Extreme MRI: Large-scale volumetric dynamic imaging from continuous non-gated acquisitions

Reproducible Research Practices in Magnetic Resonance Neuroimaging: A Review Informed by Advanced Language Models

MRI has progressed significantly with the introduction of advanced computational methods and novel imaging techniques, but their wider adoption hinges on their reproducibility. This concise review synthesizes reproducible research insights from recent MRI articles to examine the current state of reproducibility in neuroimaging, highlighting key trends and challenges. It also provides a custom generative pretrained transformer (GPT) model, designed specifically for aiding in an automated analysis and synthesis of information pertaining to the reproducibility insights associated with the articles at the core of this review.

Scan-Specific Self-Supervised Bayesian Deep Non-Linear Inversion for Undersampled MRI Reconstruction

Magnetic resonance imaging is subject to slow acquisition times due to the inherent limitations in data sampling. Recently, supervised deep learning has emerged as a promising technique for reconstructing sub-sampled MRI. However, supervised deep learning requires a large dataset of fully-sampled data. Although unsupervised or self-supervised deep learning methods have emerged to address the limitations of supervised deep learning approaches, they still require a database of images. In contrast, scan-specific deep learning methods learn and reconstruct using only the sub-sampled data from a single scan. Here, we introduce Scan-Specific Self-Supervised Bayesian Deep Non-Linear Inversion (DNLINV) that does not require an auto calibration scan region. DNLINV utilizes a Deep Image Prior-type generative modeling approach and relies on approximate Bayesian inference to regularize the deep convolutional neural network. We demonstrate our approach on several anatomies, contrasts, and sampling patterns and show improved performance over existing approaches in scan-specific calibrationless parallel imaging and compressed sensing.

Comparison of weighting algorithms to mitigate respiratory motion in free-breathing neonatal pulmonary radial UTE-MRI

Background. Thoracoabdominal MRI is limited by respiratory motion, especially in populations who cannot perform breath-holds. One approach for reducing motion blurring in radially-acquired MRI is respiratory gating. Straightforward 'hard-gating' uses only data from a specified respiratory window and suffers from reduced SNR. Proposed 'soft-gating' reconstructions may improve scan efficiency but reduce motion correction by incorporating data with nonzero weight acquired outside the specified window. However, previous studies report conflicting benefits, and importantly the choice of soft-gated weighting algorithm and effect on image quality has not previously been explored. The purpose of this study is to map how variable soft-gated weighting functions and parameters affect signal and motion blurring in respiratory-gated reconstructions of radial lung MRI, using neonates as a model population.Methods. Ten neonatal inpatients with respiratory abnormalities were imaged using a 1.5 T neonatal-sized scanner and 3D radial ultrashort echo-time (UTE) sequence. Images were reconstructed using ungated, hard-gated, and several soft-gating weighting algorithms (exponential, sigmoid, inverse, and linear weighting decay outside the period of interest), with %Nprojrepresenting the relative amount of data included. The apparent SNR (aSNR) and motion blurring (measured by the maximum derivative of image intensity at the diaphragm, MDD) were compared between reconstructions.Results. Soft-gating functions produced higher aSNR and lower MDD than hard-gated images using equivalent %Nproj, as expected. aSNR was not identical between different gating schemes for given %Nproj. While aSNR was approximately linear with %Nprojfor each algorithm, MDD performance diverged between functions as %Nprojdecreased. Algorithm performance was relatively consistent between subjects, except in images with high noise.Conclusion. The algorithm selection for soft-gating has a notable effect on image quality of respiratory-gated MRI; the timing of included data across the respiratory phase, and not simply the amount of data, plays an important role in aSNR. The specific soft-gating function and parameters should be considered for a given imaging application's requirements of signal and sharpness.

IMJENSE: Scan-Specific Implicit Representation for Joint Coil Sensitivity and Image Estimation in Parallel MRI

Parallel imaging is a commonly used technique to accelerate magnetic resonance imaging (MRI) data acquisition. Mathematically, parallel MRI reconstruction can be formulated as an inverse problem relating the sparsely sampled k-space measurements to the desired MRI image. Despite the success of many existing reconstruction algorithms, it remains a challenge to reliably reconstruct a high-quality image from highly reduced k-space measurements. Recently, implicit neural representation has emerged as a powerful paradigm to exploit the internal information and the physics of partially acquired data to generate the desired object. In this study, we introduced IMJENSE, a scan-specific implicit neural representation-based method for improving parallel MRI reconstruction. Specifically, the underlying MRI image and coil sensitivities were modeled as continuous functions of spatial coordinates, parameterized by neural networks and polynomials, respectively. The weights in the networks and coefficients in the polynomials were simultaneously learned directly from sparsely acquired k-space measurements, without fully sampled ground truth data for training. Benefiting from the powerful continuous representation and joint estimation of the MRI image and coil sensitivities, IMJENSE outperforms conventional image or k-space domain reconstruction algorithms. With extremely limited calibration data, IMJENSE is more stable than supervised calibrationless and calibration-based deep-learning methods. Results show that IMJENSE robustly reconstructs the images acquired at 5× and 6× accelerations with only 4 or 8 calibration lines in 2D Cartesian acquisitions, corresponding to 22.0% and 19.5% undersampling rates. The high-quality results and scanning specificity make the proposed method hold the potential for further accelerating the data acquisition of parallel MRI.

