SDUM: A Scalable Deep Unrolled Model for Universal MRI Reconstruction

TL;DR

SDUM addresses universal MRI reconstruction by learning a scalable, foundation-model-like unrolled architecture that generalizes across anatomies, contrasts, sampling patterns, and field strengths. It integrates a Restormer-based reconstructor, learned coil-sensitivity maps, sampling-aware data fidelity, and universal conditioning with progressive cascade expansion; the model shows predictable gains with depth and achieves state-of-the-art performance on CMRxRecon2024/2025 and fastMRI brain without task-specific fine-tuning. Ablation and scaling-law experiments validate the contributions and demonstrate robust improvements across tasks, with gains up to about 1 dB PSNR. This work enables practical deployment of universal, scalable MRI reconstruction and points toward future extensions such as 3D volumetric reconstruction and self-supervised training.

Abstract

Clinical MRI encompasses diverse imaging protocols--spanning anatomical targets (cardiac, brain, knee), contrasts (T1, T2, mapping), sampling patterns (Cartesian, radial, spiral, kt-space), and acceleration factors--yet current deep learning reconstructions are typically protocol-specific, hindering generalization and deployment. We introduce Scalable Deep Unrolled Model (SDUM), a universal framework combining a Restormer-based reconstructor, a learned coil sensitivity map estimator (CSME), sampling-aware weighted data consistency (SWDC), universal conditioning (UC) on cascade index and protocol metadata, and progressive cascade expansion training. SDUM exhibits foundation-model-like scaling behavior: reconstruction quality follows PSNR ${\sim}$ log(parameters) with correlation $r{=}0.986$ ($R^2{=}0.973$) up to 18 cascades, demonstrating predictable performance gains with model depth. A single SDUM trained on heterogeneous data achieves state-of-the-art results across all four CMRxRecon2025 challenge tracks--multi-center, multi-disease, 5T, and pediatric--without task-specific fine-tuning, surpassing specialized baselines by up to ${+}1.0$~dB. On CMRxRecon2024, SDUM outperforms the winning method PromptMR+ by ${+}0.55$~dB; on fastMRI brain, it exceeds PC-RNN by ${+}1.8$~dB. Ablations validate each component: SWDC ${+}0.43$~dB over standard DC, per-cascade CSME ${+}0.51$~dB, UC ${+}0.38$~dB. These results establish SDUM as a practical path toward universal, scalable MRI reconstruction.

Figures (7)

• Figure 1: SDUM overview. Each cascade combines a Restormer-based reconstructor, a learned CSME, sampling-aware weighted data consistency (SWDC), and universal conditioning (UC) on cascade index and protocol metadata.
• Figure 2: Example learned sampling-aware data-consistency weight maps showing spatial adaptation to sampling patterns.
• Figure 3: Scaling behavior of unrolled depth $T$ for reconstruction performance.
• Figure 4: Qualitative comparison of MRI reconstruction methods across multiple cardiac imaging modalities. Three representative cases from the CMRxRecon2024 Task 1 validation set demonstrate the superior performance of SDUM compared with PromptMR and PromptMR+.
• Figure 5: Statistical comparison of PSNR and SSIM improvements over PromptMR baseline for Task 1 (Uniform Undersampling).