Accelerating CEST MRI using Data-Driven Z-Spectral Compressed Sensing

Motivation: CEST MRI requires the collection of multiple saturated images with different saturation offsets, resulting in prolonged scan times, which hinders its clinical applications. Goal(s): we aim to reduce the scan time of CEST MRI by recovering the undersampled Z-spectrum to full-sampling counterpart using data-driven Z-spectral compressed sensing method. Approach: The modified U-Net was employed for Z-spectral recovery. Training data were generated using Bloch equation. Numerical simulations and in vivo experiments on rat brains were conducted to validate the proposed method. Results: The results demonstrate that our method outperformed conventional interpolation methods, and threefold undersampling rate can be achieved without discernible degradation in quantification. Impact: The proposed method can efficiently reduce the scan time of CEST MRI, potentially facilitating its clinical applications.