Organizations: Fudan University, Shanghai, China · University of Cambridge, Cambridge, United Kingdom · Shanghai Artificial Intelligence Laboratory, Shanghai, China
Real-world MRI reconstruction systems face the open-world challenge: test data from unseen imaging centers, anatomical structures, or acquisition protocols can differ drastically from training data, causing severe performance degradation. Existing methods struggle with this challenge. To address this, we propose BiasRecon, a bias-calibrated adaptation framework grounded in the minimal intervention principle: preserve what transfers, calibrate what does not. Concretely, BiasRecon formulates open-world adaptation as an alternating optimization framework that jointly optimizes three components: (1) frequency-guided prior calibration that introduces layer-wise calibration variables to selectively modulate frequency-specific features of the pre-trained score network via self-supervised k-space signals, (2) score-based denoising that leverages the calibrated generative prior for high-fidelity image reconstruction, and (3) adaptive regularization that employs Stein's Unbiased Risk Estimator to dynamically balance the prior-measurement trade-off, matching test-time noise characteristics without requiring ground truth. By intervening minimally and precisely through this alternating scheme, BiasRecon achieves robust adaptation with fewer than 100 tunable parameters. Extensive experiments across four datasets demonstrate state-of-the-art performance on open-world reconstruction tasks.
Figures & tables
Figure 1: Transferability analysis. (a) Open-world distribution shifts from anatomy, center, and modality changes. Models trained on FastMRI knee (inside dashed box) are tested on three OOD scenarios. (b) t-SNE visualization of features from bottom (deeper) and top (shallow) layers across different domains. (c) Frequency decomposition analysis of low-frequency and high-frequency components across different domains.
Figure 2: Overview of BiasRecon framework. The top panel illustrates the alternating optimization process of BiasRecon. The framework consists of three key components: a) Frequency-guided Prior Calibration, which adaptively adjusts feature distributions through parameter-efficient modulation; b) Score-based reverse sampling process as the foundation; and c) Regularization Parameter Adaptation, which dynamically balances measurement consistency and prior constraints.
Figure 3
Method
FastMRI Knee (In-Distribution)
FastMRI-Brain (Cross-Anatomy)
PSNR ↑
SSIM ↑
LPIPS ↓
PSNR ↑
SSIM ↑
LPIPS ↓
Zero-filling
19.83±1.52
0.691±0.025
0.329±0.031
23.72±1.83
0.785±0.028
0.236±0.022
Conventional CS method
TV
26.04±1.63
0.771±0.023
0.313±0.025
28.14±1.86
0.817±0.027
0.187±0.024
Internal deep image prior
ConvDecoder
29.97±2.15
0.745±0.029
0.257±0.027
31.27±2.73
0.876±0.022
0.141±0.022
Table 1: Open-world MRI reconstruction performance under 8 × acceleration with Gaussian 1D sampling.
Figure 5: Qualitative comparison of MRI reconstruction results under 8 × Gaussian 1D undersampling.
Method
Metric
Gaussian 1D
Uniform 1D
4 ×
8 ×
4 ×
8 ×
TV
PSNR
33.65 ± 1.29
28.14 ± 1.86
28.59 ± 1.69
25.71 ± 1.61
SSIM
0.868 ± 0.017
0.817 ± 0.027
0.801 ± 0.031
0.725 ± 0.043
LPIPS
0.136 ± 0.017
0.187 ± 0.024
0.174 ± 0.028
0.254 ± 0.043
ConvDecoder
PSNR
33.19 ± 2.41
31.27 ± 2.73
31.94 ± 2.35
26.27 ± 2.86
SSIM
0.892 ± 0.020
0.876 ± 0.022
0.860 ± 0.034
0.728 ± 0.070
Table 2: Robustness against different sampling patterns and acceleration factors on open-world FastMRI-brain dataset. Comparison includes both Gaussian 1D and Uniform 1D sampling at 4 × and 8 × acceleration rates.
