DMnet: A Reliable Magnetic Resonance Spectroscopy Quantification Network and the Verification on 5T scanners

Motivation: To create a robust and generalized artificial intelligent method for quantification of MRS. Goal(s): Our goal is to prove the reliability of the DMnet by illustrating its quantification results on data from 5T 1H-MRS in the brain. Approach: Conducting separate Bland-Altman analyses between DM and typical quantification methods before and after accelerated acquisition, and examining concentration changes under varying resolution voxel conditions. Results: The results indicate that DMNet and LCModel demonstrate high consistency in 8 healthy volunteers, with good agreement in quantified concentrations before and after acceleration by a factor of 2.58. Furthermore, the quantification results are more robust under varying resolution conditions. Impact: The DMnet can quantitate MRS more quickly and robustly, even in scenarios with lower SNRs. This method has been integrated into the CloudBrain-MRS platform for convenient one-click access by healthcare professionals, further aiding clinical treatments.