Spatial and Temporal Super-Resolution and Denoising of Intracardiac 4D Flow MRI
Authors: Yusuf A. Ozturk, Jie Zheng, Fanwei Kong
Organizations: Department of Mechanical Engineering and Materials Science, Washington University, Saint Louis, MO, USA · Mallinckrodt Institute of Radiology, Washington University School of Medicine, St. Louis, MO, USA
Intracardiac flow is driven by complex, and time-varying cardiac motion and interaction with the heart valves, producing complex flow features that may provide valuable information for disease diagnosis and treatment planning. 4D Flow MRI enables non-invasive, time-resolved measurement of three-dimensional cardiovascular velocity fields. However, substantial noise and limited spatial and temporal resolution limit its ability to accurately resolve flow features. We propose 4D-MRAR (4D Magnetic Resonance Augmented Reconstruction), a deep learning (DL) method to super-resolve and denoise 4D Flow MRI data. Unlike prior DL methods that train and infer on static vascular regions, our method predicts intracardiac flow within hearts undergoing large deformation. Moreover, it uniquely allows for simultaneous super-resolution in 3D space and time. The method combines a residual convolutional neural network, a recurrent neural network, and an implicit neural field to predict high-resolution velocity and pressure fields while efficiently enforcing the governing equations of fluid flow during training. Trained on a synthetic dataset obtained from computational fluid dynamics (CFD) simulations of left ventricle (LV) flow, our method generalizes to synthetic whole-heart 4D flow MRI data, and demonstrated promising performance on real 4D Flow MRI. The produced super-resolved flow fields exhibit significantly improved accuracy and greater consistency with the governing equations of blood flow compared with prior methods.
Figures & tables
Fig. 1: Synthetic 4D Flow MRI data generation pipeline. The CFD block computes synthetic velocity fields from patient-specific cardiac images. The FFT block transforms high-resolution images and synthetic magnitude images into k-space. The IFFT block adds noise and downsamples the k-space data before reconstructing low-resolution velocity and magnitude images.
Fig. 2: Overview of the 4D-MRAR architecture for spatiotemporal super-resolution of 4D Flow MRI. Low-resolution inputs are encoded into latent fields, temporally refined using a ConvLSTM module, and decoded at arbitrary spatiotemporal coordinates using a coordinate-based MLP with data and physics-informed losses.
Fig. 3: Example predicted velocity ( upred ) and pressure gradient fields ( ∇ppred ) of our method from low-resolution inputs ( uLR ) for the (a) finer-resolution CFD and (b) coarser-resolution CFD datasets. Columns represent different time steps ( Δt=10 ms). uGT and ∇pGT denote the ground truth. MV and AV denote LV inlet and outlet at mitral and aortic valves, respectively.
Fig. 4: Comparison between our model and baselines at an ETS.
Dataset
Metric
Time Step
Ours
4DFlowNet
Callmer et al.
A-PINN
Coarser
Abs.Err (m/s) ↓
ETS
0.0028
0.0038 ∗†
0.0042 ∗†
0.0069 ∗†
SITS
0.0032
–
0.0046 ∗†
0.0068 ∗†
MITS
0.0032
–
0.0045 ∗†
0.0066 ∗†
Rel.Err (%) ↓
ETS
15.35
19.93 ∗†
25.77 ∗†
39.63 ∗†
SITS
16.83
–
27.53 ∗†
38.32 ∗†
MITS
18.07
–
28.25 ∗†
38.98 ∗†
TABLE I: Quantitative comparison on the coarser (9 cases; 180 ETS, 360 SITS, 171 MITS patches) and finer CFD (2 cases; 30 ETS, 60 SITS, 24 MITS patches) test datasets.
Fig. 5: Predicted velocity fields on a whole heart unstructured mesh from synthetic low-resolution 4D Flow MRI generated from a whole heart CFD simulation with valves at systole and diastole frames. The low-resolution input is projected onto the whole heart mesh for comparison.
Metric
Time Step
Ours
4DFlowNet
Callmer et al.
A-PINN
Abs.Err (m/s) ↓
ETS
0.0092
0.0125 ∗†
0.0204 ∗†
0.0247 ∗†
SITS
0.0127
–
0.0225 ∗†
0.0252 ∗†
MITS
0.0148
–
0.0236 ∗†
0.0262 ∗†
Rel.Err (%) ↓
ETS
15.89
21.65 ∗†
37.68 ∗†
43.35 ∗†
SITS
21.56
–
40.90 ∗†
44.13 ∗†
MITS
24.63
–
42.03 ∗†
45.25 ∗†
TABLE II: Quantitative comparison on the whole heart dataset. Results were computed using 16 ETS, 31 SITS, and 15 MITS patches.
Fig. 6: Real 4D Flow MRI data predictions. Slices are chosen to highlight LV-aorta section of the cardiac flow. PA and LA stand for pulmonary artery and left atrium, respectively.
Patient
Ours
Callmer et al.
4DFlowNet
Linear Interp.
4D Flow MRI
Patient 1
4.79
17.15
11.59
20.03
30.46
Patient 2
6.59
16.66
8.34
15.82
21.11
Bold indicates best performance.
TABLE III: Quantitative comparison study on real 4D Flow MR data – Mean Divergence residual ( 1/s ) for two patients, computed with a shared voxel spacing across all methods.
Dataset
Metric
Time Step
Ours
Linear Interp.
