A Self-Supervised Deep Learning Method for Accelerated Quantification of High-Resolution Short-TE spectroscopic MRI Datasets

Motivation: 3D-spectroscopic MRI generates metabolite maps that are advantageous for spatially heterogeneous neurological disorders. The conventional spectral fitting algorithms use methods that are computationally burdensome, making them impractical for clinical workflow. Goal(s): To address the computational bottleneck of conventional spectral quantification methods in the clinical workflow. Approach: This study included 89 short-TE sMRI scans from two clinical trials. Comparative analysis was performed between NNFit and parametric modelling spectral quantitation method (FITT). Results: This study demonstrates that deep learning approach offers comparable performance to conventional quantification methods, but with much faster processing times and improved robustness to baseline variations at short-TE. Impact: A deep learning method for accelerated quantification of spectroscopic MRI datasets with short echo time, NNFit, achieved competitive quantitative performance in comparison to a standard parametric modelling spectral analysis method with greater computational efficiency.