Accelerating Stroke MRI with Diffusion Probabilistic Models through Large-Scale Pre-training and Target-Specific Fine-Tuning
Authors: Yamin Arefeen, Sidharth Kumar, Steven Warach, Hamidreza Saber, Jonathan Tamir
Organizations: Chandra Family Department of Electrical and Computer Engineering The University of Texas at Austin Austin, TX, USA · Department of Imaging Physics MD Anderson Cancer Center Houston, TX, USA · Dell Medical School Department of Neurology The University of Texas at Austin Austin, TX, USA · Dell Medical School Department of Neurosurgery The University of Texas at Austin Austin, TX, USA · Dell Medical School Department of Diagnostic Medicine The University of Texas at Austin Austin, TX, USA · Oden Institute for Computational Engineering and Sciences The University of Texas at Austin Austin, TX, USA
Purpose: To develop a data-efficient strategy for accelerated MRI reconstruction with Diffusion Probabilistic Generative Models (DPMs) that enables faster scan times in clinical stroke MRI when only limited fully-sampled data samples are available. Methods: Our simple training strategy, inspired by the foundation model paradigm, first trains a DPM on a large, diverse collection of publicly available brain MRI data in fastMRI and then fine-tunes on a small dataset from the target application using carefully selected learning rates and fine-tuning durations. The approach is evaluated on controlled fastMRI experiments and on clinical stroke MRI data with a blinded clinical reader study. Results: DPMs pre-trained on approximately 4000 subjects with non-FLAIR contrasts and fine-tuned on FLAIR data from only 20 target subjects achieve reconstruction performance comparable to models trained with substantially more target-domain FLAIR data across multiple acceleration factors. Experiments reveal that moderate fine-tuning with a reduced learning rate yields improved performance, while insufficient or excessive fine-tuning degrades reconstruction quality. When applied to clinical stroke MRI, a blinded reader study involving two neuroradiologists indicates that images reconstructed using the proposed approach from 2× accelerated data are non-inferior to standard-of-care in terms of image quality and structural delineation. Conclusion: Large-scale pre-training combined with targeted fine-tuning enables DPM-based MRI reconstruction in data-constrained, accelerated clinical stroke MRI. The proposed approach substantially reduces the need for large application-specific datasets while maintaining clinically acceptable image quality, supporting the use of foundation-inspired diffusion models for accelerated MRI in targeted applications.
Figures & tables
Figure 1: (A) The standard DPM training pipeline trains exclusively on the target-domain dataset, but if the target dataset is small, this can lead to degraded performance. (B) The proposed training pipeline mitigates data scarcity by first pre-training the DPM on a large, diverse external MRI dataset, followed by fine-tuning on a small target-domain dataset using a reduced learning rate and limited number of epochs. The fine-tuned model is then used for posterior sampling–based reconstruction from under-sampled k-space.
Figure 2: Quantitative results from the controlled fastMRI experiments where DPMs were fine-tuned on 20 subjects to reconstruct fastMRI flair data. NRMSE on 500 validation slices as a function of fine-tuning duration (epochs) is plotted for different learning rates and acceleration factors. Across all acceleration factors, moderate fine-tuning improves reconstruction quality relative to no fine-tuning, while excessive fine-tuning leads to degraded performance, particularly at higher learning rates.
Figure 3: A representative example from the fastMRI FLAIR experiment. (A) Quantitative NRMSE as a function of fine-tuning duration for different learning rates, highlighting that a comparatively smaller learning rate and fewer fine-tuning epochs yields the best performance. (B) Fully sampled reference FLAIR image. (C) Zoomed reconstructions obtained using different learning rates with 650 fine-tuning epochs, compared against the fully sampled reference. (D) Reconstructions obtained using a fixed learning rate of 1×10−5 across different fine-tuning durations. Appropriate fine-tuning hyperparameters lead to visibly improved structural fidelity and reduced artifacts, as highlighted by the orange arrows.
