Phase contrast MRI with minimized background phase errors.

PURPOSE: Phase contrast MRI (PC-MRI) is used clinically to measure velocities in the body, but systematic background phase errors caused by magnetic field imperfections corrupt the velocity measurements with offsets that limit clinical utility. This work aims to minimize systematic background phase errors in PC-MRI, thereby maximizing the accuracy of velocity measurements. METHODS: The MRI scanner's background phase errors from eddy currents and mechanical oscillations were modeled using the gradient impulse response function (GIRF). Gradient waveforms were then numerically optimized using the GIRF to create pulse sequences that minimize the background phase errors. The pulse sequences were tested in a static phantom where the predicted response could be compared directly to the measured background velocity. The optimized acquisitions were then tested in human subjects, where flow rates and background errors were compared to conventional PC-MRI. RESULTS: When using the GIRF-optimized gradient waveforms, the predicted background phase was within 0.6 [95% CI = -3.4, 5.4] mm/s of the measured background phase in a static phantom. Excellent agreement was seen for in vivo blood flow values (flow rate agreement r 2 $$ {r}^2 $$ = 0.96), and the background phase was reduced by 78.8 ± $$ \pm $$ 18.7%. CONCLUSION: This work shows that using a GIRF to model the effects of magnetic field imperfections combined with numerically optimized gradient waveforms enables PC-MRI waveforms to be designed to produce a minimal background phase in the most time-efficient manner. The methodology could be adapted for other MRI sequences where similar magnetic field errors affect measurements.

Pre-excitation gradients for eddy current nulled convex optimized diffusion encoding (Pre-ENCODE).

PURPOSE: To evaluate the use of pre-excitation gradients for eddy current-nulled convex optimized diffusion encoding (Pre-ENCODE) to mitigate eddy current-induced image distortions in diffusion-weighted MRI (DWI). METHODS: DWI sequences using monopolar (MONO), ENCODE, and Pre-ENCODE were evaluated in terms of the minimum achievable echo time (TE min $$ {}_{\mathrm{min}} $$ ) and eddy current-induced image distortions using simulations, phantom experiments, and in vivo DWI in volunteers ( N = 6 $$ N=6 $$ ). RESULTS: Pre-ENCODE provided a shorter TE min $$ {}_{\mathrm{min}} $$ than MONO (71.0 ± $$ \pm $$ 17.7ms vs. 77.6 ± $$ \pm $$ 22.9ms) and ENCODE (71.0 ± $$ \pm $$ 17.7ms vs. 86.2 ± $$ \pm $$ 14.2ms) in 100 % $$ \% $$ of the simulated cases for a commercial 3T MRI system with b-values ranging from 500 to 3000 s/mm 2 $$ {}^2 $$ and in-plane spatial resolutions ranging from 1.0 to 3.0mm 2 $$ {}^2 $$ . Image distortion was estimated by intravoxel signal variance between diffusion encoding directions near the phantom edges and was significantly lower with Pre-ENCODE than with MONO (10.1 % $$ \% $$ vs. 22.7 % $$ \% $$ , p = 6 - 5 $$ p={6}^{-5} $$ ) and comparable to ENCODE (10.1 % $$ \% $$ vs. 10.4 % $$ \% $$ , p = 0 . 12 $$ p=0.12 $$ ). In vivo measurements of apparent diffusion coefficients were similar in global brain pixels (0.37 [0.28,1.45] × 1 0 - 3 $$ \times 1{0}^{-3} $$ mm 2 $$ {}^2 $$ /s vs. 0.38 [0.28,1.45] × 1 0 - 3 $$ \times 1{0}^{-3} $$ mm 2 $$ {}^2 $$ /s, p = 0 . 25 $$ p=0.25 $$ ) and increased in edge brain pixels (0.80 [0.17,1.49] × 1 0 - 3 $$ \times 1{0}^{-3} $$ mm 2 $$ {}^2 $$ /s vs. 0.70 [0.18,1.48] × 1 0 - 3 $$ \times 1{0}^{-3} $$ mm 2 $$ {}^2 $$ /s, p = 0 . 02 $$ p=0.02 $$ ) for MONO compared to Pre-ENCODE. CONCLUSION: Pre-ENCODE mitigated eddy current-induced image distortions for diffusion imaging with a shorter TE min $$ {}_{\mathrm{min}} $$ than MONO and ENCODE.

