Abstract:Target Audience: Scientists and clinicians interested in highly accelerated MRI Purpose: Long acquisition times make images vulnerable to motion artifacts and limit clinical adoption. Undersampling of k-space is a widely adopted approach for fast imaging. Several reconstruction methods were proposed to recover the undersampled dataset, in which the combination of Partial Parallel Imaging (PPI) and Compressed Sensing (CS) is a promising approach to achieve low noise/artifact level . Instead of using a fixed sampling trajectory, trajectory optimization has been proposed for both PPI and CS for significantly improved reconstruction. However, the expensive computational costs of these existing methods prevent their clinical application. Here we present a clinically applicable scheme which efficiently optimizes the sampling trajectories to achieve better performance for both Parallel Imaging and Compressed Sensing. Methods: We proposed a highly efficient trajectory optimizing scheme, which uses reference data from one scan among a series of scans of a clinical exam. The steps of the algorithm are shown in Fig. 1. There are following four main features: 1) An iterative random optimization scheme (heuristic Simulated Annealing) is used; 2) A GRAPPA-operator is pre-computed based on the reference data, and the k-space extrapolation using GRAPPA-operator is pre-calculated before iterations; 3) A fast pseudo-reconstruction is implemented in each iteration; 4) L Norm of errors in image domain is used as an objective function to balance PPI and CS. Simulated Annealing is a stochastic optimization strategy widely used to search global optimal solutions using a statistical updating criterion. Pre-computation: Before iterations of optimization, a GRAPPA operator is pre-calibrated using reference k-space. Extrapolation using a 1-by-5 GRAPPA-operator can help to accurately expand a sampled k-space line and recover the two adjacent k-space lines. The extrapolation is independent of the entire sampling trajectory. Pseudo-Reconstruction: Instead of implementing the nonlinear reconstruction algorithm repeatedly, any un-acquired k-space lines can be recovered by GRAPPA-operator if there is any sampled line close-by. Given a trajectory, a fast pseudo-reconstruction can be easily generated using the pre-computed GRAPPA-operator left/right extrapolation results. At locations without adjacent acquired line, zero-filling is used. The Pseudo-reconstruction results in up-to 3 times lower reduction factor without true parallel imaging calculated in iterations. Objective Function: We used the magnitude peaks of the errors between pseudo-reconstructed image and reference image as the objective function, which is quantified using Lp Norm (ex. p=∞). Since CS is good at suppressing noise-like artifacts, minimizing Lp Norm of recovery errors can balance performance of both PPI and CS. Results: Here we implemented the trajectory optimization scheme using T1w and T2w brain datasets. The trajectory optimized from T1w data was applied to both T1w and T2w data to demonstrate the applicability of the optimized trajectory for scans with different contrasts. These two datasets (FFE sequence/230×230 mm/256×256 matrix) were acquired on a Philips 3T system (Philips Healthcare, Best, the Netherland) with an 8-channel head coil (Invivo Corporation, Gainesville, FL). The datasets were all fully sampled and retrospectively undersampled for testing. 1-D undersampling trajectory along PE direction (left-right) was optimized on T1w data using the proposed scheme with reduction factor of 5. Fig. 2 shows the change of objective function (energy in stimulated annealing) in optimization iterations and the correlation between objective function values and RMSE of reconstruction using L1-SPIRiT, CS-SENSE and L1-SPIRiT with spatially adaptive regularization . Fig. 3 shows the comparison of the reconstructed images of T1w and T2w datasets using L1-SPIRiT (20 iterations) with spatially adaptive regularization, with the proposed optimization and without (using a conventional point-spread-function based 1D variable density random trajectory instead). Optimizing the 1D random undersampling trajectory takes only 30 seconds with Matlab on a laptop with 2.4GHz Duo CPU without intended acceleration for code.

Multi PILOT: Learned Feasible Multiple Acquisition Trajectories for Dynamic MRI

PILOT: Physics-Informed Learned Optimized Trajectories for Accelerated MRI

TEAM PILOT -- Learned Feasible Extendable Set of Dynamic MRI Acquisition Trajectories

Deep Learning for Under-Sampled Non-Cartesian ASL MRI Reconstruction

Jointly Learning Non-Cartesian k-Space Trajectories and Reconstruction Networks for 2D and 3D MR Imaging through Projection

Undersampled Dynamic MRI Reconstruction by Double Sparse Spatiotemporal Dictionary.

LSST: Learned Single-Shot Trajectory and Reconstruction Network for MR Imaging

Deep Cardiac MRI Reconstruction with ADMM

Deep End-to-end Adaptive k-Space Sampling, Reconstruction, and Registration for Dynamic MRI

Learning Optimal K-space Acquisition and Reconstruction Using Physics-Informed Neural Networks

Learning to sample in Cartesian MRI

Deep learning within a priori temporal feature spaces for large-scale dynamic MR image reconstruction: Application to 5-D cardiac MR Multitasking

3D Cartesian MRI with compressed sensing and variable view sharing using complementary poisson‐disc sampling

Real Time Dynamic Magnetic Resonance Imaging Via Dictionary Learning and Combined Fourier Transform.

High-Fidelity Accelerated MRI Reconstruction by Scan-Specific Fine-Tuning of Physics-Based Neural Networks

Dynamic Magnetic Resonance Imaging Reconstruction Using Parallel Temporal Gradient Filtering

Deep Network Cascade for Dynamic Cardiac MRI Reconstruction With Motion Feature Incorporation and the Fourier Neural Attention

Joint Optimization of k-t Sampling Pattern and Reconstruction of DCE MRI for Pharmacokinetic Parameter Estimation

An efficient scheme of trajectory optimization for both Parallel Imaging and Compressed Sensing

Holistic Multi-Slice Framework for Dynamic Simultaneous Multi-Slice MRI Reconstruction

End-to-end Adaptive Dynamic Subsampling and Reconstruction for Cardiac MRI