Coil sketching for computationally efficient MR iterative reconstruction

PURPOSE

Parallel imaging and compressed sensing reconstructions of large MRI datasets often have a prohibitive computational cost that bottlenecks clinical deployment, especially for three-dimensional (3D) non-Cartesian acquisitions. One common approach is to reduce the number of coil channels actively used during reconstruction as in coil compression. While effective for Cartesian imaging, coil compression inherently loses signal energy, producing shading artifacts that compromise image quality for 3D non-Cartesian imaging. We propose coil sketching, a general and versatile method for computationally-efficient iterative MR image reconstruction.

THEORY AND METHODS

We based our method on randomized sketching algorithms, a type of large-scale optimization algorithms well established in the fields of machine learning and big data analysis. We adapt the sketching theory to the MRI reconstruction problem via a structured sketching matrix that, similar to coil compression, considers high-energy virtual coils obtained from principal component analysis. But, unlike coil compression, it also considers random linear combinations of the remaining low-energy coils, effectively leveraging information from all coils.

RESULTS

First, we performed ablation experiments to validate the sketching matrix design on both Cartesian and non-Cartesian datasets. The resulting design yielded both improved computatioanal efficiency and preserved signal-to-noise ratio (SNR) as measured by the inverse g-factor. Then, we verified the efficacy of our approach on high-dimensional non-Cartesian 3D cones datasets, where coil sketching yielded up to three-fold faster reconstructions with equivalent image quality.

CONCLUSION

Coil sketching is a general and versatile reconstruction framework for computationally fast and memory-efficient reconstruction.

Free-breathing high isotropic resolution quantitative susceptibility mapping (QSM) of liver using 3D multi-echo UTE cones acquisition and respiratory motion-resolved image reconstruction

PURPOSE

To enable free-breathing and high isotropic resolution liver quantitative susceptibility mapping (QSM) using 3D multi-echo UTE cones acquisition and respiratory motion-resolved image reconstruction.

METHODS

Using 3D multi-echo UTE cones MRI, a respiratory motion was estimated from the k-space center of the imaging data. After sorting the k-space data with estimated motion, respiratory motion state-resolved reconstruction was performed for multi-echo data followed by nonlinear least-squares fitting for proton density fat fraction (PDFF),R 2 * $$ {\mathrm{R}}_2^{\ast } $$ , and fat-corrected B0 field maps. PDFF and B0 field maps were subsequently used for QSM reconstruction. The proposed method was compared with motion-averaged (gridding) reconstruction and conventional 3D multi-echo Cartesian MRI in moving gadolinium phantom and in vivo studies. Region of interest (ROI)-based linear regression analysis was performed on these methods to investigate correlations between gadolinium concentration and QSM in the phantom study and betweenR 2 * $$ {\mathrm{R}}_2^{\ast } $$ and QSM in in vivo study.

RESULTS

Cones with motion-resolved reconstruction showed sharper image quality compared to motion-averaged reconstruction with a substantial reduction of motion artifacts in both moving phantom and in vivo studies. For ROI-based linear regression analysis of the phantom study, susceptibility values from cones with motion-resolved reconstruction (QSM ppm $$ {\mathrm{QSM}}_{\mathrm{ppm}} $$  = 0.31 × gadolinium mM + $$ \times {\mathrm{gadolinium}}_{\mathrm{mM}}+ $$ 0.05,R 2 $$ {R}^2 $$  = 0.999) and Cartesian without motion (QSM ppm $$ {\mathrm{QSM}}_{\mathrm{ppm}} $$  = 0.32× gadolinium mM + $$ \times {\mathrm{gadolinium}}_{\mathrm{mM}}+ $$ 0.04,R 2 $$ {R}^2 $$  = 1.000) showed linear relationships with gadolinium concentrations and showed good agreement with each other. For in vivo, motion-resolved reconstruction showed higher goodness of fit (QSM ppm $$ {\mathrm{QSM}}_{\mathrm{ppm}} $$  = 0.00261 × R 2 s - 1 * - $$ \times {\mathrm{R}}_{2_{{\mathrm{s}}^{-1}}}^{\ast }- $$ 0.524,R 2 $$ {R}^2 $$  = 0.977) compared to motion-averaged reconstruction (QSM ppm $$ {\mathrm{QSM}}_{\mathrm{ppm}} $$ = 0.0021 × R 2 s - 1 * - $$ \times {\mathrm{R}}_{2_{{\mathrm{s}}^{-1}}}^{\ast }- $$ 0.572,R 2 $$ {R}^2 $$ = 0.723) in ROI-based linear regression analysis betweenR 2 * $$ {\mathrm{R}}_2^{\ast } $$ and QSM.

