New Physics-Informed Method Improves MRI Reconstruction.

Chemical Exchange Saturation Transfer (CEST) MRI offers valuable metabolic insights but is hampered by lengthy acquisition times, making it less practical for clinical use. While sparse sampling can reduce scan duration, reconstructing accurate, high-resolution Z-spectra from limited data is a challenging inverse problem. Existing methods, including generic Implicit Neural Representations (INRs), often fail to incorporate physical constraints, leading to artifacts and invalid signals. This paper proposes Lorentz Encoding (LE), a novel physics-informed framework that addresses these limitations. LE frames CEST reconstruction as a self-supervised task using implicit continuous coordinate learning. Instead of generic positional encodings, LE projects sparse coordinates into a physically constrained space defined by parametric Lorentzian profiles, which are learnable basis functions. This mechanism effectively reduces noise and ensures consistency with known physical models. Experiments on human brain data demonstrate that LE significantly surpasses current state-of-the-art techniques, achieving superior PSNR and SSIM even with highly sparse sampling. Furthermore, the learned encodings form a continuous, geometrically ordered latent space, which is crucial for accurate quantitative metabolite mapping.