Virtual injections using 4D flow MRI with displacement corrections and constrained probabilistic streamlines.

PURPOSE: Streamlines from 4D-flow MRI have been used clinically for intracranial blood-flow tracking. However, deterministic and stochastic errors degrade streamline quality. The purpose of this study is to integrate displacement corrections, probabilistic streamlines, and novel fluid constraints to improve selective blood-flow tracking and emulate "virtual bolus injections." METHODS: Both displacement artifacts (deterministic) and velocity noise (stochastic) inherently occur during phase-contrast MRI acquisitions. Here, two displacement correction methods, single-step and iterative, were tested in silico with simulated displacements and were compared with ground-truth velocity fields. Next, the effects of combining displacement corrections and constrained probabilistic streamlines were performed in 10 healthy volunteers using time-averaged 4D-flow data. Measures of streamline length and depth into vasculature were then compared with streamlines generated with no corrections and displacement correction alone using one-way repeated-measures analysis of variance and Friedman's tests. Finally, virtual injections with improved streamlines were generated for three intracranial pathology cases. RESULTS: Iterative displacement correction outperformed the single-step method in silico. In volunteers, the combination of displacement corrections and constrained probabilistic streamlines allowed for significant improvements in streamline length and increased the number of streamlines entering the circle of Willis relative to streamlines with no corrections and displacement correction alone. In the pathology cases, virtual injections with improved streamlines were qualitatively similar to dynamic arterial spin labeling images and allowed for forward/reverse selective flow tracking to characterize cerebrovascular malformations. CONCLUSION: Virtual injections with improved streamlines from 4D-flow MRI allow for flexible, robust, intracranial flow tracking.

4D flow cardiovascular magnetic resonance recovery profiles following pulmonary endarterectomy in chronic thromboembolic pulmonary hypertension.

BACKGROUND: Four-dimensional flow cardiovascular magnetic resonance imaging (4D flow CMR) allows comprehensive assessment of pulmonary artery (PA) flow dynamics. Few studies have characterized longitudinal changes in pulmonary flow dynamics and right ventricular (RV) recovery following a pulmonary endarterectomy (PEA) for patients with chronic thromboembolic pulmonary hypertension (CTEPH). This can provide novel insights of RV and PA dynamics during recovery. We investigated the longitudinal trajectory of 4D flow metrics following a PEA including velocity, vorticity, helicity, and PA vessel wall stiffness. METHODS: Twenty patients with CTEPH underwent pre-PEA and > 6 months post-PEA CMR imaging including 4D flow CMR; right heart catheter measurements were performed in 18 of these patients. We developed a semi-automated pipeline to extract integrated 4D flow-derived main, left, and right PA (MPA, LPA, RPA) volumes, velocity flow profiles, and secondary flow profiles. We focused on secondary flow metrics of vorticity, volume fraction of positive helicity (clockwise rotation), and the helical flow index (HFI) that measures helicity intensity. RESULTS: Mean PA pressures (mPAP), total pulmonary resistance (TPR), and normalized RV end-systolic volume (RVESV) decreased significantly post-PEA (P < 0.002). 4D flow-derived PA volumes decreased (P < 0.001) and stiffness, velocity, and vorticity increased (P < 0.01) post-PEA. Longitudinal improvements from pre- to post-PEA in mPAP were associated with longitudinal decreases in MPA area (r = 0.68, P = 0.002). Longitudinal improvements in TPR were associated with longitudinal increases in the maximum RPA HFI (r=-0.85, P < 0.001). Longitudinal improvements in RVESV were associated with longitudinal decreases in MPA fraction of positive helicity (r = 0.75, P = 0.003) and minimum MPA HFI (r=-0.72, P = 0.005). CONCLUSION: We developed a semi-automated pipeline for analyzing 4D flow metrics of vessel stiffness and flow profiles. PEA was associated with changes in 4D flow metrics of PA flow profiles and vessel stiffness. Longitudinal analysis revealed that PA helicity was associated with pulmonary remodeling and RV reverse remodeling following a PEA.