CONCLUSION

Feasibility of free-breathing liver QSM was demonstrated with motion-resolved 3D multi-echo UTE cones MRI, achieving high isotropic resolution currently unachievable in conventional Cartesian MRI.

Time-Resolved 3D cardiopulmonary MRI reconstruction using spatial transformer network

The accurate visualization and assessment of the complex cardiac and pulmonary structures in 3D is critical for the diagnosis and treatment of cardiovascular and respiratory disorders. Conventional 3D cardiac magnetic resonance imaging (MRI) techniques suffer from long acquisition times, motion artifacts, and limited spatiotemporal resolution. This study proposes a novel time-resolved 3D cardiopulmonary MRI reconstruction method based on spatial transformer networks (STNs) to reconstruct the 3D cardiopulmonary MRI acquired using 3D center-out radial ultra-short echo time (UTE) sequences. The proposed reconstruction method employed an STN-based deep learning framework, which used a combination of data-processing, grid generator, and sampler. The reconstructed 3D images were compared against the start-of-the-art time-resolved reconstruction method. The results showed that the proposed time-resolved 3D cardiopulmonary MRI reconstruction using STNs offers a robust and efficient approach to obtain high-quality images. This method effectively overcomes the limitations of conventional 3D cardiac MRI techniques and has the potential to improve the diagnosis and treatment planning of cardiopulmonary disorders.

Motion compensated self supervised deep learning for highly accelerated 3D ultrashort Echo time pulmonary MRI

PURPOSE

To investigate motion compensated, self-supervised, model based deep learning (MBDL) as a method to reconstruct free breathing, 3D pulmonary UTE acquisitions.

THEORY AND METHODS

A self-supervised eXtra dimension MBDL architecture (XD-MBDL) was developed that combined respiratory states to reconstruct a single high-quality 3D image. Non-rigid motion fields were incorporated into this architecture by estimating motion fields from a lower resolution motion resolved (XD-GRASP) reconstruction. Motion compensated XD-MBDL was evaluated on lung UTE datasets with and without contrast and compared to constrained reconstructions and variants of self-supervised MBDL that do not account for dynamic respiratory states or leverage motion correction.

RESULTS

Images reconstructed using XD-MBDL demonstrate improved image quality as measured by apparent SNR (aSNR), contrast to noise ratio (CNR), and visual assessment relative to self-supervised MBDL approaches that do not account for dynamic respiratory states, XD-GRASP and a recently proposed motion compensated iterative reconstruction strategy (iMoCo). Additionally, XD-MBDL reduced reconstruction time relative to both XD-GRASP and iMoCo.

CONCLUSION

A method was developed to allow self-supervised MBDL to combine multiple respiratory states to reconstruct a single image. This method was combined with graphics processing unit (GPU)-based image registration to further improve reconstruction quality. This approach showed promising results reconstructing a user-selected respiratory phase from free breathing 3D pulmonary UTE acquisitions.

Latent signal models: Learning compact representations of signal evolution for improved time-resolved, multi-contrast MRI

PURPOSE

To improve time-resolved reconstructions by training auto-encoders to learn compact representations of Bloch-simulated signal evolution and inserting the decoder into the forward model.

METHODS

Building on model-based nonlinear and linear subspace techniques, we train auto-encoders on dictionaries of simulated signal evolution to learn compact, nonlinear, latent representations. The proposed latent signal model framework inserts the decoder portion of the auto-encoder into the forward model and directly reconstructs the latent representation. Latent signal models essentially serve as a proxy for fast and feasible differentiation through the Bloch equations used to simulate signal. This work performs experiments in the context of T₂ -shuffling, gradient echo EPTI, and MPRAGE-shuffling. We compare how efficiently auto-encoders represent signal evolution in comparison to linear subspaces. Simulation and in vivo experiments then evaluate if reducing degrees of freedom by incorporating our proxy for the Bloch equations, the decoder portion of the auto-encoder, into the forward model improves reconstructions in comparison to subspace constraints.

RESULTS

An auto-encoder with 1 real latent variable represents single-tissue fast spin echo, EPTI, and MPRAGE signal evolution to within 0.15% normalized RMS error, enabling reconstruction problems with 3 degrees of freedom per voxel (real latent variable + complex scaling) in comparison to linear models with 4-8 degrees of freedom per voxel. In simulated/in vivo T₂ -shuffling and in vivo EPTI experiments, the proposed framework achieves consistent quantitative normalized RMS error improvement over linear approaches. From qualitative evaluation, the proposed approach yields images with reduced blurring and noise amplification in MPRAGE-shuffling experiments.

CONCLUSION

Directly solving for nonlinear latent representations of signal evolution improves time-resolved MRI reconstructions.

