Revisiting [Formula: see text]-wavelet compressed-sensing MRI in the era of deep learning.

Following their success in numerous imaging and computer vision applications, deep-learning (DL) techniques have emerged as one of the most prominent strategies for accelerated MRI reconstruction. These methods have been shown to outperform conventional regularized methods based on compressed sensing (CS). However, in most comparisons, CS is implemented with two or three hand-tuned parameters, while DL methods enjoy a plethora of advanced data science tools. In this work, we revisit [Formula: see text]-wavelet CS reconstruction using these modern tools. Using ideas such as algorithm unrolling and advanced optimization methods over large databases that DL algorithms utilize, along with conventional insights from wavelet representations and CS theory, we show that [Formula: see text]-wavelet CS can be fine-tuned to a level close to DL reconstruction for accelerated MRI. The optimized [Formula: see text]-wavelet CS method uses only 128 parameters compared to >500,000 for DL, employs a convex reconstruction at inference time, and performs within <1% of a DL approach that has been used in multiple studies in terms of quantitative quality metrics.

Unsupervised deep learning with convolutional neural networks for static parallel transmit design: A retrospective study.

PURPOSE: To mitigate B 1 + $$ {B}_1^{+} $$ inhomogeneity at 7T for multi-channel transmit arrays using unsupervised deep learning with convolutional neural networks (CNNs). METHODS: Deep learning parallel transmit (pTx) pulse design has received attention, but such methods have relied on supervised training and did not use CNNs for multi-channel B 1 + $$ {B}_1^{+} $$ maps. In this work, we introduce an alternative approach that facilitates the use of CNNs with multi-channel B 1 + $$ {B}_1^{+} $$ maps while performing unsupervised training. The multi-channel B 1 + $$ {B}_1^{+} $$ maps are concatenated along the spatial dimension to enable shift-equivariant processing amenable to CNNs. Training is performed in an unsupervised manner using a physics-driven loss function that minimizes the discrepancy of the Bloch simulation with the target magnetization, which eliminates the calculation of reference transmit RF weights. The training database comprises 3824 2D sagittal, multi-channel B 1 + $$ {B}_1^{+} $$ maps of the healthy human brain from 143 subjects. B 1 + $$ {B}_1^{+} $$ data were acquired at 7T using an 8Tx/32Rx head coil. The proposed method is compared to the unregularized magnitude least-squares (MLS) solution for the target magnetization in static pTx design. RESULTS: The proposed method outperformed the unregularized MLS solution for RMS error and coefficient-of-variation and had comparable energy consumption. Additionally, the proposed method did not show local phase singularities leading to distinct holes in the resulting magnetization unlike the unregularized MLS solution. CONCLUSION: Proposed unsupervised deep learning with CNNs performs better than unregularized MLS in static pTx for speed and robustness.

Large-scale 3D non-Cartesian coronary MRI reconstruction using distributed memory-efficient physics-guided deep learning with limited training data.

OBJECT: To enable high-quality physics-guided deep learning (PG-DL) reconstruction of large-scale 3D non-Cartesian coronary MRI by overcoming challenges of hardware limitations and limited training data availability. MATERIALS AND METHODS: While PG-DL has emerged as a powerful image reconstruction method, its application to large-scale 3D non-Cartesian MRI is hindered by hardware limitations and limited availability of training data. We combine several recent advances in deep learning and MRI reconstruction to tackle the former challenge, and we further propose a 2.5D reconstruction using 2D convolutional neural networks, which treat 3D volumes as batches of 2D images to train the network with a limited amount of training data. Both 3D and 2.5D variants of the PG-DL networks were compared to conventional methods for high-resolution 3D kooshball coronary MRI. RESULTS: Proposed PG-DL reconstructions of 3D non-Cartesian coronary MRI with 3D and 2.5D processing outperformed all conventional methods both quantitatively and qualitatively in terms of image assessment by an experienced cardiologist. The 2.5D variant further improved vessel sharpness compared to 3D processing, and scored higher in terms of qualitative image quality. DISCUSSION: PG-DL reconstruction of large-scale 3D non-Cartesian MRI without compromising image size or network complexity is achieved, and the proposed 2.5D processing enables high-quality reconstruction with limited training data.