Reproducibility of global and segmental myocardial strain using cine DENSE at 3 T: a multicenter cardiovascular magnetic resonance study in healthy subjects and patients with heart disease.

BACKGROUND: While multiple cardiovascular magnetic resonance (CMR) methods provide excellent reproducibility of global circumferential and global longitudinal strain, achieving highly reproducible segmental strain is more challenging. Previous single-center studies have demonstrated excellent reproducibility of displacement encoding with stimulated echoes (DENSE) segmental circumferential strain. The present study evaluated the reproducibility of DENSE for measurement of whole-slice or global circumferential (Ecc), longitudinal (Ell) and radial (Err) strain, torsion, and segmental Ecc at multiple centers. METHODS: Six centers participated and a total of 81 subjects were studied, including 60 healthy subjects and 21 patients with various types of heart disease. CMR utilized 3 T scanners, and cine DENSE images were acquired in three short-axis planes and in the four-chamber long-axis view. During one imaging session, each subject underwent two separate DENSE scans to assess inter-scan reproducibility. Each subject was taken out of the scanner and repositioned between the scans. Intra-user, inter-user-same-site, inter-user-different-site, and inter-user-Human-Deep-Learning (DL) comparisons assessed the reproducibility of different users analyzing the same data. Inter-scan comparisons assessed the reproducibility of DENSE from scan to scan. The reproducibility of whole-slice or global Ecc, Ell and Err, torsion, and segmental Ecc were quantified using Bland-Altman analysis, the coefficient of variation (CV), and the intraclass correlation coefficient (ICC). CV was considered excellent for CV ≤ 10%, good for 10% < CV ≤ 20%, fair for 20% < CV ≤ 40%, and poor for CV > 40. ICC values were considered excellent for ICC > 0.74, good for ICC 0.6 < ICC ≤ 0.74, fair for ICC 0.4 < ICC ≤ 0.59, poor for ICC < 0.4. RESULTS: Based on CV and ICC, segmental Ecc provided excellent intra-user, inter-user-same-site, inter-user-different-site, inter-user-Human-DL reproducibility and good-excellent inter-scan reproducibility. Whole-slice Ecc and global Ell provided excellent intra-user, inter-user-same-site, inter-user-different-site, inter-user-Human-DL and inter-scan reproducibility. The reproducibility of torsion was good-excellent for all comparisons. For whole-slice Err, CV was in the fair-good range, and ICC was in the good-excellent range. CONCLUSIONS: Multicenter data show that 3 T CMR DENSE provides highly reproducible whole-slice and segmental Ecc, global Ell, and torsion measurements in healthy subjects and heart disease patients.

CG-SENSE revisited: Results from the first ISMRM reproducibility challenge.

PURPOSE: The aim of this work is to shed light on the issue of reproducibility in MR image reconstruction in the context of a challenge. Participants had to recreate the results of "Advances in sensitivity encoding with arbitrary k-space trajectories" by Pruessmann et al. METHODS: The task of the challenge was to reconstruct radially acquired multicoil k-space data (brain/heart) following the method in the original paper, reproducing its key figures. Results were compared to consolidated reference implementations created after the challenge, accounting for the two most common programming languages used in the submissions (Matlab/Python). RESULTS: Visually, differences between submissions were small. Pixel-wise differences originated from image orientation, assumed field-of-view, or resolution. The reference implementations were in good agreement, both visually and in terms of image similarity metrics. DISCUSSION AND CONCLUSION: While the description level of the published algorithm enabled participants to reproduce CG-SENSE in general, details of the implementation varied, for example, density compensation or Tikhonov regularization. Implicit assumptions about the data lead to further differences, emphasizing the importance of sufficient metadata accompanying open datasets. Defining reproducibility quantitatively turned out to be nontrivial for this image reconstruction challenge, in the absence of ground-truth results. Typical similarity measures like NMSE of SSIM were misled by image intensity scaling and outlier pixels. Thus, to facilitate reproducibility, researchers are encouraged to publish code and data alongside the original paper. Future methodological papers on MR image reconstruction might benefit from the consolidated reference implementations of CG-SENSE presented here, as a benchmark for methods comparison.