Deep Learning-Based Reconstruction for Cardiac MRI: A Review

Cardiac magnetic resonance (CMR) is an essential clinical tool for the assessment of cardiovascular disease. Deep learning (DL) has recently revolutionized the field through image reconstruction techniques that allow unprecedented data undersampling rates. These fast acquisitions have the potential to considerably impact the diagnosis and treatment of cardiovascular disease. Herein, we provide a comprehensive review of DL-based reconstruction methods for CMR. We place special emphasis on state-of-the-art unrolled networks, which are heavily based on a conventional image reconstruction framework. We review the main DL-based methods and connect them to the relevant conventional reconstruction theory. Next, we review several methods developed to tackle specific challenges that arise from the characteristics of CMR data. Then, we focus on DL-based methods developed for specific CMR applications, including flow imaging, late gadolinium enhancement, and quantitative tissue characterization. Finally, we discuss the pitfalls and future outlook of DL-based reconstructions in CMR, focusing on the robustness, interpretability, clinical deployment, and potential for new methods.

4D Golden-Angle Radial MRI at Subsecond Temporal Resolution

Intraframe motion blurring, as a major challenge in free-breathing dynamic MRI, can be reduced if high temporal resolution can be achieved. To address this challenge, this work proposes a highly accelerated 4D (3D + time) dynamic MRI framework with subsecond temporal resolution that does not require explicit motion compensation. The method combines standard stack-of-stars golden-angle radial sampling and tailored GRASP-Pro (Golden-angle RAdial Sparse Parallel imaging with imProved performance) reconstruction. Specifically, 4D dynamic MRI acquisition is performed continuously without motion gating or sorting. The k-space centers in stack-of-stars radial data are organized to guide estimation of a temporal basis, with which GRASP-Pro reconstruction is employed to enforce joint low-rank subspace and sparsity constraints. This new basis estimation strategy is the new feature proposed for subspace-based reconstruction in this work to achieve high temporal resolution (e.g., subsecond/3D volume). It does not require sequence modification to acquire additional navigation data, it is compatible with commercially available stack-of-stars sequences, and it does not need an intermediate reconstruction step. The proposed 4D dynamic MRI approach was tested in abdominal motion phantom, free-breathing abdominal MRI, and dynamic contrast-enhanced MRI (DCE-MRI). Our results have shown that GRASP-Pro reconstruction with the new basis estimation strategy enables highly-accelerated 4D dynamic imaging at subsecond temporal resolution (with five spokes or less for each dynamic frame per image slice) for both free-breathing non-DCE-MRI and DCE-MRI. In the abdominal phantom, better image quality with lower root mean square error and higher structural similarity index was achieved using GRASP-Pro compared with standard GRASP. With the ability to acquire each 3D image in less than 1 s, intraframe respiratory blurring can be intrinsically reduced for body applications with our approach, which eliminates the need for explicit motion detection and motion compensation.

4D lung MRI with high-isotropic-resolution using half-spoke (UTE) and full-spoke 3D radial acquisition and temporal compressed sensing reconstruction

Objective. To develop a respiratory motion-resolved four-dimensional (4D) magnetic resonance imaging (MRI) technique with high-isotropic-resolution (1.1 mm) using 3D radial sampling, camera-based respiratory motion sensing, and temporal compressed sensing reconstruction for lung cancer imaging.Approach. Free-breathing half- and full-spoke 3D golden-angle radial acquisitions were performed on eight healthy volunteers and eight patients with lung tumors of varying size. A back-and-forth k-space ordering between consecutive interleaves of the 3D radial acquisition was performed to minimize eddy current-related artifacts. Data were sorted into respiratory motion states using camera-based motion navigation and 4D images were reconstructed using temporal compressed sensing to reduce scan time. Normalized sharpness indices of the diaphragm, apparent signal-to-noise ratio (aSNR) and contrast-to-noise ratio (CNR) of the lung tumor (patients only), liver, and aortic arch were compared between half- and full-spoke 4D MRI images to evaluate the impact of respiratory motion and image contrast on 4D MRI image quality. Respiration-induced changes in lung volumes and center of mass shifts were compared between half- and full-spoke 4D MRI measurements. In addition, the motion measurements from 4D MRI and the same-day 4D CT were presented in one of the lung tumor patients.Main results. Half-spoke 4D MRI provides better visualization of the lung parenchyma, while full-spoke 4D MRI presents sharper diaphragm images and higher aSNR and CNR in the lung tumor, liver, and aortic arch. Lung volume changes and center of mass shifts measured by half- and full-spoke 4D MRI were not statistically different. For the patient with 4D MRI and same-day 4D CT, lung volume changes and center of mass shifts were generally comparable.Significance. This work demonstrates the feasibility of a motion-resolved 4D MRI technique with high-isotropic-resolution using 3D radial acquisition, camera-based respiratory motion sensing, and temporal compressed sensing reconstruction for treatment planning and motion monitoring in radiotherapy of lung cancer.

