Reducing clustering of readouts in non-Cartesian cine magnetic resonance imaging.

BACKGROUND: Non-Cartesian magnetic resonance imaging trajectories at golden angle increments have the advantage of allowing motion correction and gating using intermediate real-time reconstructions. However, when the acquired data are cardiac binned for cine imaging, trajectories can cluster together at certain heart rates (HR) causing image artifacts. Here, we demonstrate an approach to reduce clustering by inserting additional angular increments within the trajectory, and optimizing them while still allowing for intermediate reconstructions. METHODS: Three acquisition models were simulated under constant and variable HR: golden angle (Mtrd), random additional angles (Mrnd), and optimized additional angles (Mopt). The standard deviations of trajectory angular differences (STAD) were compared through their interquartile ranges (IQR) and the Kolmogorov-Smirnov test (significance level: p = 0.05). Agreement between an image reconstructed with uniform sampling and images from Mtrd, Mrnd, and Mopt was analyzed using the structural similarity index measure (SSIM). Mtrd and Mopt were compared in three adults at high, low, and no HR variability. RESULTS: STADs from Mtrd were significantly different (p < 0.05) from Mopt and Mrnd. STAD (IQR × 10-2 rad) showed that Mopt (0.5) and Mrnd (0.5) reduced clustering relative to Mtrd (1.9) at constant HR. For variable HR, Mopt (0.5) and Mrnd (0.5) outperformed Mtrd (0.9). The SSIM (IQR) showed that Mopt (0.011) produced the best image quality, followed by Mrnd (0.014), and Mtrd (0.030). Mopt outperformed Mtrd at reduced HR variability in in-vivo studies. At high HR variability, both models performed well. CONCLUSION: This approach reduces clustering in k-space and improves image quality.

Magnetization-prepared shells trajectory with automated gradient waveform design.

PURPOSE: To develop a fully automated trajectory and gradient waveform design for the non-Cartesian shells acquisition, and to develop a magnetization-prepared (MP) shells acquisition to achieve an efficient three-dimensional acquisition with improved gray-to-white brain matter contrast. METHODS: After reviewing the shells k-space trajectory, a novel, fully automated trajectory design is developed that allows for gradient waveforms to be automatically generated for specified acquisition parameters. Designs for two types of shells are introduced, including fully sampled and undersampled/accelerated shells. Using those designs, an MP-Shells acquisition is developed by adjusting the acquisition order of shells interleaves to synchronize the center of k-space sampling with the peak of desired gray-to-white matter contrast. The feasibility of the proposed design and MP-Shells is demonstrated using simulation, phantom, and volunteer subject experiments, and the performance of MP-Shells is compared with a clinical Cartesian magnetization-prepared rapid gradient echo acquisition. RESULTS: Initial experiments show that MP-Shells produces excellent image quality with higher data acquisition efficiency and improved gray-to-white matter contrast-to-noise ratio (by 36%) compared with the conventional Cartesian magnetization-prepared rapid gradient echo acquisition. CONCLUSION: We demonstrated the feasibility of a three-dimensional MP-Shells acquisition and an automated trajectory design to achieve an efficient acquisition with improved gray-to-white matter contrast. Magn Reson Med 79:2024-2035, 2018. © 2017 International Society for Magnetic Resonance in Medicine.

Simultaneous auto-calibration and gradient delays estimation (SAGE) in non-Cartesian parallel MRI using low-rank constraints.

PURPOSE: To correct gradient timing delays in non-Cartesian MRI while simultaneously recovering corruption-free auto-calibration data for parallel imaging, without additional calibration scans. METHODS: The calibration matrix constructed from multi-channel k-space data should be inherently low-rank. This property is used to construct reconstruction kernels or sensitivity maps. Delays between the gradient hardware across different axes and RF receive chain, which are relatively benign in Cartesian MRI (excluding EPI), lead to trajectory deviations and hence data inconsistencies for non-Cartesian trajectories. These in turn lead to higher rank and corrupted calibration information which hampers the reconstruction. Here, a method named Simultaneous Auto-calibration and Gradient delays Estimation (SAGE) is proposed that estimates the actual k-space trajectory while simultaneously recovering the uncorrupted auto-calibration data. This is done by estimating the gradient delays that result in the lowest rank of the calibration matrix. The Gauss-Newton method is used to solve the non-linear problem. The method is validated in simulations using center-out radial, projection reconstruction and spiral trajectories. Feasibility is demonstrated on phantom and in vivo scans with center-out radial and projection reconstruction trajectories. RESULTS: SAGE is able to estimate gradient timing delays with high accuracy at a signal to noise ratio level as low as 5. The method is able to effectively remove artifacts resulting from gradient timing delays and restore image quality in center-out radial, projection reconstruction, and spiral trajectories. CONCLUSION: The low-rank based method introduced simultaneously estimates gradient timing delays and provides accurate auto-calibration data for improved image quality, without any additional calibration scans.

