Objective: Magnetic Resonance Elastography (MRE) is a non-invasive imaging technique for mapping biomechanical properties of in vivo tissue, including shear wave speed (SWS), but involves intrinsically slow data acquisition and an ill-posed wave inversion. Instead of relying on handcrafted image priors, we propose a data-driven approach jointly combining image reconstruction and MRE inversion for robust SWS estimation from undersampled k-space data. Methods: Our physics-informed reconstruction framework comprises two blocks: a model-based neural network (NN)-regularized reconstruction module and a phase-gradient inversion (k-MDEV) calculating SWS from the reconstructed images. Concatenating both blocks yields an end-to-end trainable method to estimate SWS directly from measured k-space data. We evaluated the method on retrospectively highly undersampled brain MRE data and compared it to a total variation (TV) minimization-based approach. We assessed the impact of end-to-end training (qualitative images and SWS maps as targets) versus pre-training (qualitative images as targets) and applied the method also to in vivo data. Results: Our approach significantly reduces NRMSE by 30% compared to TV. End-to-end training improves SWS estimation over separate image reconstruction and SWS calculation. Conclusion: Accurate SWS quantification is possible at acceleration factors up to 19. Our method significantly outperforms TV, highlighting the need for data-driven regularization in this challenging MR problem. Further, our approach successfully generalizes to in vivo data. Significance: We present the first end-to-end trainable MRE reconstruction method for estimating SWS maps directly from k-space. NN-based reconstruction can enable rapid stiffness mapping for dynamic studies, functional imaging, and real-time clinical feedback.
Figures & tables
Fig. 1 : Yellow block : Image reconstruction block Recθ : Starting from undersampled spiral k -space data, an initial MR image estimate is computed by approximating the pseudo-inverse solution ( purple block ). Grey block : An image is fed into a CNN, whose output is refined via a DC step according to ( 10 ). CNN and DC steps are repeated T times. Pre-training minimizes the loss between reconstructed and target images. Orange block : k -MDEV: The phase is extracted, unwrapped, and temporally Fourier transformed. The harmonic component of the wave field is filtered and inverted to yield SWS maps, as described in ( 13 ). Green block : The entire reconstruction network, comprising the image reconstruction and the k -MDEV module, is fine-tuned by minimizing a combined loss that uses reference images and reference SWS maps.
Fig. 2 : Training and validation loss of complex MR images and SWS maps over 100 epochs with T=6 . Perpendicular loss ( 16 ) was used for images, RMSE combined with SSIM for SWS maps ( 17 ). The first 50 epochs correspond to pre-training with the qualitative images as reference, followed by 50 epochs of fine-tuning with α=0.1 according to ( 15 ).
Fig. 3 : SSIM, NRMSE, and PSNR computed on the test set with pre-trained and fine-tuned model ( T=6 , α=0.1 ) applied. The metrics were calculated on qualitative images and SWS with Nmeg=3 . Except for the qualitative images, the fine-tuned model outperformed the pre-trained one.
Fig. 4 : Magnitude images (left), phase images (middle), and resulting SWS maps (right) from reconstructions using various methods on retrospectively undersampled single-spiral data. From top to bottom: Adjoint, iterative SENSE [ 29 ] , the TV minimization-based comparison method, and the proposed method. Fully sampled data (bottom row) serves as reference.
Spirals
Method
SSIM
NRMSE
PSNR (dB)
1
Adjoint
0.75 ± 0.072
0.43 ± 0.054
21.10 ± 2.96
it. SENSE
0.83 ± 0.051
0.23 ± 0.040
26.79 ± 2.69
TV
0.87 ± 0.049
0.20 ± 0.041
27.81 ± 2.58
Proposed
0.93 ± 0.030
0.14 ± 0.033
30.92 ± 2.34
2
Adjoint
0.83 ± 0.055
0.26 ± 0.030
25.37 ± 2.88
it. SENSE
0.92 ± 0.035
0.16 ± 0.036
30.10 ± 2.65
TABLE I : Comparison of quantitative metrics (SSIM, NRMSE, PSNR) computed from SWS maps reconstructed using different methods, applied to simulated undersampled data with 1, 2, and 3 spirals. Masked metrics are shown, using the brain mask as explained in IV-C2 . Best values shown in bold.
Fig. 5 : Comparison between different numbers of unrolled iterations ( T=1 vs T=6 ) as well as different hyper-parameters α≥0 for the loss function ( 15 ) used for fine-tuning the entire end-to-end trainable network. Employing a larger T yields a lower validation error. Additionally, reducing the effect of catastrophic forgetting with respect to the task of image reconstruction is possible by setting α>0 . The proposed parameter combination T=6 and α=0.1 yields a good tradeoff between image- and SWS-accuracy.
T
α
SSIM
NRMSE
PSNR (dB)
Image
1
0.00
0.90 ± 0.041
0.16 ± 0.031
31.45 ± 2.64
1
0.10
0.91 ± 0.042
0.14 ± 0.033
31.69 ± 2.74
1
0.15
0.91 ± 0.041
0.14 ± 0.032
31.75 ± 2.75
1
0.50
0.91 ± 0.041
0.14 ± 0.033
31.90 ± 2.75
6
0.00
0.95 ± 0.030
0.10 ± 0.029
34.97 ± 3.07
6
0.10
0.95 ± 0.029
0.09 ± 0.028
35.34 ± 3.09
TABLE II : Reconstruction metrics for different T and α configurations applied to the 19-fold undersampled test set. SWS maps are calculated for all MEGs and frequencies combined.