A gradient optimization toolbox for general purpose time-optimal MRI gradient waveform design.

PURPOSE: To introduce and demonstrate a software library for time-optimal gradient waveform optimization with a wide range of applications. The software enables direct on-the-fly gradient waveform design on the scanner hardware for multiple vendors. METHODS: The open-source gradient optimization (GrOpt) toolbox was implemented in C with both Matlab and Python wrappers. The toolbox enables gradient waveforms to be generated based on a set of constraints that define the features and encodings for a given acquisition. The GrOpt optimization routine is based on the alternating direction method of multipliers (ADMM). Additional constraints enable error corrections to be added, or patient comfort and safety to be adressed. A range of applications and compute speed metrics are analyzed. Finally, the method is implemented and tested on scanners from different vendors. RESULTS: Time-optimal gradient waveforms for different pulse sequences and the constraints that define them are shown. Additionally, the ability to add, arbitrary motion (gradient moment) compensation or limit peripheral nerve stimulation is demonstrated. There exists a trade-off between computation time and gradient raster time, but it was observed that acceptable gradient waveforms could be generated in 1-40 ms. Gradient waveforms generated and run on the different scanners were functionally equivalent, and the images were comparable. CONCLUSIONS: GrOpt is an open source toolbox that enables on-the-fly optimization of gradient waveform design, subject to a set of defined constraints. GrOpt was presented for a range of imaging applications, analyzed in terms of computational complexity, and implemented to run on the scanner for a multi-vendor demonstration.

Optimization methods for magnetic resonance imaging gradient waveform design.

The development and implementation of novel MRI pulse sequences remains challenging and laborious. Gradient waveforms are typically designed using a combination of analytical and ad hoc methods to construct each gradient waveform axis independently. This strategy makes coding the pulse sequence complicated, in addition to being time inefficient. As a consequence, nearly all commercial MRI pulse sequences fail to maximize use of the available gradient hardware or efficiently mitigate physiological effects. This results in expensive MRI systems that underperform relative to their inherent hardware capabilities. To address this problem, a software solution is proposed that incorporates numerical optimization methods into MRI pulse sequence programming. Examples are shown for rotational variant vs. invariant waveform designs, acceleration nulled velocity encoding gradients, and mitigation of peripheral nerve stimulation for diffusion encoding. The application of optimization methods to MRI pulse sequence design incorporates gradient hardware limits and the prescribed MRI protocol parameters (e.g. field-of-view, resolution, gradient moments, and/or b-value) to simultaneously construct time-optimal gradient waveforms. In many cases, the resulting constrained gradient waveform design problem is convex and can be solved on-the-fly on the MRI scanner. The result is a set of multi-axis time-optimal gradient waveforms that satisfy the design constraints, thereby increasing SNR-efficiency. These optimization methods can also be used to mitigate imaging artifacts (e.g. eddy currents) or account for peripheral nerve stimulation. The result of the optimization method is to enable easier pulse sequence gradient waveform design and permit on-the-fly implementation for a range of MRI pulse sequences.

4D Flow MR Imaging to Improve Microwave Ablation Prediction Models: A Feasibility Study in an In Vivo Porcine Liver.