Motion-corrected 4D-Flow MRI for neurovascular applications

Neurovascular 4D-Flow MRI has emerged as a powerful tool for comprehensive cerebrovascular hemodynamic characterization. Clinical studies in at risk populations such as aging adults indicate hemodynamic markers can be confounded by motion-induced bias. This study develops and characterizes a high fidelity 3D self-navigation approach for retrospective rigid motion correction of neurovascular 4D-Flow data. A 3D radial trajectory with pseudorandom ordering was combined with a multi-resolution low rank regularization approach to enable high spatiotemporal resolution self-navigators from extremely undersampled data. Phantom and volunteer experiments were performed at 3.0T to evaluate the ability to correct for different amounts of induced motions. In addition, the approach was applied to clinical-research exams from ongoing aging studies to characterize performance in the clinical setting. Simulations, phantom and volunteer experiments with motion correction produced images with increased vessel conspicuity, reduced image blurring, and decreased variability in quantitative measures. Clinical exams revealed significant changes in hemodynamic parameters including blood flow rates, flow pulsatility index, and lumen areas after motion correction in probed cerebral arteries (Flow: P<0.001 Lt ICA, P=0.002 Rt ICA, P=0.004 Lt MCA, P=0.004 Rt MCA; Area: P<0.001 Lt ICA, P<0.001 Rt ICA, P=0.004 Lt MCA, P=0.004 Rt MCA; flow pulsatility index: P=0.042 Rt ICA, P=0.002 Lt MCA). Motion induced bias can lead to significant overestimation of hemodynamic markers in cerebral arteries. The proposed method reduces measurement bias from rigid motion in neurovascular 4D-Flow MRI in challenging populations such as aging adults.

Dynamic Imaging Using Deep Bi-Linear Unsupervised Representation (DEBLUR)

Bilinear models such as low-rank and dictionary methods, which decompose dynamic data to spatial and temporal factor matrices are powerful and memory-efficient tools for the recovery of dynamic MRI data. Current bilinear methods rely on sparsity and energy compaction priors on the factor matrices to regularize the recovery. Motivated by deep image prior, we introduce a novel bilinear model, whose factor matrices are generated using convolutional neural networks (CNNs). The CNN parameters, and equivalently the factors, are learned from the undersampled data of the specific subject. Unlike current unrolled deep learning methods that require the storage of all the time frames in the dataset, the proposed approach only requires the storage of the factors or compressed representation; this approach allows the direct use of this scheme to large-scale dynamic applications, including free breathing cardiac MRI considered in this work. To reduce the run time and to improve performance, we initialize the CNN parameters using existing factor methods. We use sparsity regularization of the network parameters to minimize the overfitting of the network to measurement noise. Our experiments on free-breathing and ungated cardiac cine data acquired using a navigated golden-angle gradient-echo radial sequence show the ability of our method to provide reduced spatial blurring as compared to classical bilinear methods as well as a recent unsupervised deep-learning approach.

Three-dimensional ultrashort echo time magnetic resonance imaging in assessment of idiopathic pulmonary fibrosis, in comparison with high-resolution computed tomography

Background

We aimed to evaluate the image quality, feasibility, and diagnostic performance of three-dimensional ultrashort echo time magnetic resonance imaging (3D UTE-MRI) to assess idiopathic pulmonary fibrosis (IPF) compared with high-resolution computed tomography (HRCT) and half-Fourier single-shot turbo spin-echo (HASTE) MRI.

Methods

A total of 36 patients with IPF (34 men; mean age: 62±8 years, age range: 43 to 78 years) were prospectively included and underwent HRCT and chest MRI on the same day. Chest MRI was performed with a free-breathing 3D spiral UTE pulse sequence and HASTE sequence on a 1.5 T MRI. Two radiologists independently evaluated the image quality of the HRCT, HASTE, and 3D UTE-MRI. They assessed the representative imaging features of IPF, including honeycombing, reticulation, traction bronchiectasis, and ground-glass opacities. Image quality of the 3D UTE-MRI, HASTE, and HRCT were assessed using a 5-point visual scoring method. Kappa and weighted kappa analysis were used to measure intra- and inter-observer and inter-method agreements. Sensitivity (SE), specificity (SP), and accuracy (AC) were used to assess the performance of 3D UTE-MRI for detecting image features of IPF and monitoring the extent of pulmonary fibrosis. Linear regressions and Bland-Altman plots were generated to assess the correlation and agreement between the assessment of the extent of pulmonary fibrosis made by the 2 observers.