Fig. 6 : SWS maps resulting from the application of different methods to in vivo data acquired with one, two or three spirals compared to the target SWS maps of an acquisition with 25 spirals.
To develop a deep-learning method for achieving fast high-resolution MR elastography from highly undersampled data without the need of high-quality training dataset. We first framed the deep neural network representation as a nonlinear extension of the linear subspace model, then used it to represent and reconstruct MRE image repetitions from undersampled k-space data. The network weights were learned using a multi-level k-space consistent loss. To further enhance reconstruction quality, phase-contrast specific magnitude and phase priors were incorporated, including the similarity of anatomical structures and smoothness of wave-induced harmonic displacement. Experiments were conducted using both 3D gradient-echo spiral and multi-slice spin-echo spiral MRE datasets. Compared to the conventional linear subspace-based approaches, the nonlinear network representation method was able to produce superior image reconstruction with suppressed noise and artifacts from a single in-plane spiral arm per MRE repetition (e.g., 2mm isotropic resolution in 1 min with a total R=10), yielding comparable stiffness estimation to the fully sampled data. This work demonstrated the feasibility of using deep network representations to model and reconstruct MRE images from highly-undersampled data, a nonlinear extension of the subspace-based approaches.
Xi Peng
Departments of Radiology, Biomedical Engineering, Electronic and Computer Engineering, and Mechanical Engineering, University of Iowa, Iowa City, IA 52246 · Department of Radiology, University of Iowa, Iowa City, IA, 52246
Magnetic resonance elastography (MRE) is a noninvasive imaging modality for quantifying the viscoelastic properties of soft tissues from shear wave propagation. Recovering the complex-valued shear modulus from measured displacement fields leads to a severely ill-posed inverse problem, particularly in the presence of noise and limited boundary excitations. We investigate adjoint-based optimization, convolutional neural network (CNN) reconstruction, and a hybrid framework combining both approaches. The forward model is based on a scalar form of the modified stationary Stokes system with a complex shear modulus. We establish well-posedness of the forward problem, existence of minimizers, and first-order optimality conditions for the adjoint-based formulation, and implement a nonlinear conjugate-gradient method with Armijo line search. While PDE-constrained optimization can accurately refine coefficient reconstructions, its performance depends strongly on initialization. We therefore construct a two-dimensional CNN that maps complex-valued displacement measurements to spatially varying complex shear modulus fields and provides rapid, informative initial reconstructions. The proposed hybrid method uses the CNN reconstruction to initialize the adjoint-based optimization, yielding faster convergence and improved accuracy. The CNN is trained on coefficient fields containing individual perturbations and tested on both individual and previously unseen combined configurations. Numerical experiments demonstrate that the CNN generalizes to these more challenging configurations, while subsequent PDE-constrained optimization further refines the reconstructed coefficient. These results demonstrate the potential of combining data-driven initialization with physics-based optimization for efficient and accurate MRE reconstruction.
Anwesa Dey, Johann Rudi, Elena Cherkaev
Department of Mathematics, The University of Utah, address USA. · Department of Mathematics, Virginia Tech, address USA. · Department of Mathematics, The University of Utah, address, address, USA.
Purpose: To develop a phase-aware self-supervised framework for accelerated high-resolution multi-echo gradient-echo (mGRE) MRI that preserves anatomical detail and complex echo information for quantitative mapping. Methods: We developed multi-stage phase-aware self-supervised reconstruction (MSPR-GRE), integrating progressive k-space expansion with phase estimation and refinement. Sequential unrolled networks reconstruct complex multi-echo images and update echo-dependent phase priors. A relaxation-informed temporal subspace is used for acquisitions with at least four echoes, while shorter echo trains are reconstructed in the image domain. Training uses held-out acquired k-space samples without fully sampled targets. Evaluation included 9.4 T ex vivo mouse-brain data 30um eight-echo and 18um dual-echo) and retrospectively undersampled 7 T AHEAD human-brain data, with supervised ablations. Results:MSPR-GRE reduced artifacts in regions of rapid phase variation and preserved fine mouse-brain anatomy. Across five AHEAD test datasets at R = 4, 8, and 12, it generally improved magnitude-image, T2* map, and QSM metrics, particularly at higher acceleration. At R = 12, fourth-echo magnitude NRMSE was 12.09±0.60% versus 21.55±1.12% for SSDU. Ablations indicated benefits of phase refinement, while supervised experiments supported multi-stage phase-aware reconstruction. Conclusion: MSPR-GRE improves accelerated mGRE reconstruction and quantitative-map agreement without fully sampled training targets. Further validation of generalizability and phase-estimation accuracy is needed.
Juan Liu, Nian Wang
Advanced Imaging Research Center, University of Texas Southwestern Medical Center, USA · Department of Biomedical Engineering, University of Texas Southwestern Medical Center, USA · Peter O’Donnell Jr. Brain Institute, University of Texas Southwestern Medical Center, USA