Evaluating increases in sensitivity from NORDIC for diverse fMRI acquisition strategies.

As the neuroimaging field moves towards detecting smaller effects at higher spatial resolutions, and faster sampling rates, there is increased attention given to the deleterious contribution of unstructured, thermal noise. Here, we critically evaluate the performance of a recently developed reconstruction method, termed NORDIC, for suppressing thermal noise using datasets acquired with various field strengths, voxel sizes, sampling rates, and task designs. Following minimal preprocessing, statistical activation (t-values) of NORDIC processed data was compared to the results obtained with alternative denoising methods. Additionally, we examined the consistency of the estimates of task responses at the single-voxel, single run level, using a finite impulse response (FIR) model. To examine the potential impact on effective image resolution, the overall smoothness of the data processed with different methods was estimated. Finally, to determine if NORDIC alters or removes temporal information important for modeling responses, we employed an exhaustive leave-p-out cross validation approach, using FIR task responses to predict held out timeseries, quantified using R2. After NORDIC, the t-values are increased, an improvement comparable to what could be achieved by 1.5 voxels smoothing, and task events are clearly visible and have less cross-run error. These advantages are achieved with smoothness estimates increasing by less than 4%, while 1.5 voxel smoothing is associated with increases of over 140%. Cross-validated R2s based on the FIR models show that NORDIC is not measurably distorting the temporal structure of the data under this approach and is the best predictor of non-denoised time courses. The results demonstrate that analyzing 1 run of data after NORDIC produces results equivalent to using 2 to 3 original runs and that NORDIC performs equally well across a diverse array of functional imaging protocols. Significance Statement: For functional neuroimaging, the increasing availability of higher field strengths and ever higher spatiotemporal resolutions has led to concomitant increase in concerns about the deleterious effects of thermal noise. Historically this noise source was suppressed using methods that reduce spatial precision such as image blurring or averaging over a large number of trials or sessions, which necessitates large data collection efforts. Here, we critically evaluate the performance of a recently developed reconstruction method, termed NORDIC, which suppresses thermal noise. Across datasets varying in field strength, voxel sizes, sampling rates, and task designs, NORDIC produces substantial gains in data quality. Both conventional t-statistics derived from general linear models and coefficients of determination for predicting unseen data are improved. These gains match or even exceed those associated with 1 voxel Full Width Half Max image smoothing, however, even such small amounts of smoothing are associated with a 52% reduction in estimates of spatial precision, whereas the measurable difference in spatial precision is less than 4% following NORDIC.

Signal intensity informed multi-coil encoding operator for physics-guided deep learning reconstruction of highly accelerated myocardial perfusion CMR.

PURPOSE: To develop a physics-guided deep learning (PG-DL) reconstruction strategy based on a signal intensity informed multi-coil (SIIM) encoding operator for highly-accelerated simultaneous multislice (SMS) myocardial perfusion cardiac MRI (CMR). METHODS: First-pass perfusion CMR acquires highly-accelerated images with dynamically varying signal intensity/SNR following the administration of a gadolinium-based contrast agent. Thus, using PG-DL reconstruction with a conventional multi-coil encoding operator leads to analogous signal intensity variations across different time-frames at the network output, creating difficulties in generalization for varying SNR levels. We propose to use a SIIM encoding operator to capture the signal intensity/SNR variations across time-frames in a reformulated encoding operator. This leads to a more uniform/flat contrast at the output of the PG-DL network, facilitating generalizability across time-frames. PG-DL reconstruction with the proposed SIIM encoding operator is compared to PG-DL with conventional encoding operator, split slice-GRAPPA, locally low-rank (LLR) regularized reconstruction, low-rank plus sparse (L + S) reconstruction, and regularized ROCK-SPIRiT. RESULTS: Results on highly accelerated free-breathing first pass myocardial perfusion CMR at three-fold SMS and four-fold in-plane acceleration show that the proposed method improves upon the reconstruction methods use for comparison. Substantial noise reduction is achieved compared to split slice-GRAPPA, and aliasing artifacts reduction compared to LLR regularized reconstruction, L + S reconstruction and PG-DL with conventional encoding. Furthermore, a qualitative reader study indicated that proposed method outperformed all methods. CONCLUSION: PG-DL reconstruction with the proposed SIIM encoding operator improves generalization across different time-frames /SNRs in highly accelerated perfusion CMR.

