Imaging data analysis

Cross-sectional VBM processing To identify regional GMV differences between patients with OSA, before and after surgical treatment, and the healthy control group, structural T1-weighted images were processed using statistical parametric mapping (SPM8; http://www.fil.ion.ucl.ac.uk/spm; Wellcome Institute of Neurology, University College London, UK) and the VBM8 toolbox (http://dbm.neuro.uni-jena.de) with default settings as described in the manual. The procedure for the cross-sectional VBM pipeline followed that of previous cross-sectional based VBM studies from our group [26, 27] (additional details available in the Method of the online “Additional file 1”).

Longitudinal VBM processing The default longitudinal batch script in the VBM8 toolbox was used to identify longitudinal effects to GMV in patients with OSA before and after surgical treatment. In this pipeline the modulation step was not used because our focus was on relative tissue differences between different time-points within the same participant [27]. First, follow-up (after surgery) T1-weighted scan was registered to the baseline scan (before surgery). Second, the mean anatomic image was calculated using the realigned images and served as a reference image for realignment of baseline and follow-up scans for each participant. Third, the individual realigned baseline and follow-up scans were bias corrected to account for field inhomogeneities regarding the corresponding mean anatomical scan. Fourth, the resultant bias-corrected mean anatomical scan and realigned images were segmented into GM, WM, and CSF tissue segments using the VBM8 segmentation approach. Fifth, the DARTEL registration parameters were estimated using the GM tissue segments of the bias-corrected mean anatomical scan. The resulting registration parameters were applied to the corresponding tissue segments of the realigned baseline and follow-up anatomical scans. The resulting normalized GM segments for each time point for each participant were smoothed using an 8-mm FWHM Gaussian kernel and served as inputs for the subsequent longitudinal statistical model.