1H-MRS can quantify brain metabolites noninvasively. However, in a typical clinical setting, human brain spectra are indispensably degraded due to low SNR, line-broadening, and unknown spectral baseline, and consequently, quantification of brain metabolites is challenging even with the current state-of-the-art software. Given the recent accomplishment of deep-learning in a variety of different tasks, we developed a convolutional-neural-network (CNN) that maps the degraded brain spectra into noise-free, line-narrowed, baseline-removed, metabolite-only spectra. The robust performance of the proposed method as validated on both simulated and in vivo human brain spectra strongly supports the potential of deep learning in 1H-MRS of human brain.
This abstract and the presentation materials are available to members only; a login is required.