Robust cardiac T 1 ρ $$ {\mathrm{T}}_{1_{\boldsymbol{\rho}}} $$ mapping at 3T using adiabatic spin-lock preparations.

PURPOSE: The aim of this study is to develop and optimize an adiabatic T 1 ρ $$ {\mathrm{T}}_{1\uprho} $$ ( T 1 ρ , adiab $$ {\mathrm{T}}_{1\uprho, \mathrm{adiab}} $$ ) mapping method for robust quantification of spin-lock (SL) relaxation in the myocardium at 3T. METHODS: Adiabatic SL (aSL) preparations were optimized for resilience against B 0 $$ {\mathrm{B}}_0 $$ and B 1 + $$ {\mathrm{B}}_1^{+} $$ inhomogeneities using Bloch simulations. Optimized B 0 $$ {\mathrm{B}}_0 $$ -aSL, Bal-aSL and B 1 $$ {\mathrm{B}}_1 $$ -aSL modules, each compensating for different inhomogeneities, were first validated in phantom and human calf. Myocardial T 1 ρ $$ {\mathrm{T}}_{1\uprho} $$ mapping was performed using a single breath-hold cardiac-triggered bSSFP-based sequence. Then, optimized T 1 ρ , adiab $$ {\mathrm{T}}_{1\uprho, \mathrm{adiab}} $$ preparations were compared to each other and to conventional SL-prepared T 1 ρ $$ {\mathrm{T}}_{1\uprho} $$ maps (RefSL) in phantoms to assess repeatability, and in 13 healthy subjects to investigate image quality, precision, reproducibility and intersubject variability. Finally, aSL and RefSL sequences were tested on six patients with known or suspected cardiovascular disease and compared with LGE, T 1 $$ {\mathrm{T}}_1 $$ , and ECV mapping. RESULTS: The highest T 1 ρ , adiab $$ {\mathrm{T}}_{1\uprho, \mathrm{adiab}} $$ preparation efficiency was obtained in simulations for modules comprising 2 HS pulses of 30 ms each. In vivo T 1 ρ , adiab $$ {\mathrm{T}}_{1\uprho, \mathrm{adiab}} $$ maps yielded significantly higher quality than RefSL maps. Average myocardial T 1 ρ , adiab $$ {\mathrm{T}}_{1\uprho, \mathrm{adiab}} $$ values were 183.28 ± $$ \pm $$ 25.53 ms, compared with 38.21 ± $$ \pm $$ 14.37 ms RefSL-prepared T 1 ρ $$ {\mathrm{T}}_{1\uprho} $$ . T 1 ρ , adiab $$ {\mathrm{T}}_{1\uprho, \mathrm{adiab}} $$ maps showed a significant improvement in precision (avg. 14.47 ± $$ \pm $$ 3.71% aSL, 37.61 ± $$ \pm $$ 19.42% RefSL, p < 0.01) and reproducibility (avg. 4.64 ± $$ \pm $$ 2.18% aSL, 47.39 ± $$ \pm $$ 12.06% RefSL, p < 0.0001), with decreased inter-subject variability (avg. 8.76 ± $$ \pm $$ 3.65% aSL, 51.90 ± $$ \pm $$ 15.27% RefSL, p < 0.0001). Among aSL preparations, B 0 $$ {\mathrm{B}}_0 $$ -aSL achieved the better inter-subject variability. In patients, B 1 $$ {\mathrm{B}}_1 $$ -aSL preparations showed the best artifact resilience among the adiabatic preparations. T 1 ρ , adiab $$ {\mathrm{T}}_{1\uprho, \mathrm{adiab}} $$ times show focal alteration colocalized with areas of hyper-enhancement in the LGE images. CONCLUSION: Adiabatic preparations enable robust in vivo quantification of myocardial SL relaxation times at 3T.

Physics-Driven Deep Learning for Computational Magnetic Resonance Imaging: Combining physics and machine learning for improved medical imaging.

