Comparative evaluation of a deep learning method QNet and LCModel for MRS quantification on the cloud computing platform CloudBrain-MRS

Reliable magnetic resonance (MR) spectroscopy (MRS) quantification is key to accurate clinical diagnosis. This study aimed to statistically compare the metabolite quantification of human brain MRS between the deep learning method QNet and the classical method LCModel via an easy-to-use intelligent cloud computing platform CloudBrain-MRS. In this retrospective study, 15 healthy volunteers (12 females and 3 males, age range: 21–35 years, mean age ± standard deviation: 27.4 ± 3.9 years) were recruited. In September and October 2021, two 3 T MRI scanners each collected 61 in vivo 1H MR spectra from the brain region of pregenual anterior cingulate cortex of the healthy participants. The analyses of Bland-Altman, Pearson correlation and reasonability were performed to assess the degree of agreement, linear correlation, and reasonability, respectively, between the two quantification methods. The analyses of Bland-Altman, Pearson correlation and reasonability showed very high to moderate consistency (relative half interval of limits of agreement = 3.04%, 9.31%, and 18.50%, respectively) and very strong to moderate correlation (Pearson correlation coefficient r = 0.78, 0.93, and 0.47, respectively) between the two methods for quantifying total N-acetylaspartate (tNAA), total choline (tCho), and inositol (Ins). In addition, the quantification results of QNet were generally closer to the previously reported average values than those of LCModel. There were very high to moderate degrees of consistency between the quantification results of QNet and LCModel for tNAA, tCho, and Ins, and QNet generally had more reasonable quantification than LCModel.