Zero-shot Low-Field MRI Enhancement via Diffusion-Based Adaptive Contrast Transport

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.

💡 Research Summary

The paper tackles the longstanding challenge of enhancing low‑field (LF) magnetic resonance images, which suffer from low signal‑to‑noise ratio (SNR) and, more critically, a non‑linear distortion of tissue contrast caused by field‑dependent relaxation dynamics. Because paired LF–high‑field (HF) scans are rarely available, the authors pursue a zero‑shot approach that does not rely on supervised training data. Their solution, DACT (Diffusion‑Based Adaptive Contrast Transport), combines a pre‑trained HF diffusion model with a physically‑inspired, differentiable contrast‑transport module based on optimal transport (OT).

Problem formulation
The degradation from HF to LF is expressed as y = H Φ(x) + n, where H denotes spatial down‑sampling, Φ is an unknown, non‑linear contrast mapping, and n is noise. Φ cannot be analytically modeled because T1/T2 relaxation times change non‑linearly with magnetic field strength. Consequently, reconstructing an HF‑equivalent image from LF data is a severely ill‑posed inverse problem that requires both an accurate forward model and a strong image prior.

Methodology

1. Diffusion prior – An unconditional diffusion model is trained on a large HF (3 T) dataset (HCP). During inference, the model provides a generative prior ˆx₀ that constrains the solution to the natural HF image manifold.
2. Adaptive contrast transport – Instead of trying to write down Φ, the authors treat the intensity distributions of LF and HF images as two probability measures. Using kernel density estimation they obtain soft histograms h_LF and h_HF, then compute an OT plan with the Sinkhorn algorithm (regularized with a squared Euclidean cost). The plan is collapsed into a 1‑D look‑up table (LUT) that maps each HF intensity to a corrected LF intensity. This LUT is differentiable, allowing gradient‑based optimization.
3. Spatially‑adaptive weighting – A learnable pixel‑wise weight map α∈