Physics-driven deep learning methods have emerged as a powerful tool for computational magnetic resonance imaging (MRI) problems, pushing reconstruction performance to new limits. This article provides an overview of the recent developments in incorporating physics information into learning-based MRI reconstruction. We consider inverse problems with both linear and non-linear forward models for computational MRI, and review the classical approaches for solving these. We then focus on physics-driven deep learning approaches, covering physics-driven loss functions, plug-and-play methods, generative models, and unrolled networks. We highlight domain-specific challenges such as real- and complex-valued building blocks of neural networks, and translational applications in MRI with linear and non-linear forward models. Finally, we discuss common issues and open challenges, and draw connections to the importance of physics-driven learning when combined with other downstream tasks in the medical imaging pipeline.

Residual RAKI: A hybrid linear and non-linear approach for scan-specific k-space deep learning.

Parallel imaging is the most clinically used acceleration technique for magnetic resonance imaging (MRI) in part due to its easy inclusion into routine acquisitions. In k-space based parallel imaging reconstruction, sub-sampled k-space data are interpolated using linear convolutions. At high acceleration rates these methods have inherent noise amplification and reduced image quality. On the other hand, non-linear deep learning methods provide improved image quality at high acceleration, but the availability of training databases for different scans, as well as their interpretability hinder their adaptation. In this work, we present an extension of Robust Artificial-neural-networks for k-space Interpolation (RAKI), called residual-RAKI (rRAKI), which achieves scan-specific machine learning reconstruction using a hybrid linear and non-linear methodology. In rRAKI, non-linear CNNs are trained jointly with a linear convolution implemented via a skip connection. In effect, the linear part provides a baseline reconstruction, while the non-linear CNN that runs in parallel provides further reduction of artifacts and noise arising from the linear part. The explicit split between the linear and non-linear aspects of the reconstruction also help improve interpretability compared to purely non-linear methods. Experiments were conducted on the publicly available fastMRI datasets, as well as high-resolution anatomical imaging, comparing GRAPPA and its variants, compressed sensing, RAKI, Scan Specific Artifact Reduction in K-space (SPARK) and the proposed rRAKI. Additionally, highly-accelerated simultaneous multi-slice (SMS) functional MRI reconstructions were also performed, where the proposed rRAKI was compred to Read-out SENSE-GRAPPA and RAKI. Our results show that the proposed rRAKI method substantially improves the image quality compared to conventional parallel imaging, and offers sharper images compared to SPARK and ℓ1-SPIRiT. Furthermore, rRAKI shows improved preservation of time-varying dynamics compared to both parallel imaging and RAKI in highly-accelerated SMS fMRI.

Multi-mask self-supervised learning for physics-guided neural networks in highly accelerated magnetic resonance imaging.

Self-supervised learning has shown great promise because of its ability to train deep learning (DL) magnetic resonance imaging (MRI) reconstruction methods without fully sampled data. Current self-supervised learning methods for physics-guided reconstruction networks split acquired undersampled data into two disjoint sets, where one is used for data consistency (DC) in the unrolled network, while the other is used to define the training loss. In this study, we propose an improved self-supervised learning strategy that more efficiently uses the acquired data to train a physics-guided reconstruction network without a database of fully sampled data. The proposed multi-mask self-supervised learning via data undersampling (SSDU) applies a holdout masking operation on the acquired measurements to split them into multiple pairs of disjoint sets for each training sample, while using one of these pairs for DC units and the other for defining loss, thereby more efficiently using the undersampled data. Multi-mask SSDU is applied on fully sampled 3D knee and prospectively undersampled 3D brain MRI datasets, for various acceleration rates and patterns, and compared with the parallel imaging method, CG-SENSE, and single-mask SSDU DL-MRI, as well as supervised DL-MRI when fully sampled data are available. The results on knee MRI show that the proposed multi-mask SSDU outperforms SSDU and performs as well as supervised DL-MRI. A clinical reader study further ranks the multi-mask SSDU higher than supervised DL-MRI in terms of signal-to-noise ratio and aliasing artifacts. Results on brain MRI show that multi-mask SSDU achieves better reconstruction quality compared with SSDU. The reader study demonstrates that multi-mask SSDU at R = 8 significantly improves reconstruction compared with single-mask SSDU at R = 8, as well as CG-SENSE at R = 2.

Unsupervised Deep Learning Methods for Biological Image Reconstruction and Enhancement: An overview from a signal processing perspective.

