UEPS: Robust and Efficient MRI Reconstruction
This addresses a critical barrier to clinical adoption of MRI reconstruction methods by improving robustness for diverse clinical shifts, though it is incremental as it builds on existing deep unrolled models.
The paper tackled the problem of deep unrolled models' lack of robustness under domain shift in accelerated MRI reconstruction by proposing UEPS, a novel architecture that eliminates coil sensitivity map dependency, resulting in consistent and substantial outperformance across 10 out-of-distribution test sets with low-latency inference.
Deep unrolled models (DUMs) have become the state of the art for accelerated MRI reconstruction, yet their robustness under domain shift remains a critical barrier to clinical adoption. In this work, we identify coil sensitivity map (CSM) estimation as the primary bottleneck limiting generalization. To address this, we propose UEPS, a novel DUM architecture featuring three key innovations: (i) an Unrolled Expanded (UE) design that eliminates CSM dependency by reconstructing each coil independently; (ii) progressive resolution, which leverages k-space-to-image mapping for efficient coarse-to-fine refinement; and (iii) sparse attention tailored to MRI's 1D undersampling nature. These physics-grounded designs enable simultaneous gains in robustness and computational efficiency. We construct a large-scale zero-shot transfer benchmark comprising 10 out-of-distribution test sets spanning diverse clinical shifts -- anatomy, view, contrast, vendor, field strength, and coil configurations. Extensive experiments demonstrate that UEPS consistently and substantially outperforms existing DUM, end-to-end, diffusion, and untrained methods across all OOD tests, achieving state-of-the-art robustness with low-latency inference suitable for real-time deployment.