Fast magnetic resonance fingerprinting

Abstract

Magnetic resonance fingerprinting (MRF) is a novel and efficient technique in quantitative magnetic resonance imaging (MRI). The efficiency of MRF relies on the short scan time and multiple parametric quantification in a single scan. On the one hand, the rapid acquisition of MRF is enabled by fast sequence such as balanced steady state free-precession (bSSFP) and fast imaging with steady state precession (FISP), and efficient readout trajectory such as spiral. On the other hand, quantification of multiple parameters is based on the recognition of distinct signal evolutions of different tissues, which is generated with variable scan parameters such as flip angle and repetition time (TR). The entire MRF framework includes data acquisition, image reconstruction and pattern recognition. The principles and technical details of MRF design including scan parameters, readout trajectories, sequence choices and matching methods are thus introduced with our in-house MRF framework. To further accelerate MRF, several methods are proposed in this thesis with different emphases.

First, a parallel imaging based simultaneous multi-slice (SMS) MRF method namely temporal slicing and sharing for simultaneous multi-slice magnetic resonance fingerprinting (tsSMS-MRF) is proposed. Slice unaliasing of SMS-MRF is often ill-conditioned due to the lack of slice-wise coil sensitivity variation, so in tsSMS-MRF, slice dependent image shift is included to provide better performance of slice unaliasing. Such shift is enabled with temporal slicing and sharing to transfer temporal phase variation to k-space phase pattern. After dictionary matching, better quantitative maps can be obtained with this improved slice unaliasing method.

Second, a sensitivity-free SMS/3D MRF method is proposed, using spiral in-out readout with additional $k_z$ encoding. An extra $G_z$ blip is added in between of spiral-in and spiral-out readout to provide an additional $k_z$ encoding step, leading to an optimized sampling pattern in stack-of-spiral SMS/3D acquisition. This sampling pattern causes more sparsely distributed frequency components of different slices and slice aliasing can be resolved in dictionary matching with such a sparsity constraint. With the theory of three-dimensional Fourier encoding of simultaneously acquired slices, this sampling pattern can be used in both SMS and 3D MRF. A modified sliding window reconstruction is used to further reduce the frequency interference.

And last, an alternating direction method of multipliers (ADMM) based SMS-MRF method is proposed which utilizes both parallel imaging and sparsity constraints. The optimization problem of SMS-MRF is split to subproblems of data consistency validation and dictionary matching which are iteratively solved, and parametric maps can be obtained until convergence. Several compression strategies are used in this reconstruction method to reduce the computational burden. Artifacts and noises are significantly reduced according to the results.

All these methods are demonstrated by in-vivo experiments with related acceleration factors. Specifically, tsSMS-MRF provides better matching results with robustness of coil sensitivity compared to previous sensitivity-encoding based method. Spiral in-out SMS/3D MRF improves the sampling strategy of sensitivity-free SMS/3D MRF methods for higher acceleration factor. ADMM-SMS-MRF provides matching with higher acceleration factor and reduced artifacts iteratively. As a conclusion, MRF is proved to be effectively accelerated with these proposed methods.