Recently, deep learning approaches have become the main research frontier for biological image reconstruction and enhancement problems thanks to their high performance, along with their ultra-fast inference times. However, due to the difficulty of obtaining matched reference data for supervised learning, there has been increasing interest in unsupervised learning approaches that do not need paired reference data. In particular, self-supervised learning and generative models have been successfully used for various biological imaging applications. In this paper, we overview these approaches from a coherent perspective in the context of classical inverse problems, and discuss their applications to biological imaging, including electron, fluorescence and deconvolution microscopy, optical diffraction tomography and functional neuroimaging.

NOise reduction with DIstribution Corrected (NORDIC) PCA in dMRI with complex-valued parameter-free locally low-rank processing.

Diffusion-weighted magnetic resonance imaging (dMRI) has found great utility for a wide range of neuroscientific and clinical applications. However, high-resolution dMRI, which is required for improved delineation of fine brain structures and connectomics, is hampered by its low signal-to-noise ratio (SNR). Since dMRI relies on the acquisition of multiple different diffusion weighted images of the same anatomy, it is well-suited for denoising methods that utilize correlations across the image series to improve the apparent SNR and the subsequent data analysis. In this work, we introduce and quantitatively evaluate a comprehensive framework, NOise Reduction with DIstribution Corrected (NORDIC) PCA method for processing dMRI. NORDIC uses low-rank modeling of g-factor-corrected complex dMRI reconstruction and non-asymptotic random matrix distributions to remove signal components which cannot be distinguished from thermal noise. The utility of the proposed framework for denoising dMRI is demonstrated on both simulations and experimental data obtained at 3 Tesla with different resolutions using human connectome project style acquisitions. The proposed framework leads to substantially enhanced quantitative performance for estimating diffusion tractography related measures and for resolving crossing fibers as compared to a conventional/state-of-the-art dMRI denoising method.

Improved simultaneous multislice cardiac MRI using readout concatenated k-space SPIRiT (ROCK-SPIRiT).

PURPOSE: To develop and evaluate a simultaneous multislice (SMS) reconstruction technique that provides noise reduction and leakage blocking for highly accelerated cardiac MRI. METHODS: ReadOut Concatenated k-space SPIRiT (ROCK-SPIRiT) uses the concept of readout concatenation in image domain to represent SMS encoding, and performs coil self-consistency as in SPIRiT-type reconstruction in an extended k-space, while allowing regularization for further denoising. The proposed method is implemented with and without regularization, and validated on retrospectively SMS-accelerated cine imaging with three-fold SMS and two-fold in-plane acceleration. ROCK-SPIRiT is compared with two leakage-blocking SMS reconstruction methods: readout-SENSE-GRAPPA and split slice-GRAPPA. Further evaluation and comparisons are performed using prospectively SMS-accelerated cine imaging. RESULTS: Results on retrospectively three-fold SMS and two-fold in-plane accelerated cine imaging show that ROCK-SPIRiT without regularization significantly improves on existing methods in terms of PSNR (readout-SENSE-GRAPPA: 33.5 ± 3.2, split slice-GRAPPA: 34.1 ± 3.8, ROCK-SPIRiT: 35.0 ± 3.3) and SSIM (readout-SENSE-GRAPPA: 84.4 ± 8.9, split slice-GRAPPA: 85.0 ± 8.9, ROCK-SPIRiT: 88.2 ± 6.6 [in percentage]). Regularized ROCK-SPIRiT significantly outperforms all methods, as characterized by these quantitative metrics (PSNR: 37.6 ± 3.8, SSIM: 94.2 ± 4.1 [in percentage]). The prospectively five-fold SMS and two-fold in-plane accelerated data show that ROCK-SPIRiT and regularized ROCK-SPIRiT have visually improved image quality compared with existing methods. CONCLUSION: The proposed ROCK-SPIRiT technique reduces noise and interslice leakage in accelerated SMS cardiac cine MRI, improving on existing methods both quantitatively and qualitatively.

Free-breathing simultaneous T1 , T2 , and T2∗ quantification in the myocardium.