Results

The image quality of HRCT was higher than that of HASTE and UTE-MRI (HRCT vs. UTE-MRI vs. HASTE: 4.9±0.3 vs. 4.1±0.7 vs. 3.0±0.3; P<0.001). Interobserver agreement of HRCT, HASTE, and 3D UTE-MRI when assessing pulmonary fibrosis was substantial and excellent (HRCT: 0.727≤ κ ≤1, P<0.001; HASTE: 0.654≤ κ ≤1, P<0.001; 3D UTE-MRI: 0.719≤ κ ≤0.824, P<0.001). In addition, reticulation (SE: 97.1%; SP: 100%; AC: 97.2%; κ =0.654), honeycombing (SE: 83.3%; SP: 100%; AC: 86.1%; κ =0.625) patterns, and traction bronchiectasis (SE: 94.1%; SP: 100%; AC: 94.4%, κ =0.640) were also well-visualized on 3D UTE-MRI, which was significantly superior to HASTE. Compared with HRCT, the sensitivity of 3D UTE-MRI to detect signs of pulmonary fibrosis (n=35) was 97.2%. The interobserver agreement in elevation of the extent of pulmonary fibrosis with HRCT and 3D UTE-MRI was R²=0.84 (P<0.001) and R²=0.84 (P<0.001), respectively. The extent of pulmonary fibrosis assessed with 3D UTE-MRI [median =9, interquartile range (IQR): 6.25 to 10.00] was lower than that from HRCT (median =12, IQR: 9.25 to 13.00; U=320.00, P<0.001); however, they had a positive correlation (R=0.72, P<0.001).

Conclusions

As a radiation-free non-contrast enhanced imaging method, although the image quality of 3D UTE-MRI is inferior to that of HRCT, it has high reproducibility to identify the imaging features of IPF and evaluate the extent of pulmonary fibrosis.

Cardiac MR: From Theory to Practice

Cardiovascular disease (CVD) is the leading single cause of morbidity and mortality, causing over 17. 9 million deaths worldwide per year with associated costs of over $800 billion. Improving prevention, diagnosis, and treatment of CVD is therefore a global priority. Cardiovascular magnetic resonance (CMR) has emerged as a clinically important technique for the assessment of cardiovascular anatomy, function, perfusion, and viability. However, diversity and complexity of imaging, reconstruction and analysis methods pose some limitations to the widespread use of CMR. Especially in view of recent developments in the field of machine learning that provide novel solutions to address existing problems, it is necessary to bridge the gap between the clinical and scientific communities. This review covers five essential aspects of CMR to provide a comprehensive overview ranging from CVDs to CMR pulse sequence design, acquisition protocols, motion handling, image reconstruction and quantitative analysis of the obtained data. (1) The basic MR physics of CMR is introduced. Basic pulse sequence building blocks that are commonly used in CMR imaging are presented. Sequences containing these building blocks are formed for parametric mapping and functional imaging techniques. Commonly perceived artifacts and potential countermeasures are discussed for these methods. (2) CMR methods for identifying CVDs are illustrated. Basic anatomy and functional processes are described to understand the cardiac pathologies and how they can be captured by CMR imaging. (3) The planning and conduct of a complete CMR exam which is targeted for the respective pathology is shown. Building blocks are illustrated to create an efficient and patient-centered workflow. Further strategies to cope with challenging patients are discussed. (4) Imaging acceleration and reconstruction techniques are presented that enable acquisition of spatial, temporal, and parametric dynamics of the cardiac cycle. The handling of respiratory and cardiac motion strategies as well as their integration into the reconstruction processes is showcased. (5) Recent advances on deep learning-based reconstructions for this purpose are summarized. Furthermore, an overview of novel deep learning image segmentation and analysis methods is provided with a focus on automatic, fast and reliable extraction of biomarkers and parameters of clinical relevance.

Data-Consistent non-Cartesian deep subspace learning for efficient dynamic MR image reconstruction

Non-Cartesian sampling with subspace-constrained image reconstruction is a popular approach to dynamic MRI, but slow iterative reconstruction limits its clinical application. Data-consistent (DC) deep learning can accelerate reconstruction with good image quality, but has not been formulated for non-Cartesian subspace imaging. In this study, we propose a DC non-Cartesian deep subspace learning framework for fast, accurate dynamic MR image reconstruction. Four novel DC formulations are developed and evaluated: two gradient decent approaches, a directly solved approach, and a conjugate gradient approach. We applied a U-Net model with and without DC layers to reconstruct T1-weighted images for cardiac MR Multitasking (an advanced multidimensional imaging method), comparing our results to the iteratively reconstructed reference. Experimental results show that the proposed framework significantly improves reconstruction accuracy over the U-Net model without DC, while significantly accelerating the reconstruction over conventional iterative reconstruction.

Optimized multi-axis spiral projection MR fingerprinting with subspace reconstruction for rapid whole-brain high-isotropic-resolution quantitative imaging

PURPOSE

To improve image quality and accelerate the acquisition of 3D MR fingerprinting (MRF).