PURPOSE: To characterize the effect of hepatic vessel flow using 4-dimensional (4D) flow magnetic resonance (MR) imaging and correlate their effect on microwave ablation volumes in an in vivo non-cirrhotic porcine liver model. MATERIALS AND METHODS: Microwave ablation antennas were placed under ultrasound guidance in each liver lobe of swine (n = 3 in each animal) for a total of 9 ablations. Pre- and post-ablation 4D flow MR imaging was acquired to quantify flow changes in the hepatic vasculature. Flow measurements, along with encompassed vessel size and vessel-antenna spacing, were then correlated with final ablation volume from segmented MR images. RESULTS: The linear regression model demonstrated that the preablation measurement of encompassed hepatic vein size (ß = -0.80 ± 0.25, 95% confidence interval [CI] -1.15 to -0.22; P = .02) was significantly correlated to final ablation zone volume. The addition of hepatic vein flow rate found via 4D flow MRI (ß = -0.83 ± 0.65, 95% CI -2.50 to 0.84; P = .26), and distance from antenna to hepatic vein (ß = 0.26 ± 0.26, 95% CI -0.40 to 0.92; P = .36) improved the model accuracy but not significantly so (multivariate adjusted R2 = 0.70 vs univariate (vessel size) adjusted R2 = 0.63, P = .24). CONCLUSIONS: Hepatic vein size in an encompassed ablation zone was found to be significantly correlated with final ablation zone volume. Although the univariate 4D flow MR imaging-acquired measurements alone were not found to be statistically significant, its addition to hepatic vein size improved the accuracy of the ablation volume regression model. Pre-ablation 4D flow MR imaging of the liver may assist in prospectively optimizing thermal ablation treatment.

Time-optimized 4D phase contrast MRI with real-time convex optimization of gradient waveforms and fast excitation methods.

PURPOSE: To shorten 4D flow acquisitions by shortening TRs with fast RF pulses and gradient waveforms. Real-time convex optimization is used to generate these gradients waveforms on the scanner. THEORY AND METHODS: RF and slab-select waveforms were shortened with a minimum phase SLR excitation and the time-optimal variable-rate selective excitation method. Real-time convex optimization was used to shorten bipolar and spoiler gradients by finding the shortest gradient waveforms that satisfied constraints on scan parameters, gradient hardware, M0 , M1 , and peripheral nerve stimulation. Waveforms were calculated and TE and/or TR values were compared for a range of scan parameters and compared to a conventional 4D flow sequence. The method was tested in flow phantoms, and in the aorta and neurovasculature of volunteers (N = 10). Additionally, eddy current error was measured in a large phantom. RESULTS: TEs and TRs were shortened by 21-32% and 20-34%, respectively, compared to the conventional sequence over a range of scan parameters. Bland-Altman analysis of 2 flow phantom configurations showed flow rate bias of 0.3 mL/s and limits of agreement (LOA) of [-6.9, 7.5] mL/s for a cardiac phantom and a bias of -0.1 mL/s with LOA = [-0.4, 0.2] mL/s for a neuro phantom. Similar agreement was also seen for flow measurements in volunteers (bias = -1.0 and -0.1 mL/s, LOA = [-34.9, 33.0] and [-0.7, 0.6] mL/s). Measured eddy currents were 39% larger with the CVX + mpVERSE method. CONCLUSION: The real-time optimized 4D flow gradients and fast slab-selection excitation methods produced up to 34% faster TRs with excellent flow measurement agreement compared to a conventional 4D flow sequence.

Velocity reconstruction with nonconvex optimization for low-velocity-encoding phase-contrast MRI.