Method
PSNR ↑
SSIM ↑
LPIPS ↓
Baseline
30.23±3.57
0.844±0.035
0.187±0.028
w/o RPA
31.28±3.22(↑1.05)
0.855±0.035(↑.011)
0.183±0.029(↓.004)
w/o FPC
32.51±4.04(↑2.28)
0.853±0.058(↑.009)
0.157±0.065(↓.030)
Ours (FPC + RPA)
33.17±3.11(↑2.94)
0.866±0.041(↑.022)
0.138±0.027(↓.049)
Table 3: Ablation study on Cross-Center shift under 4 × Gaussian 1D acceleration. FPC: Frequency-guided Prior Calibration; RPA: Regularization Parameter Adaptation.
Dormant Awakening (DA) specializes a magnetic resonance imaging (MRI) reconstruction model by fitting a small set of zero weights to one labeled slice. Across fourteen U-shaped convolutional network (U-Net) and vision transformer (ViT) sources, fitting 1,004 or 10,560 weights gives mean peak signal-to-noise ratio (PSNR) gains of .430 and .509 dB. We analyze how scaling the fitted output correction changes evaluation PSNR, then use the calibration correction size to cap changes on new images. The cap bounds changes in root mean squared error (RMSE) relative to the source without an evaluation reference. At 10% adaptation budget, we compare fixed halving and dynamic capping for DA, unrestricted sparse adaptation and low-rank adaptation (LoRA). The controls increase both measured calibration-evaluation correlations across all six tested architecture-adapter settings. Experiments concern one constructed fastMRI-to-M4Raw shift with fixed source checkpoints and participant panels.
Mohammed Wattad, Tamir Shor, Alexander M. Bronstein
Faculty of Computer Science, Technion-Israel Institute of Technology, Haifa, Israel
We develop a novel transfer learning framework to tackle the challenge of limited training data in image reconstruction problems. The proposed framework consists of two training steps, both of which are formed as bi-level optimizations. In the first step, we train a powerful universal feature-extractor that is capable of learning important knowledge from large, heterogeneous data sets in various domains. In the second step, we train a task-specific domain-adapter for a new target domain or task with only a limited amount of data available for training. Then the composition of the adapter and the universal feature-extractor effectively explores feature which serve as an important component of image regularization for the new domains, and this leads to high-quality reconstruction despite the data limitation issue. We apply this framework to reconstruct under-sampled MR images with limited data by using a collection of diverse data samples from different domains, such as images of other anatomies, measurements of various sampling ratios, and even different image modalities, including natural images. Experimental results demonstrate a promising transfer learning capability of the proposed method.
Yunmei Chen, Chi Ding, Xiaojing Ye
Department of Mathematics, University of Florida, Gainesville, FL 32611, USA. · Department of Mathematics and Statistics, Georgia State University, Atlanta, GA 30303, USA.
MRI reconstruction is an inherently ill-posed inverse problem, since incomplete measurements admit many plausible solutions. This ambiguity becomes more severe under high acceleration, where pixel-domain continuous predictors tend to average over feasible reconstructions and suppress high-frequency anatomy. We address this limitation by moving reconstruction to discrete multi-scale latent space and posing it as autoregressive next-acceleration-scale prediction. Leveraging discrete priors proven effective in visual autoregressive modeling, our method restricts the solution to compact sequences of codebook tokens, enabling sharp reconstructions even from extremely sparse measurements. This discrete autoregressive formulation also aligns naturally with modern large language model post-training techniques. Building on this observation, we introduce on-policy privileged information distillation for visual autoregressive modeling, where a teacher is provided training only privileged context that is unavailable at inference, in our case fully sampled acquisitions, and supervises a student trained on its own rollouts, leading to consistent reconstruction gains. Through extensive experiments on the fastMRI benchmark, we show that our approach delivers improved reconstruction performance across diverse sampling patterns under extreme undersampling. Project website is here.