METHODS

Building on the multi-axis spiral-projection MRF technique, a subspace reconstruction with locally low-rank constraint and a modified spiral-projection spatiotemporal encoding scheme called tiny golden-angle shuffling were implemented for rapid whole-brain high-resolution quantitative mapping. Reconstruction parameters such as the locally low-rank regularization parameter and the subspace rank were tuned using retrospective in vivo data and simulated examinations. B₀ inhomogeneity correction using multifrequency interpolation was incorporated into the subspace reconstruction to further improve the image quality by mitigating blurring caused by off-resonance effect.

RESULTS

The proposed MRF acquisition and reconstruction framework yields high-quality 1-mm isotropic whole-brain quantitative maps in 2 min at better quality compared with 6-min acquisitions of prior approaches. The proposed method was validated to not induce bias in T₁ and T₂ mapping. High-quality whole-brain MRF data were also obtained at 0.66-mm isotropic resolution in 4 min using the proposed technique, where the increased resolution was shown to improve visualization of subtle brain structures.

CONCLUSIONS

The proposed tiny golden-angle shuffling, MRF with optimized spiral-projection trajectory and subspace reconstruction enables high-resolution quantitative mapping in ultrafast acquisition time.

Scan-specific artifact reduction in k-space (SPARK) neural networks synergize with physics-based reconstruction to accelerate MRI

PURPOSE

To develop a scan-specific model that estimates and corrects k-space errors made when reconstructing accelerated MRI data.

METHODS

Scan-specific artifact reduction in k-space (SPARK) trains a convolutional-neural-network to estimate and correct k-space errors made by an input reconstruction technique by back-propagating from the mean-squared-error loss between an auto-calibration signal (ACS) and the input technique's reconstructed ACS. First, SPARK is applied to generalized autocalibrating partially parallel acquisitions (GRAPPA) and demonstrates improved robustness over other scan-specific models, such as robust artificial-neural-networks for k-space interpolation (RAKI) and residual-RAKI. Subsequent experiments demonstrate that SPARK synergizes with residual-RAKI to improve reconstruction performance. SPARK also improves reconstruction quality when applied to advanced acquisition and reconstruction techniques like 2D virtual coil (VC-) GRAPPA, 2D LORAKS, 3D GRAPPA without an integrated ACS region, and 2D/3D wave-encoded imaging.

RESULTS

SPARK yields SSIM improvement and 1.5 - 2× root mean squared error (RMSE) reduction when applied to GRAPPA and improves robustness to ACS size for various acceleration rates in comparison to other scan-specific techniques. When applied to advanced reconstruction techniques such as residual-RAKI, 2D VC-GRAPPA and LORAKS, SPARK achieves up to 20% RMSE improvement. SPARK with 3D GRAPPA also improves RMSE performance by ~2×, SSIM performance, and perceived image quality without a fully sampled ACS region. Finally, SPARK synergizes with non-Cartesian, 2D and 3D wave-encoding imaging by reducing RMSE between 20% and 25% and providing qualitative improvements.

CONCLUSION

SPARK synergizes with physics-based acquisition and reconstruction techniques to improve accelerated MRI by training scan-specific models to estimate and correct reconstruction errors in k-space.

Real-Time Non-Rigid 3D Respiratory Motion Estimation for MR-Guided Radiotherapy Using MR-MOTUS

The MR-Linac is a combination of an MR-scanner and radiotherapy linear accelerator (Linac) which holds the promise to increase the precision of radiotherapy treatments with MR-guided radiotherapy by monitoring motion during radiotherapy with MRI, and adjusting the radiotherapy plan accordingly. Optimal MR-guidance for respiratory motion during radiotherapy requires MR-based 3D motion estimation with a latency of 200-500 ms. Currently this is still challenging since typical methods rely on MR-images, and are therefore limited by the 3D MR-imaging latency. In this work, we present a method to perform non-rigid 3D respiratory motion estimation with 170 ms latency, including both acquisition and reconstruction. The proposed method called real-time low-rank MR-MOTUS reconstructs motion-fields directly from k -space data, and leverages an explicit low-rank decomposition of motion-fields to split the large scale 3D+t motion-field reconstruction problem posed in our previous work into two parts: (I) a medium-scale offline preparation phase and (II) a small-scale online inference phase which exploits the results of the offline phase for real-time computations. The method was validated on free-breathing data of five volunteers, acquired with a 1.5T Elekta Unity MR-Linac. Results show that the reconstructed 3D motion-field are anatomically plausible, highly correlated with a self-navigation motion surrogate ( R=0.975 ±0.0110 ), and can be reconstructed with a total latency of 170 ms that is sufficient for real-time MR-guided abdominal radiotherapy.

A data-driven semantic segmentation model for direct cardiac functional analysis based on undersampled radial MR cine series

PURPOSE

Image acquisition and subsequent manual analysis of cardiac cine MRI is time-consuming. The purpose of this study was to train and evaluate a 3D artificial neural network for semantic segmentation of radially undersampled cardiac MRI to accelerate both scan time and postprocessing.