PURPOSE: To implement a free-breathing sequence for simultaneous quantification of T1 , T2 , and T2∗ for comprehensive tissue characterization of the myocardium in a single scan using a multi-gradient-echo readout with saturation and T2 preparation pulses. METHODS: In the proposed Saturation And T2 -prepared Relaxometry with Navigator-gating (SATURN) technique, a series of multi-gradient-echo (GRE) images with different magnetization preparations was acquired during free breathing. A total of 35 images were acquired in 26.5 ± 14.9 seconds using multiple saturation times and T2 preparation durations and with imaging at 5 echo times. Bloch simulations and phantom experiments were used to validate a 5-parameter fit model for accurate relaxometry. Free-breathing simultaneous T1 , T2 , and T2∗ measurements were performed in 10 healthy volunteers and 2 patients using SATURN at 3T and quantitatively compared to conventional single-parameter methods such as SASHA for T1 , T2 -prepared bSSFP, and multi-GRE for T2∗ . RESULTS: Simulations confirmed accurate fitting with the 5-parameter model. Phantom measurements showed good agreement with the reference methods in the relevant range for in vivo measurements. Compared to single-parameter methods comparable accuracy was achieved. SATURN produced in vivo parameter maps that were visually comparable to single-parameter methods. No significant difference between T1 , T2 , and T2∗ times acquired with SATURN and single-parameter methods was shown in quantitative measurements (SATURN T1=1573±86ms , T2=33.2±3.6ms , T2∗=25.3±6.1ms ; conventional methods: T1=1544±107ms , T2=33.2±3.6ms , T2∗=23.8±5.5ms ; P>.2 ) CONCLUSION: SATURN enables simultaneous quantification of T1 , T2 , and T2∗ in the myocardium for comprehensive tissue characterization with co-registered maps, in a single scan with good agreement to single-parameter methods.

Diffusion Imaging in the Post HCP Era.

Diffusion imaging is a critical component in the pursuit of developing a better understanding of the human brain. Recent technical advances promise enabling the advancement in the quality of data that can be obtained. In this review the context for different approaches relative to the Human Connectome Project are compared. Significant new gains are anticipated from the use of high-performance head gradients. These gains can be particularly large when the high-performance gradients are employed together with ultrahigh magnetic fields. Transmit array designs are critical in realizing high accelerations in diffusion-weighted (d)MRI acquisitions, while maintaining large field of view (FOV) coverage, and several techniques for optimal signal-encoding are now available. Reconstruction and processing pipelines that precisely disentangle the acquired neuroanatomical information are established and provide the foundation for the application of deep learning in the advancement of dMRI for complex tissues. Level of Evidence: 3 Technical Efficacy Stage: Stage 3.

A field-monitoring-based approach for correcting eddy-current-induced artifacts of up to the 2nd spatial order in human-connectome-project-style multiband diffusion MRI experiment at 7T: A pilot study.

Over the recent years, significant advances in Spin-Echo (SE) Echo-Planar (EP) Diffusion MRI (dMRI) have enabled improved fiber tracking conspicuity in the human brain. At the same time, pushing the spatial resolution and using higher b-values inherently expose the acquired images to further eddy-current-induced distortion and blurring. Recently developed data-driven correction techniques, capable of significantly mitigating these defects, are included in the reconstruction pipelines developed for the Human Connectome Project (HCP) driven by the NIH BRAIN initiative. In this case, however, corrections are derived from the original diffusion-weighted (DW) magnitude images affected by distortion and blurring. Considering the complexity of k-space deviations in the presence of time varying high spatial order eddy currents, distortion and blurring may not be fully reversed when relying on magnitude DW images only. An alternative approach, consisting of iteratively reconstructing DW images based on the actual magnetic field spatiotemporal evolution measured with a magnetic field monitoring camera, has been successfully implemented at 3T in single band dMRI (Wilm et al., 2017, 2015). In this study, we aim to demonstrate the efficacy of this eddy current correction method in the challenging context of HCP-style multiband (MB â€‹= â€‹2) dMRI protocol. The magnetic field evolution was measured during the EP-dMRI readout echo train with a field monitoring camera equipped with 16 19F NMR probes. The time variation of 0th, 1st and 2nd order spherical field harmonics were used to reconstruct DW images. Individual DW images reconstructed with and without field correction were compared. The impact of eddy current correction was evaluated by comparing the corresponding direction-averaged DW images and fractional anisotropy (FA) maps. 19F field monitoring data confirmed the existence of significant field deviations induced by the diffusion-encoding gradients, with variations depending on diffusion gradient amplitude and direction. In DW images reconstructed with the field correction, residual aliasing artifacts were reduced or eliminated, and when high b-values were applied, better gray/white matter delineation and sharper gyri contours were observed, indicating reduced signal blurring. The improvement in image quality further contributed to sharper contours and better gray/white matter delineation in mean DW images and FA maps. In conclusion, we demonstrate that up-to-2nd-order-eddy-current-induced field perturbation in multiband, in-plane accelerated HCP-style dMRI acquisition at 7T can be corrected by integrating the measured field evolution in image reconstruction.

