Advancing In Vivo Molecular Bioimaging With Optimal Frequency Offset Selection and Deep Learning Reconstruction for CEST MRI.

Chemical exchange saturation transfer (CEST) magnetic resonance imaging (MRI) is an emerging non-invasive molecular imaging technique offering significant potential for biomedical research and clinical applications. However, CEST MRI data acquisition requires prolonged scanning times, as data need to be collected at multiple frequency offsets to capture necessary information for accurate analysis of biological compounds. Faster CEST MRI will improve molecular imaging, advancing biomedical pre-clinical research studies and clinical applications. Thus, herein, we accelerate CEST MRI data acquisition using a two-step approach. Firstly, we use an optimization algorithm to identify a set of optimal sparse frequency offsets for data collection. Secondly, we apply a deep learning algorithm to reconstruct the high-resolution CEST MRI Z-spectra from the low-resolution Z-spectra. CEST MRI data acquired on adult mice brains (n = 19) were utilized. The optimization technique efficiently selected down to 10% of the total frequency offset points, and the deep learning algorithm accurately reconstructed dense Z-spectra. The performance metrics, root mean square errors (RMSE), mean absolute error (MAE), and Pearson's correlation were calculated for various Z-spectra reconstructions. The minimum, maximum, and average RMSE values achieved when the lowest 10% of frequency offsets were used were 0.0065, 0.0133, and 0.0094, respectively. The proposed CEST MRI approach, involving optimal frequency offset selection followed by deep learning reconstruction, achieves an acceleration by 10 times while maintaining high-quality data. This approach expands the applications of CEST MRI, potentially advancing