METHODS

A database of Cartesian short-axis MR images of the heart (148,500 images, 484 examinations) was assembled from an openly accessible database and radial undersampling was simulated. A 3D U-Net architecture was pretrained for segmentation of undersampled spatiotemporal cine MRI. Transfer learning was then performed using samples from a second database, comprising 108 non-Cartesian radial cine series of the midventricular myocardium to optimize the performance for authentic data. The performance was evaluated for different levels of undersampling by the Dice similarity coefficient (DSC) with respect to reference labels, as well as by deriving ventricular volumes and myocardial masses.

RESULTS

Without transfer learning, the pretrained model performed moderately on true radial data [maximum number of projections tested, P = 196; DSC = 0.87 (left ventricle), DSC = 0.76 (myocardium), and DSC =0.64 (right ventricle)]. After transfer learning with authentic data, the predictions achieved human level even for high undersampling rates (P = 33, DSC = 0.95, 0.87, and 0.93) without significant difference compared with segmentations derived from fully sampled data.

CONCLUSION

A 3D U-Net architecture can be used for semantic segmentation of radially undersampled cine acquisitions, achieving a performance comparable with human experts in fully sampled data. This approach can jointly accelerate time-consuming cine image acquisition and cumbersome manual image analysis.

Real-time 3D motion estimation from undersampled MRI using multi-resolution neural networks

Purpose: To enable real‐time adaptive magnetic resonance imaging–guided radiotherapy (MRIgRT) by obtaining time‐resolved three‐dimensional (3D) deformation vector fields (DVFs) with high spatiotemporal resolution and low latency (<500 ms).

Theory and Methods: Respiratory‐resolved T1‐weighted 4D‐MRI of 27 patients with lung cancer were acquired using a golden‐angle radial stack‐of‐stars readout. A multiresolution convolutional neural network (CNN) called TEMPEST was trained on up to 32× retrospectively undersampled MRI of 17 patients, reconstructed with a nonuniform fast Fourier transform, to learn optical flow DVFs. TEMPEST was validated using 4D respiratory‐resolved MRI, a digital phantom, and a physical motion phantom. The time‐resolved motion estimation was evaluated in‐vivo using two volunteer scans, acquired on a hybrid MR‐scanner with integrated linear accelerator. Finally, we evaluated the model robustness on a publicly‐available four‐dimensional computed tomography (4D‐CT) dataset.

Results: TEMPEST produced accurate DVFs on respiratory‐resolved MRI at 20‐fold acceleration, with the average end‐point‐error <2 mm, both on respiratory‐sorted MRI and on a digital phantom. TEMPEST estimated accurate time‐resolved DVFs on MRI of a motion phantom, with an error <2 mm at 28× undersampling. On two volunteer scans, TEMPEST accurately estimated motion compared to the self‐navigation signal using 50 spokes per dynamic (366× undersampling). At this undersampling factor, DVFs were estimated within 200 ms, including MRI acquisition. On fully sampled CT data, we achieved a target registration error of 1.87±1.65 mm without retraining the model.

Conclusion: A CNN trained on undersampled MRI produced accurate 3D DVFs with high spatiotemporal resolution for MRIgRT.

Nonrigid 3D motion estimation at high temporal resolution from prospectively undersampled k-space data using low-rank MR-MOTUS

Purpose

With the recent introduction of the MR‐LINAC, an MR‐scanner combined with a radiotherapy LINAC, MR‐based motion estimation has become of increasing interest to (retrospectively) characterize tumor and organs‐at‐risk motion during radiotherapy. To this extent, we introduce low‐rank MR‐MOTUS, a framework to retrospectively reconstruct time‐resolved nonrigid 3D+t motion fields from a single low‐resolution reference image and prospectively undersampled k‐space data acquired during motion.

Theory

Low‐rank MR‐MOTUS exploits spatiotemporal correlations in internal body motion with a low‐rank motion model, and inverts a signal model that relates motion fields directly to a reference image and k‐space data. The low‐rank model reduces the degrees‐of‐freedom, memory consumption, and reconstruction times by assuming a factorization of space‐time motion fields in spatial and temporal components.

Methods

Low‐rank MR‐MOTUS was employed to estimate motion in 2D/3D abdominothoracic scans and 3D head scans. Data were acquired using golden‐ratio radial readouts. Reconstructed 2D and 3D respiratory motion fields were, respectively, validated against time‐resolved and respiratory‐resolved image reconstructions, and the head motion against static image reconstructions from fully sampled data acquired right before and right after the motion.

Results

Results show that 2D+t respiratory motion can be estimated retrospectively at 40.8 motion fields per second, 3D+t respiratory motion at 7.6 motion fields per second and 3D+t head‐neck motion at 9.3 motion fields per second. The validations show good consistency with image reconstructions.

Conclusions

The proposed framework can estimate time‐resolved nonrigid 3D motion fields, which allows to characterize drifts and intra and inter‐cycle patterns in breathing motion during radiotherapy, and could form the basis for real‐time MR‐guided radiotherapy.