Self-calibrated interpolation of non-Cartesian data with GRAPPA in parallel imaging.

PURPOSE: To develop a non-Cartesian k-space reconstruction method using self-calibrated region-specific interpolation kernels for highly accelerated acquisitions. METHODS: In conventional non-Cartesian GRAPPA with through-time GRAPPA (TT-GRAPPA), the use of region-specific interpolation kernels has demonstrated improved reconstruction quality in dynamic imaging for highly accelerated acquisitions. However, TT-GRAPPA requires the acquisition of a large number of separate calibration scans. To reduce the overall imaging time, we propose Self-calibrated Interpolation of Non-Cartesian data with GRAPPA (SING) to self-calibrate region-specific interpolation kernels from dynamic undersampled measurements. The SING method synthesizes calibration data to adapt to the distinct shape of each region-specific interpolation kernel geometry, and uses a novel local k-space regularization through an extension of TT-GRAPPA. This calibration approach is used to reconstruct non-Cartesian images at high acceleration rates while mitigating noise amplification. The reconstruction quality of SING is compared with conjugate-gradient SENSE and TT-GRAPPA in numerical phantoms and in vivo cine data sets. RESULTS: In both numerical phantom and in vivo cine data sets, SING offers visually and quantitatively similar reconstruction quality to TT-GRAPPA, and provides improved reconstruction quality over conjugate-gradient SENSE. Furthermore, temporal fidelity in SING and TT-GRAPPA is similar for the same acceleration rates. G-factor evaluation over the heart shows that SING and TT-GRAPPA provide similar noise amplification at moderate and high rates. CONCLUSION: The proposed SING reconstruction enables significant improvement of acquisition efficiency for calibration data, while matching the reconstruction performance of TT-GRAPPA.

Self-supervised learning of physics-guided reconstruction neural networks without fully sampled reference data.

PURPOSE: To develop a strategy for training a physics-guided MRI reconstruction neural network without a database of fully sampled data sets. METHODS: Self-supervised learning via data undersampling (SSDU) for physics-guided deep learning reconstruction partitions available measurements into two disjoint sets, one of which is used in the data consistency (DC) units in the unrolled network and the other is used to define the loss for training. The proposed training without fully sampled data is compared with fully supervised training with ground-truth data, as well as conventional compressed-sensing and parallel imaging methods using the publicly available fastMRI knee database. The same physics-guided neural network is used for both proposed SSDU and supervised training. The SSDU training is also applied to prospectively two-fold accelerated high-resolution brain data sets at different acceleration rates, and compared with parallel imaging. RESULTS: Results on five different knee sequences at an acceleration rate of 4 shows that the proposed self-supervised approach performs closely with supervised learning, while significantly outperforming conventional compressed-sensing and parallel imaging, as characterized by quantitative metrics and a clinical reader study. The results on prospectively subsampled brain data sets, in which supervised learning cannot be used due to lack of ground-truth reference, show that the proposed self-supervised approach successfully performs reconstruction at high acceleration rates (4, 6, and 8). Image readings indicate improved visual reconstruction quality with the proposed approach compared with parallel imaging at acquisition acceleration. CONCLUSION: The proposed SSDU approach allows training of physics-guided deep learning MRI reconstruction without fully sampled data, while achieving comparable results with supervised deep learning MRI trained on fully sampled data.

Self-navigation for 3D multishot EPI with data-reference.

