Organizations: School of Information Science and Technology, ShanghaiTech University, Shanghai, China · School of Biomedical Engineering, Shanghai Jiao Tong University, Shanghai, China
Low-field (LF) magnetic resonance imaging (MRI) democratizes access to diagnostic imaging but is fundamentally limited by low signal-to-noise ratio and significant tissue contrast distortion due to field-dependent relaxation dynamics. Reconstructing high-field (HF) quality images from LF data is a blind inverse problem, severely challenged by the scarcity of paired training data and the unknown, non-linear contrast transformation operator. Existing zero-shot methods, which assume simplified linear degradation, often fail to recover authentic tissue contrast. In this paper, we propose DACT(Diffusion-Based Adaptive Contrast Transport), a novel zero-shot framework that restores HF-quality images without paired supervision. DACT synergizes a pre-trained HF diffusion prior to ensure anatomical fidelity with a physically-informed adaptive forward model. Specifically, we introduce a differentiable Sinkhorn optimal transport module that explicitly models and corrects the intensity distribution shift between LF and HF domains during the reverse diffusion process. This allows the framework to dynamically learn the intractable contrast mapping while preserving topological consistency. Extensive experiments on simulated and real clinical LF datasets demonstrate that DACT achieves state-of-the-art performance, yielding reconstructions with superior structural detail and correct tissue contrast.
Figures & tables
Figure 1: Overview of the DACT . ( a ) Soft histograms of input LF image y and HF estimate z are generated via kernel density estimation (KDE). The Sinkhorn algorithm computes an optimal transport plan for contrast alignment, while a learnable weight map α adaptively balances contrast transformation and structural detail preservation. The known spatial degradation H is then applied to obtain the reconstructed LF estimate y^ . ( b ) A high-field prior x^0 generated by a pretrained diffusion model is incorporated to jointly optimize the target HF image and the forward degradation process.
Category
Method
Synthetic T1w
Synthetic T2w
Real T2w
PSNR ↑
SSIM ↑
LPIPS ↓
PSNR ↑
SSIM ↑
LPIPS ↓
NIQE ↓
BRISQUE ↓
FID ↓
INR-based
ULFINR
22.06
0.717
0.219
20.26
0.569
0.270
6.614
29.53
169.65
Supervised
PF-SR
21.92
0.705
0.213
21.21
0.654
0.200
5.452
24.18
118.37
Diffusion- based
DiffDeuR
23.17
0.684
0.261
23.66
0.809
0.262
5.887
20.67
130.12
GDP
23.42
0.679
0.217
23.92
0.766
0.228
4.821
24.45
115.96
TAO
23.61
0.689
0.216
24.18
0.782
0.209
5.000
19.91
110.93
Table 1: Quantitative comparisons of DACT with baselines on both the synthetic and real-world LF dataset. The best performances are highlighted in bold .
Figure 2: Qualitative comparisons of DACT with baselines on two representative synthetic LF data. The bottom row shows the downstream segmentation masks for GM, WM, and CSF based on the T2w results, with the average Dice scores.
Figure 3: Qualitative reconstruction (top) and tissue segmentation (bottom) on real LF data. DACT yields sharp cortical contrast and highly accurate segmentation.
Figure 4: Convergence analysis. (Left) All methods successfully minimize fidelity error. (Right) Validation error against the ground truth reveals baseline overfitting, whereas DACT stably converges to the authentic contrast transformation.
Department of Bioengineering and Imperial-X, Imperial College London, London, UK · Department of Electrical and Computer Systems Engineering, Monash University, Melbourne, Australia · National Heart and Lung Institute, Imperial College London, London SW7 2AZ, UK +2