Dual-domain self-supervised learning for accelerated non-Cartesian MRI reconstruction.

While enabling accelerated acquisition and improved reconstruction accuracy, current deep MRI reconstruction networks are typically supervised, require fully sampled data, and are limited to Cartesian sampling patterns. These factors limit their practical adoption as fully-sampled MRI is prohibitively time-consuming to acquire clinically. Further, non-Cartesian sampling patterns are particularly desirable as they are more amenable to acceleration and show improved motion robustness. To this end, we present a fully self-supervised approach for accelerated non-Cartesian MRI reconstruction which leverages self-supervision in both k-space and image domains. In training, the undersampled data are split into disjoint k-space domain partitions. For the k-space self-supervision, we train a network to reconstruct the input undersampled data from both the disjoint partitions and from itself. For the image-level self-supervision, we enforce appearance consistency obtained from the original undersampled data and the two partitions. Experimental results on our simulated multi-coil non-Cartesian MRI dataset demonstrate that DDSS can generate high-quality reconstruction that approaches the accuracy of the fully supervised reconstruction, outperforming previous baseline methods. Finally, DDSS is shown to scale to highly challenging real-world clinical MRI reconstruction acquired on a portable low-field (0.064 T) MRI scanner with no data available for supervised training while demonstrating improved image quality as compared to traditional reconstruction, as determined by a radiologist study.

Advances in spiral fMRI: A high-resolution dataset.

We present data collected for the research article "Advances in Spiral fMRI: A High-resolution Study with Single-shot Acquisition" (Kasper et al. 2022). All data was acquired on a 7T ultra-high field MR system (Philips Achieva), equipped with a concurrent magnetic field monitoring setup based on 16 NMR probes. For task-based fMRI, a visual quarterfield stimulation paradigm was employed, alongside physiological monitoring via peripheral recordings. This data collection contains different datasets pertaining to different purposes: (1) Measured magnetic field dynamics (k0, spiral k-space trajectories, 2nd order spherical harmonics, concomitant fields) during ultra-high field fMRI sessions from six subjects, as well as concurrent temperature curves of the gradient coil, to explore MR system and subject-induced variability of field fluctuations and assess the impact of potential correction methods. (2) MR Raw Data, i.e., coil and concurrent encoding magnetic field (trajectory) data, of a single subject, as well as nominal spiral gradient waveforms, precomputed B0 and coil sensitivity maps, to enable testing of alternative image reconstruction approaches for spiral fMRI data. (3) Reconstructed image time series of the same subject alongside behavioral and physiological logfiles, to reproduce the fMRI preprocessing and analysis, as well as figures presented in the research article related to this article, using the published analysis code repository. All data is provided in standardized formats for the respective research area. In particular, ISMRMRD (HDF5) is used to store raw coil data and spiral trajectories, as well as measured field dynamics. Likewise, the NIfTI format is used for all imaging data (anatomical MRI and spiral fMRI, B0 and coil sensitivity maps).

Optimizing trajectory ordering for fast radial ultra-short TE (UTE) acquisitions.

PURPOSE: Additional spoiler gradients are required in 3D UTE sequences with random view ordering to suppress magnetization refocusing. By leveraging the encoding gradient induced spoiling effect, the spoiler gradients could potentially be reduced or removed to shorten the TR and increase encoding efficiency. An analysis framework is built that models the gradient spoiling effects and a new ordering scheme is proposed for fast 3D UTE acquisition. THEORY AND METHODS: UTE signal evolution and spatial encoding gradient induced spoiling effect are derived from the Bloch equations. And the concept is validated in 2D radial UTE simulation. Then an optimized ordering scheme, named reordered 2D golden angle (r2DGA) scheme, for 3D UTE acquisition is proposed. The r2DGA scheme is compared to the sequential and 3D golden angle schemes in both phantom and volunteer studies. RESULTS: The proposed r2DGA ordering scheme was applied to two applications, single breath-holding and free breathing 3D lung MRI. With r2DGA ordering scheme, breath-holding lung MRI scan increased 60% scan efficiency by removing the spoiler gradients and the free breathing scan reduced 20% scan time compared to the 3D golden angle scheme by reducing the spoiler gradients. CONCLUSIONS: The proposed r2DGA ordering scheme UTE acquisition reduces the need of spoiler gradients and increases the encoding efficiency, and shows improvements in both breath-holding and free breathing lung MRI applications.

Real-Time Mapping of Tissue Properties for Magnetic Resonance Fingerprinting.