Figure 4: (A) DPMs are compared across methods with access to different amounts of target-domain FLAIR data. Methods 1–3 are trained using FLAIR data from all 344 subjects, while Methods 5 and 6 are trained using 20 FLAIR subjects and the external dataset. The proposed approach (Method 4), pre-trained on a large external dataset and fine-tuned on 20 FLAIR subjects, achieves comparable NRMSE to Methods 1-3, despite using substantially less target domain data and consistently outperforms Methods 5,6 trained on the same limited data. (B) displays example reconstructions from this experiment.
Figure 5: Using the best fine-tuned DPM, average NRMSE on fastMRI FLAIR validation data is plotted as a function of the data-consistency step size ζ for multiple acceleration factors. Red markers denote the default choice ζ=1 , while green markers indicate the value of ζ that minimizes reconstruction error at each acceleration. The optimal ζ increases with acceleration.
Figure 6: (A) NRMSE of SWI reconstructions averaged over five validation subjects from data retrospectively accelerated by R≈3.75 . (B) Fully-sampled reference. (C) Zoomed-in reconstructions obtained with fixed epochs and varying learning rate and (D) fixed learning rate and varying epochs. Similar to the fastMRI FLAIR experiments, insufficient or excessive fine-tuning duration and higher learning rates consistently result in worse reconstructions, demonstrating that the proposed training strategy is effective for clinical stroke SWI data.
Figure 7: (A) NRMSE of DWI reconstructions averaged over five validation subjects from data retrospectively accelerated by R≈3.75 . (B) Fully-sampled reference. (C) Zoomed-in reconstructions obtained with fixed epochs and varying learning rate and (D) fixed learning rate and varying epochs. Similar to the fastMRI FLAIR experiments, insufficient or excessive fine-tuning duration and higher learning rates consistently result in worse reconstructions, demonstrating that the proposed training strategy is effective for clinical stroke DWI data.
Figure 8: Representative images from a clinical stroke protocol are shown for a single subject, comparing standard-of-care reconstructions (top row) with images retrospectively under-sampled by an additional factor of two and reconstructed using the proposed DPM-based approach (bottom row). The protocol includes diffusion-weighted imaging (shown as derived trace and apparent diffusion coefficient (ADC) maps), susceptibility-weighted imaging (SWI), FLAIR, and MPRAGE. Across all contrasts, the proposed accelerated reconstructions preserve structural detail and contrast comparable to the standard-of-care.
Figure 9: Mean reader scores for structural delineation (left) and image quality metrics (right) are shown for two board-certified neuroradiologists evaluating full stroke-protocol image sets reconstructed from standard-of-care data (GT) and from retrospectively accelerated data reconstructed using the proposed DPM-based approach (Recon). The top row reports results from Reader 1, who evaluated all 80 subjects, while the bottom row reports results from Reader 2, who evaluated 21 subjects. For Reader 1, the proposed accelerated reconstructions achieved statistically significant higher scores for several image quality metrics and for white matter and ventricular delineation. For Reader 2, no significant differences were observed except for SNR, where the standard-of-care reconstruction scored higher; however, mean scores for both methods remained high across all categories.
