k ‐Space and Image Domain Hybrid Reconstruction Using Low‐Rank Plus Sparse Model (KILS) for Accelerated CEST MRI

As a promising in vivo metabolic imaging method, chemical exchange saturation transfer (CEST) MRI requires collecting a series of images using varied saturation frequencies ( ω ). The low‐rankness feature was utilized for acquisition acceleration and data denoising but primarily in the image domain. Herein, we aim to utilize such features in both k ‐space and the image domain, for achieving faster imaging and higher quality reconstruction. According to Parseval's theorem, CEST images at each ω have equal total energy to their k ‐space counterparts; that is, k ‐space signal along ω exhibited a similar valley shape as Z ‐spectra in the image domain, whereas the central k ‐space and peripheral regions reflect different spatial‐frequency components, corresponding to varied low‐rankness features. Because k ‐space series contain sparsity and redundancy complementary to the image domain, we proposed a hybrid k ‐space and image domain reconstruction for CEST MRI based on low‐rank plus sparse (KILS). Both retrospective and prospective experiments were conducted at 3 T, on a BSA phantom, 11 healthy volunteers and 15 glioma patients. KILS was also retrospectively evaluated in human liver at 3 T and rat brains at 9.4 T. In retrospective experiments of phantoms and human brains, KILS demonstrated the best quantitative metrics among all reconstruction methods, with acceleration factors (AFs) ranging from 2 to 8. The ablation study demonstrated that KILS well preserved both CEST contrast and image anatomy. Prospectively, a 31‐offset spectral scan of 2‐mm isotropic whole‐brain images was demonstrated, taking 5.7 min. Additionally, KILS proved effective in retrospective reconstruction for human liver at 3 T (AF = 6) and ischemic rat brain at 9.4 T (AF = 6). KILS demonstrated accurate and robust reconstruction of undersampled CEST MRI. The complementary nature of k ‐space and image domain may enable KILS to be readily applicable to multiple application scenarios.