Topology-Aware Co-Domain Learning for Ultra-Low-Field MRI Enhancement.

Ultra-low-field (ULF) MRI offers a cost effective and portable imaging alternative for High-Field (HF) MR imaging, but remains limited by its low signal to-noise ratio (SNR) and compressed contrast resolution. Existing enhancement methods have shown promising improvements yet often fail to recover fine anatomical details or over-smooth high-frequency components, limiting their reliability for a high-fidelity synthesis process. To address these challenges, we propose a co-domain diffusion frame work that jointly leverages spatial and frequency representations for ULF enhancement. Our framework integrates two diffusion generators trained collaboratively guided by a topology-aware alignment that preserves the hierarchical organization of brain anatomy. Comprehensive evaluation across multi-contrast imaging demonstrates that our framework achieves comparatively better quantitative and qualitative performance with stronger regional correspondence measured via Dice, RANSAC, and volume correlations. Furthermore, age-volume consistency analysis shows that the synthesized contrasts preserve realistic morphological trends in actual HF images. Evaluation on an external dataset and robustness analysis under controlled input corruption demonstrate reliable performance beyond the training conditions. Moreover, a blinded radiological evaluation demonstrated the clinical reliability of the synthetic results. Overall, the proposed approach yields anatomically coherent and structurally consistent results, offering a robust solution for reliable ULF MRI enhancement.