PURPOSE: In this study, we sought to develop a self-navigation strategy for improving the reconstruction of diffusion weighted 3D multishot echo planar imaging (EPI). We propose a method for extracting the phase correction information from the acquisition itself, eliminating the need for a 2D navigator, further accelerating the acquisition. METHODS: In-vivo acquisitions at 3T with 0.9 mm and 1.5 mm isotropic resolutions were used to evaluate the performance of the self-navigation strategy. Sensitivity to motion was tested using a large difference in pitch position of the head. Using a multishell diffusion weighted acquisition, tractography results were obtained at (0.9 mm)3 to validate the quality with conventional acquisition. RESULTS: The use of 3D multislab EPI with self-navigation enables 3D diffusion-weighted spin echo EPI acquisitions that have the same efficiency as 2D single-shot acquisition. For matched acquisition time the image signal-to-noise ratio (SNR) between 3D and 2D acquisition is shown to be comparable for whole-brain coverage with (1.5 mm)3 resolution and for (0.9 mm)3 resolution the 3D acquisition has higher SNR than what can be obtained with 2D acquisitions using current state-of-art multiband techniques. The self-navigation technique was shown to be stable under inter-volume motion. In tractography analysis, the higher resolution afforded by our technique enabled clear delineation of the tapetum and posterior corona radiata. CONCLUSION: The proposed self-navigation approach utilized a self-consistent phase in 3D diffusion weighted acquisitions. Its efficiency and stability were demonstrated for a plurality of common acquisitions. The proposed self-navigation approach allows for faster acquisition of 3D multishot EPI desirable for large field of view and/or higher resolution.

Deep-Learning Methods for Parallel Magnetic Resonance Imaging Reconstruction: A Survey of the Current Approaches, Trends, and Issues.

Following the success of deep learning in a wide range of applications, neural network-based machine learning techniques have received interest as a means of accelerating magnetic resonance imaging (MRI). A number of ideas inspired by deep learning techniques from computer vision and image processing have been successfully applied to non-linear image reconstruction in the spirit of compressed sensing for both low dose computed tomography and accelerated MRI. The additional integration of multi-coil information to recover missing k-space lines in the MRI reconstruction process, is still studied less frequently, even though it is the de-facto standard for currently used accelerated MR acquisitions. This manuscript provides an overview of the recent machine learning approaches that have been proposed specifically for improving parallel imaging. A general background introduction to parallel MRI is given that is structured around the classical view of image space and k-space based methods. Both linear and non-linear methods are covered, followed by a discussion of recent efforts to further improve parallel imaging using machine learning, and specifically using artificial neural networks. Image-domain based techniques that introduce improved regularizers are covered as well as k-space based methods, where the focus is on better interpolation strategies using neural networks. Issues and open problems are discussed as well as recent efforts for producing open datasets and benchmarks for the community.

Scan-specific robust artificial-neural-networks for k-space interpolation (RAKI) reconstruction: Database-free deep learning for fast imaging.

PURPOSE: To develop an improved k-space reconstruction method using scan-specific deep learning that is trained on autocalibration signal (ACS) data. THEORY: Robust artificial-neural-networks for k-space interpolation (RAKI) reconstruction trains convolutional neural networks on ACS data. This enables nonlinear estimation of missing k-space lines from acquired k-space data with improved noise resilience, as opposed to conventional linear k-space interpolation-based methods, such as GRAPPA, which are based on linear convolutional kernels. METHODS: The training algorithm is implemented using a mean square error loss function over the target points in the ACS region, using a gradient descent algorithm. The neural network contains 3 layers of convolutional operators, with 2 of these including nonlinear activation functions. The noise performance and reconstruction quality of the RAKI method was compared with GRAPPA in phantom, as well as in neurological and cardiac in vivo data sets. RESULTS: Phantom imaging shows that the proposed RAKI method outperforms GRAPPA at high (≥4) acceleration rates, both visually and quantitatively. Quantitative cardiac imaging shows improved noise resilience at high acceleration rates (rate 4:23% and rate 5:48%) over GRAPPA. The same trend of improved noise resilience is also observed in high-resolution brain imaging at high acceleration rates. CONCLUSION: The RAKI method offers a training database-free deep learning approach for MRI reconstruction, with the potential to improve many existing reconstruction approaches, and is compatible with conventional data acquisition protocols.