Magnetic resonance fingerprinting (MRF) is a relatively new multi-parametric quantitative imaging method that involves a two-step process: (i) reconstructing a series of time frames from highly-undersampled non-Cartesian spiral k-space data and (ii) pattern matching using the time frames to infer tissue properties (e.g., T1 and T2 relaxation times). In this paper, we introduce a novel end-to-end deep learning framework to seamlessly map the tissue properties directly from spiral k-space MRF data, thereby avoiding time-consuming processing such as the non-uniform fast Fourier transform (NUFFT) and the dictionary-based fingerprint matching. Our method directly consumes the non-Cartesian k-space data, performs adaptive density compensation, and predicts multiple tissue property maps in one forward pass. Experiments on both 2D and 3D MRF data demonstrate that quantification accuracy comparable to state-of-the-art methods can be accomplished within 0.5 s, which is 1,100 to 7,700 times faster than the original MRF framework. The proposed method is thus promising for facilitating the adoption of MRF in clinical settings.

SSFP fMRI at 3 tesla: Efficiency of polar acquisition-reconstruction technique.

SSFP-based fMRI techniques, known for their high specificity and low geometrical distortion, look promising for high-resolution brain mapping. Nevertheless, they suffer from lack of speed and sensitivity, leading them to be exploited mostly in high-field scanners. Radial acquisition can help with these inefficiencies through better tSNR and more effective coverage of the spatial frequencies. Here, we present a SSFP-fMRI approach and experimentally investigate it at 3 T scanners using radial readout for acquisition. In particular, the visual activity is mapped through three bSSFP techniques: 1- Cartesian, 2- Radial with re-gridding reconstruction, 3- Radial with Polar Fourier Transform (PFT) reconstruction. In the PFT technique streaking artifacts, generated at high acceleration rates by re-gridding reconstruction, are avoided and pixel size in the final framework is retrospectively selectable. General agreement, but better tSNR of Radial reading, was first confirmed for these techniques in detection of neural activities at 2 × 2 mm2 in-plane resolution for all 28 subjects,. Next the outcome of the PFT algorithm with 1 × 1 mm2 pixel size was compared to images reconstructed by re-gridding (from the same raw data) with the identical pixel size through interpolation. The localization of the activity showed improvement in PFT over interpolation both qualitatively (i.e., well-fitting in gray-matter) and quantitatively (i.e., higher z-scores and tSNR). The proposed technique can therefore be considered as a remedy for lack of speed and sensitivity in SSFP-based fMRI, in conventional field strengths. The proposed approach is particularly useful in task-based studies when we concentrate on a ROI considerably smaller than FOV, without sacrificing spatial resolution.

Mean square optimal NUFFT approximation for efficient non-Cartesian MRI reconstruction.

The fast evaluation of the discrete Fourier transform of an image at non-uniform sampling locations is key to efficient iterative non-Cartesian MRI reconstruction algorithms. Current non-uniform fast Fourier transform (NUFFT) approximations rely on the interpolation of oversampled uniform Fourier samples. The main challenge is high memory demand due to oversampling, especially when multidimensional datasets are involved. The main focus of this work is to design an NUFFT algorithm with minimal memory demands. Specifically, we introduce an analytical expression for the expected mean square error in the NUFFT approximation based on our earlier work. We then introduce an iterative algorithm to design the interpolator and scale factors. Experimental comparisons show that the proposed optimized NUFFT scheme provides considerably lower approximation errors than the previous designs [1] that rely on worst case error metrics. The improved approximations are also seen to considerably reduce the errors and artifacts in non-Cartesian MRI reconstruction.

Structured errors in reconstruction methods for Non-Cartesian MR data.

BACKGROUND: Reconstruction methods for Non-Cartesian magnetic resonance imaging have often been analyzed using the root mean square error (RMSE). However, RMSE is not able to measure the level of structured error associated with the reconstruction process. METHODS: An index for geometric information loss was presented using the 2D autocorrelation function. The performances of Least Squares Non Uniform Fast Fourier Transform (LS-NUFFT) and gridding reconstruction (GR) methods were compared. The Direct Summation method (DS) was used as reference. For both methods, RMSE and the loss in geometric information were calculated using a digital phantom and a hyperpolarized (13)C dataset. RESULTS: The performance of the geometric information loss index was analyzed in the presence of noise. Comparing to GR, LS-NUFFT obtained a lower RMSE, but its error image appeared more structured. This was observed in both phantom and in vivo experiments. DISCUSSION: The evaluation of geometric information loss together with the reconstruction error was important for an appropriate performance analysis of the reconstruction methods. The use of geometric information loss was helpful to determine that LS-NUFFT loses relevant information in the reconstruction process, despite the low RMSE.