PURPOSE: To introduce and demonstrate a nonconvex optimization method for reconstructing velocity data from low-velocity-encoding (Venc ) phase-contrast MRI data. THEORY AND METHODS: Solving for velocity values from phase-contrast MRI data was formulated as a nonconvex optimization problem. Weighting was added to account for intravoxel dephasing, and a Laplacian-based regularization was used to account for residual velocity aliasing. The reconstruction was tested with two low-Venc schemes: dual-Venc and a multidirectional high-moment encoding. The reconstruction method was tested in a digital simulation, in flow phantoms, and in healthy volunteers (N = 5). RESULTS: The nonconvex-optimization reconstruction velocity error was lower than the conventional reconstruction in simulations (4.6 versus 3.0 cm/s for multidirectional high moment, 8.3 versus 3.8 cm/s for dual-Venc ) and in flow phantoms (23.9 versus 5.9 cm/s for multidirectional high moment, 15.2 versus 6.4 cm/s for dual-Venc ). Qualitative assessment of velocity fields in all experiments, including healthy volunteers, showed decreased apparent noise in the velocity fields and fewer phase wraps. No additional velocity bias in measured velocities was seen in volunteers with the proposed method. CONCLUSIONS: The proposed nonconvex-optimization reconstruction method incorporates additional information to solve for velocities when using any type of low-Venc (high-moment) acquisition. The method reduces the amount of residual phase aliasing, and decreases velocity errors that result from intravoxel dephasing. These improvements allow for more robust acquisitions, and for Venc to be lowered 2 to 4 times relative to conventional acquisitions, thereby increasing the velocity-to-noise ratio. Magn Reson Med, 2017. © 2017 International Society for Magnetic Resonance in Medicine. Magn Reson Med 80:42-52, 2018. © 2017 International Society for Magnetic Resonance in Medicine.

Accelerating 4D flow MRI by exploiting vector field divergence regularization.

PURPOSE: To improve velocity vector field reconstruction from undersampled four-dimensional (4D) flow MRI by penalizing divergence of the measured flow field. THEORY AND METHODS: Iterative image reconstruction in which magnitude and phase are regularized separately in alternating iterations was implemented. The approach allows incorporating prior knowledge of the flow field being imaged. In the present work, velocity data were regularized to reduce divergence, using either divergence-free wavelets (DFW) or a finite difference (FD) method using the ℓ1-norm of divergence and curl. The reconstruction methods were tested on a numerical phantom and in vivo data. Results of the DFW and FD approaches were compared with data obtained with standard compressed sensing (CS) reconstruction. RESULTS: Relative to standard CS, directional errors of vector fields and divergence were reduced by 55-60% and 38-48% for three- and six-fold undersampled data with the DFW and FD methods. Velocity vector displays of the numerical phantom and in vivo data were found to be improved upon DFW or FD reconstruction. CONCLUSION: Regularization of vector field divergence in image reconstruction from undersampled 4D flow data is a valuable approach to improve reconstruction accuracy of velocity vector fields.

Phase unwrapping in 4D MR flow with a 4D single-step laplacian algorithm.

PURPOSE: To introduce and demonstrate a method for unwrapping 4D flow data by utilizing continuity constraints in all four available dimensions. MATERIALS AND METHODS: A Laplacian-based algorithm for unwrapping phase data was expanded to unwrap along the temporal dimension in addition to all three spatial dimensions. The method was tested on simulated blood flow under varying vessel diameters and velocity encoding (Venc ) values. The algorithm was also tested in the aorta of five volunteers, with wrapped data acquired with Venc = 80 cm/s and 40 cm/s. Unwrapping performance was measured visually and in comparison to a high Venc reference free of phase wrapping. Ten patients with aortic coarctations with clinical Venc values and lower-Venc reconstructions were corrected and scored by blinded reviewers on a 0-3 scale. RESULTS: Simulated data were completely unwrapped for most clinically relevant levels of velocity aliasing using the proposed method. In vivo data in the aorta were completely unwrapped for cases of moderate wrapping (Venc = 80 cm/s, peak velocities = ∼160 cm/s), while residual aliasing remained for the more considerably aliased datasets (Venc = 40 cm/s). Improvements were seen in scoring (mean score improved by 1.1 and 2.2 for clinical and low-Venc datasets, respectively) by the blinded reviewers in the patient cohort for both standard and low-Venc reconstructions. CONCLUSION: A computationally fast, fully automated, easy to use, and parameter-free single-step method for unwrapping 4D flow data is shown to be effective for use in most common clinical occurrences of velocity aliasing.

