AUTHOR=Spurny Benjamin , Heckova Eva , Seiger Rene , Moser Philipp , Klöbl Manfred , Vanicek Thomas , Spies Marie , Bogner Wolfgang , Lanzenberger Rupert 

TITLE=Automated ROI-Based Labeling for Multi-Voxel Magnetic Resonance Spectroscopy Data Using FreeSurfer

JOURNAL=Frontiers in Molecular Neuroscience

VOLUME=12

YEAR=2019

URL=https://www.frontiersin.org/journals/molecular-neuroscience/articles/10.3389/fnmol.2019.00028

DOI=10.3389/fnmol.2019.00028

ISSN=1662-5099

ABSTRACT=<p><bold>Purpose</bold>: Advanced analysis methods for multi-voxel magnetic resonance spectroscopy (MRS) are crucial for neurotransmitter quantification, especially for neurotransmitters showing different distributions across tissue types. So far, only a handful of studies have used region of interest (ROI)-based labeling approaches for multi-voxel MRS data. Hence, this study aims to provide an automated ROI-based labeling tool for 3D-multi-voxel MRS data.</p><p><bold>Methods</bold>: MRS data, for automated ROI-based labeling, was acquired in two different spatial resolutions using a spiral-encoded, LASER-localized 3D-MRS imaging sequence with and without MEGA-editing. To calculate the mean metabolite distribution within selected ROIs, masks of individual brain regions were extracted from structural T<sub>1</sub>-weighted images using FreeSurfer. For reliability testing of automated labeling a comparison to manual labeling and single voxel selection approaches was performed for six different subcortical regions.</p><p><bold>Results</bold>: Automated ROI-based labeling showed high consistency [intra-class correlation coefficient (ICC) &gt; 0.8] for all regions compared to manual labeling. Higher variation was shown when selected voxels, chosen from a multi-voxel grid, uncorrected for voxel composition, were compared to labeling methods using spatial averaging based on anatomical features within gray matter (GM) volumes.</p><p><bold>Conclusion</bold>: We provide an automated ROI-based analysis approach for various types of 3D-multi-voxel MRS data, which dramatically reduces hands-on time compared to manual labeling without any possible inter-rater bias.</p>