On The Role of K-Space Acquisition in MRI Reconstruction Domain-Generalization

TL;DR

The paper addresses domain generalization in MRI reconstruction by treating k-space acquisition trajectories as an active lever for robustness. It systematically compares learned versus fixed trajectories across Cartesian and Radial sampling and introduces acquisition uncertainty during training through random and adversarial trajectory noise, plus image-domain noise. The results show trajectory learning yields cross-domain gains and that trajectory-based perturbations provide more reliable generalization than image-domain noise, highlighting the role of acquisition design as a regularizer. The findings suggest integrating trajectory optimization with domain-adversarial or meta-learning strategies could further enhance cross-domain MRI reconstruction without target-domain supervision.

Abstract

Recent work has established learned k-space acquisition patterns as a promising direction for improving reconstruction quality in accelerated Magnetic Resonance Imaging (MRI). Despite encouraging results, most existing research focuses on acquisition patterns optimized for a single dataset or modality, with limited consideration of their transferability across imaging domains. In this work, we demonstrate that the benefits of learned k-space sampling can extend beyond the training domain, enabling superior reconstruction performance under domain shifts. Our study presents two main contributions. First, through systematic evaluation across datasets and acquisition paradigms, we show that models trained with learned sampling patterns exhibitimproved generalization under cross-domain settings. Second, we propose a novel method that enhances domain robustness by introducing acquisition uncertainty during training-stochastically perturbing k-space trajectories to simulate variability across scanners and imaging conditions. Our results highlight the importance of treating kspace trajectory design not merely as an acceleration mechanism, but as an active degree of freedom for improving domain generalization in MRI reconstruction.

Figures (7)

• Figure 1: Fixed Radial Trajectory.
• Figure 2: Learned Radial Trajectories (PT-ViT-L).
• Figure 3: Standard traj. noise (PT-ViT-L).
• Figure 4: Adversarial traj. noise (PT-ViT-L).
• Figure 5: PT-ViT-L reconstructions without TL on Cartesian; trajectory noise outperforms image noise.