CNN Decoder
No Physics
Coarser
Abs.Err (m/s) ↓
ETS
0.0028
0.0032 ∗†
0.0027 ∗
0.0031 ∗†
SITS
0.0032
0.0035 ∗†
0.0035 ∗†
0.0034 ∗†
MITS
0.0032
0.0035 ∗†
0.0034 ∗
0.0035 ∗†
Rel.Err (%) ↓
ETS
15.35
17.96 ∗†
15.35
17.43 ∗†
SITS
16.83
18.76 ∗†
17.90 ∗†
18.33 ∗†
MITS
18.07
19.77 ∗†
18.58 ∗
19.40 ∗†
TABLE IV: Quantitative comparison of ablation variants on the coarser- and finer-resolution CFD datasets.
Fig. 7: Comparison between the coordinate-based MLP and the CNN decoder on the whole heart dataset. Velocity fields and absolute error maps are shown on an aortic-root cross-section at early systole and a longitudinal ventricular plane at early diastole.Ground-truth velocity fields are compared with MLP-decoder and trilinearly projected CNN-decoder predictions.
Four-dimensional (4D) flow magnetic resonance imaging (MRI) is a powerful non-invasive technique for visualizing and quantifying complex blood flow patterns in vivo. Despite its clinical promise, broader adoption is limited by low spatial resolution and sensitivity to noise, which restrict accurate assessment of critical hemodynamic biomarkers such as wall shear stress, pressure gradients, and turbulent kinetic energy. To overcome these challenges, we propose a deep learning-based super-resolution framework that integrates multi-scale feature extraction and attention mechanisms to enhance the quality of 4D flow MRI data. The model was trained on a dataset of 120 patients with 240 stenosed carotid arteries. High-resolution ground truth data were generated using patient-specific computational fluid dynamics (CFD) simulations based on segmented vascular geometries and physiologically realistic boundary conditions, and the resulting velocity fields served as targets for supervised learning. The proposed architecture uses convolutional block attention modules (CBAM) to guide the network toward clinically relevant spatial features and to suppress noise in low-resolution inputs. Quantitative results show that the attention-guided model substantially reduces the root mean square error (RMSE) compared with a baseline model without attention, and qualitative velocity contour analysis confirms improved reconstruction of intricate flow patterns. These findings highlight the capacity of the model to restore high-fidelity flow fields under noisy conditions and support the use of deep learning to extend the clinical utility of 4D flow MRI for non-invasive hemodynamic assessment.
Ali Mokhtari, Dominik Obrist
ARTORG Center for Biomedical Engineering Research, University of Bern, Bern, Switzerland
Objective: To improve the spatial and temporal resolution of aortic 4D flow MRI through a unified deep learning framework for post-acquisition super-resolution (SR) of reconstructed velocity vector fields. Methods: We propose CURRENT, a recurrent neural network that performs spatial and temporal SR with a single network architecture. Supervised training was conducted using paired low- and high-resolution datasets generated by embedding computational fluid dynamics simulations of aortic flow into in vivo thoracic MRI backgrounds, followed by controlled spatial and temporal downsampling. CURRENT was evaluated for two-fold spatial and temporal SR against 4DFlowNet and for two-fold spatiotemporal SR against sequential spatial and temporal SR pipelines based on 4DFlowNet. Image-based and k-space-based quantitative metrics were considered and conventional interpolation were used as baselines. In addition, the impact of CURRENT's spatiotemporal SR on physics-informed relative pressure estimation was assessed. Finally, the ability of CURRENT to generalize from synthetic training data to in vivo acquisitions was investigated. Results: CURRENT achieved superior spatial SR performance and comparable temporal SR performance to dedicated 4DFlowNet models while providing a unified solution for both tasks. Simultaneous spatiotemporal SR using CURRENT yielded more accurate relative pressure estimates than either spatial or temporal SR alone and outperformed sequential 4DFlowNet-based spatial and temporal SR approaches. Furthermore, the proposed model demonstrated promising generalization to in vivo 4D flow MRI data despite being trained exclusively on synthetic datasets. Conclusion: CURRENT provides a unified framework for spatial and temporal SR of 4D flow MRI. The proposed approach has the potential to enhance physics-based analysis and broaden the applicability of accelerated 4D flow MRI.
Luuk Jacobs, Pietro Dirix, Sebastian Kozerke
Institute for Biomedical Engineering, University and ETH Zurich, Zurich, Switzerland
Super-resolution is widely used in medical imaging to enhance low-quality data, reducing scan time and improving abnormality detection. Conventional super-resolution approaches typically rely on paired datasets of downsampled and original high resolution images, training models to reconstruct high resolution images from their artificially degraded counterparts. However, in real-world clinical settings, low resolution data often arise from acquisition mechanisms that differ significantly from simple downsampling. As a result, these inputs may lie outside the domain of the training data, leading to poor model generalization due to domain shift. To address this limitation, we propose a distributional deep learning framework that improves model robustness and domain generalization. We develop this approch for enhancing the resolution of 4D Flow MRI (4DF). This is a novel imaging modality that captures hemodynamic flow velocity and clinically relevant metrics such as vessel wall stress. These metrics are critical for assessing aneurysm rupture risk. Our model is initially trained on high resolution computational fluid dynamics (CFD) simulations and their downsampled counterparts. It is then fine-tuned on a small, harmonized dataset of paired 4D Flow MRI and CFD samples. We derive the theoretical properties of our distributional estimators and demonstrate that our framework significantly outperforms traditional deep learning approaches through real data applications. This highlights the effectiveness of distributional learning in addressing domain shift and improving super-resolution performance in clinically realistic scenarios.
Xiaoyi Wen, Fei Jiang
Department of Epidemiology and Biostatistics, University of California, San Francisco