Myocardial Segmentation of Tagged Magnetic Resonance Images with Transfer Learning Using Generative Cine-To-Tagged Dataset Transformation.

The use of deep learning (DL) segmentation in cardiac MRI has the potential to streamline the radiology workflow, particularly for the measurement of myocardial strain. Recent efforts in DL motion tracking models have drastically reduced the time needed to measure the heart's displacement field and the subsequent myocardial strain estimation. However, the selection of initial myocardial reference points is not automated and still requires manual input from domain experts. Segmentation of the myocardium is a key step for initializing reference points. While high-performing myocardial segmentation models exist for cine images, this is not the case for tagged images. In this work, we developed and compared two novel DL models (nnU-net and Segmentation ResNet VAE) for the segmentation of myocardium from tagged CMR images. We implemented two methods to transform cardiac cine images into tagged images, allowing us to leverage large public annotated cine datasets. The cine-to-tagged methods included (i) a novel physics-driven transformation model, and (ii) a generative adversarial network (GAN) style transfer model. We show that pretrained models perform better (+2.8 Dice coefficient percentage points) and converge faster (6×) than models trained from scratch. The best-performing method relies on a pretraining with an unpaired, unlabeled, and structure-preserving generative model trained to transform cine images into their tagged-appearing equivalents. Our state-of-the-art myocardium segmentation network reached a Dice coefficient of 0.828 and 95th percentile Hausdorff distance of 4.745 mm on a held-out test set. This performance is comparable to existing state-of-the-art segmentation networks for cine images.

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.

Hemodynamic Effects of Entry and Exit Tear Size in Aortic Dissection Evaluated with In Vitro Magnetic Resonance Imaging and Fluid-Structure Interaction Simulation.

Understanding the complex interplay between morphologic and hemodynamic features in aortic dissection is critical for risk stratification and for the development of individualized therapy. This work evaluates the effects of entry and exit tear size on the hemodynamics in type B aortic dissection by comparing fluid-structure interaction (FSI) simulations with in vitro 4D-flow magnetic resonance imaging (MRI). A baseline patient-specific 3D-printed model and two variants with modified tear size (smaller entry tear, smaller exit tear) were embedded into a flow- and pressure-controlled setup to perform MRI as well as 12-point catheter-based pressure measurements. The same models defined the wall and fluid domains for FSI simulations, for which boundary conditions were matched with measured data. Results showed exceptionally well matched complex flow patterns between 4D-flow MRI and FSI simulations. Compared to the baseline model, false lumen flow volume decreased with either a smaller entry tear (-17.8 and -18.5 %, for FSI simulation and 4D-flow MRI, respectively) or smaller exit tear (-16.0 and -17.3 %). True to false lumen pressure difference (initially 11.0 and 7.9 mmHg, for FSI simulation and catheter-based pressure measurements, respectively) increased with a smaller entry tear (28.9 and 14.6 mmHg), and became negative with a smaller exit tear (-20.6 and -13.2 mmHg). This work establishes quantitative and qualitative effects of entry or exit tear size on hemodynamics in aortic dissection, with particularly notable impact observed on FL pressurization. FSI simulations demonstrate acceptable qualitative and quantitative agreement with flow imaging, supporting its deployment in clinical studies.

Hemodynamic effects of entry and exit tear size in aortic dissection evaluated with in vitro magnetic resonance imaging and fluid-structure interaction simulation.

