关键词: 磁共振指纹成像, 定量磁共振成像, 流形结构化, 局部低秩

Abstract:

Magnetic resonance fingerprinting (MRF) has demonstrated considerable potential for the quantitative assessment of tissue susceptibility in diverse diseases. However, the accurate estimation of high-accuracy parameter maps from highly undersampled measurements remains a primary challenge in MRF. In this paper, we propose a novel MRF reconstruction framework leveraging manifold structured data priors. This approach models fingerprint signals and tissue quantitative parameters as data points residing on manifolds, and reveals the intrinsic topological consistency between the fingerprint manifold and the parameter manifold. Based on this key observation, we introduce a manifold structured data regularization constraint for MRF reconstruction. By enforcing structural consistency between the fingerprint manifold and the parameter manifold during reconstruction, the proposed constraint effectively improves reconstruction quality. Furthermore, to fully exploit the inherent data correlations, we integrate a locally low-rank prior into our reconstruction framework, which further enhances reconstruction performance. Experimental results demonstrate that the proposed method achieves significantly improved reconstruction quality with substantially reduced computational time compared to state-of-the-art methods, highlighting its promising potential for clinical translation in high-quality MRF imaging.

Key words: Magnetic resonance fingerprinting, Quantitative MRI, Manifold structured data, Locally low-rank