Figure 10: The proposed DPM, pre-trained on fastMRI data and fine-tuned on FLAIR or SWI data from 20 subjects, is applied to reconstruct both retrospectively and prospectively under-sampled clinical stroke data. For each contrast, the fully sampled reference, retrospective reconstructions, and prospective reconstructions are displayed along with their effective acceleration factors. Across both contrasts, the proposed method qualitatively reduces reconstruction artifacts relative to the ℓ1 baseline, and achieves comparable image quality between retrospectively and prospectively under-sampled acquisitions
Diffusion models (DMs) have emerged as powerful generative priors for MRI reconstruction with promising results. Yet DM-based methods require extensive iterative refinement, limiting their practical deployment. Consistency models (CMs) provide a compelling alternative, aiming to map out the diffusion trajectory in a single pass, enabling faster generation. In this work, we propose CM-RED, a novel MRI reconstruction method that integrates a pretrained CM into the regularization by denoising (RED) scheme. Our method builds on accelerated proximal gradient RED (RED-APG), and further incorporates controlled noise injection during the update steps to enhance generative diversity and accelerate convergence. Extensive experiments on the fastMRI knee and brain datasets demonstrate that CM-RED achieves high-quality reconstructions across multiple anatomies, contrast weights, acceleration factors, and undersampling patterns, using only 4 network function evaluations (NFEs). The proposed method consistently outperforms existing DM- and CM-based approaches in both quantitative metrics and visual fidelity, and exhibits strong robustness to hyperparameter variations, highlighting CM-RED as an efficient and effective generative framework for accelerated MRI reconstruction. The source code and pretrained models are publicly available at https://github.com/MerveGulle/CM-RED.
Merve Gülle, Junno Yun, Yaşar Utku Alçalar +1
Department of Electrical and Computer Engineering, University of Minnesota, MN, USA · Center for Magnetic Resonance Research, University of Minnesota, MN, USA
Diffusion MRI (dMRI) probes water diffusion within the brain, providing unique information about tissue microstructure and white matter connectivity. High-quality dMRI enables advanced and accurate analyses but demands dense acquisitions, resulting in long scan times and hardware constraints that are often infeasible. Further, dMRI requires dedicated computational methods to support diverse analysis tasks (e.g., parcellation, registration, and synthesis) for both neuroscientific and clinical applications. Here, we present dMRI-F, a foundation model that unifies dMRI data enhancement and downstream tasks across diverse acquisitions and populations, while requiring only minimal acquisition input. dMRI-F is pre-trained with a q-space masked reconstruction task: predicting densely sampled multi-shell data from only six volumes. It learns generalizable latent feature representations of dMRI data, which serve as a core foundation for adapting to diverse downstream tasks, including brain parcellation, brain registration, and cross-modal synthesis. Across nine independently acquired datasets, we demonstrate that dMRI-F outperforms state-of-the-art methods with robust generalization across acquisitions, populations, and tasks. Our results demonstrate dMRI-F's versatility as a general-purpose framework that enables diverse dMRI applications from low-cost, sparse acquisitions.
Xi Zhu, Yijie Li, Xiaofan Wang +4
University of Electronic Science and Technology of China.
Accelerated magnetic resonance imaging (MRI) enabled by the training of deep learning (DL)-based image recon. models requires large and diverse raw k-space datasets. In most clinical MRI applications, due to storage and patient privacy concerns, raw k-space data is discarded and magnitude-only images are the only component saved. Consequently, a large portion of the DL-based MRI recon. literature has either relied on small training datasets or has used one of the few available open-source k-space datasets. At the same time, the growing number of anonymized magnitude-only image registries/databases motivates the development of techniques that can use them as training datasets for generalizable DL-based recon. models. Here we propose to address this challenge by employing a generative approach based on conditional score-based diffusion models (SBDMs): given a magnitude-only MR image, it synthesizes a phase map (in the image domain) that realistically corresponds to the magnitude-only image. We evaluate its generative capabilities in a downstream DL-based recon. task whereby a large k-space dataset is generated by combining the SBDM-synthesized phase-maps and the corresponding magnitude-only images, and this k-space dataset is then used to train a DL model for accelerated MRI recon. We compare the performance of the resulting DL model versus those trained according to (a) a naive approach that uses smooth phase, (b) a k-space training dataset generated using synthesized phase maps derived from a generative adversarial network, and (c) the ground truth k-space data. Our results suggest that the DL model trained from SBDM-synthesized k-space data outperforms the other approaches in terms of quantitative metrics as well as qualitatively observed recon. fidelity, i.e., whether the reconstructed images include erroneous or hallucinated features that could adversely impact diagnostic accuracy.
M. Berk Sahin, Dilek Yalcinkaya, Abolfazl Hashemi +1