Learning Implicit Neural Representation with Explicit Physical Priors for Accelerated Quantitative $T1ρ$ Mapping

Motivation: Magnetic resonance $T1ρ$ mapping provides critical insights into tissue properties for early disease detection, but its clinical use is hindered by long scan times needed for acquiring multiple $$T1ρ$-weighted images. Goal(s): This study proposes an unsupervised implicit neural representation (INR) framework for precise $$T1ρ$ map generation. Approach: A subject-specific unsupervised method that learns an implicit neural representation of the $$T1ρ$-weighted images, simultaneously capturing the relationships among $$T1ρ$-weighted images and multi-channel $$k$-space data. Results: LINEAR achieves 14-fold acceleration with high accuracy in $$T1ρ$ map generation, outperforming state-of-the-art unsupervised and traditional methods in artifact suppression and error reduction. Impact: This study enables accelerated, high-quality $$T1ρ$ mapping, improving diagnostic efficiency and providing a foundation for future advancements in rapid quantitative imaging, with potential applications across diverse clinical and research fields.