coil(VCC).A second-order virtual coil is further constructed by nonlinear mapping the VCC-TGRAPPA relationship into a high-order feature space.Nonlinear bias induced by noise in the calibration data is suppressed.The proposed NL-VCC-TGRAPPA method can suppress noise efficiently at 8-fold acceleration which can improve the temporal resolution of RT-cine to ?45 ms/frame.Comparing with the iterative CS approach,NL-VCC-TGRAPPA achieves comparable reconstruction quality while providing a higher computational efficiency without the need for iteration.2.Joint intracranial and carotid VWI demands high spatial resolution and sharp wall depiction.This thesis improves the conventional combined CS and PI acceleration by using an equidistant sampling scheme to avoid the less sampling of the high-frequency region of k-space in conventional variable-density random sampling,which is crucial to the image sharpness.Iterative feature refinement module is further integrated into CS iteration to restore high-resolution image details which are potentially deteriorated by preceding nonlinear filtering step.The scan duration of joint intracranial and carotid VWI at an isotropic spatial resolution of 0.55 mm is reduced by 5-fold acceleration to only 5 minutes.The imaging quality of normal vessel walls on 8 healthy volunteers and the diagnosis value of wall lesions on 20 patients are both comparable to the results from 2.7-fold PI VWI scans which take approximately 9 minutes.3.The scanner integration and clinical deployment of proposed fast imaging techniques require online and efficient reconstruction with low latency.Open-sourced image reconstruction platform Gadgetron and raw data format ISMRMRD are introduced to achieve efficient implementation and online running of proposed fast imaging techniques.The virtual coil-based implementation of non-iterative NL-VCC-TGRAPA can reuse the code for conventional TGRAPPA and makes the scanner integration easy.However,the high computational burden of iterative CS VWI limits its implementation on a commercial reconstruction system.Based on the external Gadgetron workstation with customized hardware,this thesis reduced the online reconstruction time of CS VWI into a clinically acceptable 2 minutes by coil compression and parallel computing on multiple GPUs.The achieved CS VWI was deployed into clinical environments with domestic MR scanners to improve the clinical scan efficiency and robustness to patient motion of VWI.
Keywords/Search Tags:Fast Imaging, Parallel Imaging, Compressed Sensing, MR Vessel Wall Imaging, MR Real-Time Cardiac Cine Imaging
PDF Full Text Request
Related items