Understanding the complex interplay between morphologic and hemodynamic features in aortic dissection is critical for risk stratification and for the development of individualized therapy. This work evaluates the effects of entry and exit tear size on the hemodynamics in type B aortic dissection by comparing fluid-structure interaction (FSI) simulations with in vitro 4D-flow magnetic resonance imaging (MRI). A baseline patient-specific 3D-printed model and two variants with modified tear size (smaller entry tear, smaller exit tear) were embedded into a flow- and pressure-controlled setup to perform MRI as well as 12-point catheter-based pressure measurements. The same models defined the wall and fluid domains for FSI simulations, for which boundary conditions were matched with measured data. Results showed exceptionally well matched complex flow patterns between 4D-flow MRI and FSI simulations. Compared to the baseline model, false lumen flow volume decreased with either a smaller entry tear (- 17.8 and - 18.5%, for FSI simulation and 4D-flow MRI, respectively) or smaller exit tear (- 16.0 and - 17.3%). True to false lumen pressure difference (initially 11.0 and 7.9 mmHg, for FSI simulation and catheter-based pressure measurements, respectively) increased with a smaller entry tear (28.9 and 14.6 mmHg), and became negative with a smaller exit tear (- 20.6 and - 13.2 mmHg). This work establishes quantitative and qualitative effects of entry or exit tear size on hemodynamics in aortic dissection, with particularly notable impact observed on FL pressurization. FSI simulations demonstrate acceptable qualitative and quantitative agreement with flow imaging, supporting its deployment in clinical studies.

Using synthetic data generation to train a cardiac motion tag tracking neural network.

A CNN based method for cardiac MRI tag tracking was developed and validated. A synthetic data simulator was created to generate large amounts of training data using natural images, a Bloch equation simulation, a broad range of tissue properties, and programmed ground-truth motion. The method was validated using both an analytical deforming cardiac phantom and in vivo data with manually tracked reference motion paths. In the analytical phantom, error was investigated relative to SNR, and accurate results were seen for SNR>10 (displacement error <0.3 mm). Excellent agreement was seen in vivo for tag locations (mean displacement difference = -0.02 pixels, 95% CI [-0.73, 0.69]) and calculated cardiac circumferential strain (mean difference = 0.006, 95% CI [-0.012, 0.024]). Automated tag tracking with a CNN trained on synthetic data is both accurate and precise.

Arbitrary Point Tracking with Machine Learning to Measure Cardiac Strains in Tagged MRI.

Cardiac tagged MR images allow for deformation fields to be measured in the heart by tracking the motion of tag lines throughout the cardiac cycle. Machine learning (ML) algorithms enable accurate and robust tracking of tag lines. Herein, the use of a massive synthetic physics-driven training dataset with known ground truth was used to train an ML network to enable tracking any number of points at arbitrary positions rather than anchored to the tag lines themselves. The tag tracking and strain calculation methods were investigated in a computational deforming cardiac phantom with known (ground truth) strain values. This enabled both tag tracking and strain accuracy to be characterized for a range of image acquisition and tag tracking parameters. The methods were also tested on in vivo volunteer data. Median tracking error was <0.26mm in the computational phantom, and strain measurements were improved in vivo when using the arbitrary point tracking for a standard clinical protocol.

On the impact of vessel wall stiffness on quantitative flow dynamics in a synthetic model of the thoracic aorta.

Aortic wall stiffening is a predictive marker for morbidity in hypertensive patients. Arterial pulse wave velocity (PWV) correlates with the level of stiffness and can be derived using non-invasive 4D-flow magnetic resonance imaging (MRI). The objectives of this study were twofold: to develop subject-specific thoracic aorta models embedded into an MRI-compatible flow circuit operating under controlled physiological conditions; and to evaluate how a range of aortic wall stiffness impacts 4D-flow-based quantification of hemodynamics, particularly PWV. Three aorta models were 3D-printed using a novel photopolymer material at two compliant and one nearly rigid stiffnesses and characterized via tensile testing. Luminal pressure and 4D-flow MRI data were acquired for each model and cross-sectional net flow, peak velocities, and PWV were measured. In addition, the confounding effect of temporal resolution on all metrics was evaluated. Stiffer models resulted in increased systolic pressures (112, 116, and 133 mmHg), variations in velocity patterns, and increased peak velocities, peak flow rate, and PWV (5.8-7.3 m/s). Lower temporal resolution (20 ms down to 62.5 ms per image frame) impacted estimates of peak velocity and PWV (7.31 down to 4.77 m/s). Using compliant aorta models is essential to produce realistic flow dynamics and conditions that recapitulated in vivo hemodynamics.