Preprocessing of Resting-State fMRI Data

Preprocessing was conducted using SPM12 again separately for both samples but following the same protocol. In a first step, we discarded the first 10 volumes from each resting-state session due to known stabilization effects of the magnetic field. All remaining EPI volumes were realigned to the first volume and subsequently coregistered to the high-resolution T1-weighted images in native space. None of the scanned subjects showed excessive head motion (translation: >3 mm; rotation: >3°). We further tested whether diagnostic groups (HC vs. MCI) showed differences in head motion. To this end we computed the average frame-wise displacement for each individual following a previously described protocol (Power et al., 2014). When comparing the average frame-wise displacement across diagnostic groups using a two-sample t-test, we found no group differences between HC and MCI [ISD: t(54) = -0.790, p = 0.433; TUM: t(39) = -1.664, p = 0.104]. For spatial normalization to MNI space, the non-linear DARTEL and affine registration parameters that were estimated during preprocessing of the T1-weighted images were combined and applied to the coregistered EPI volumes. All EPI images were subsequently smoothed using an 8 mm FWHM Gaussian kernel, detrended and band-pass filtered, using a frequency band of 0.01–0.08 Hz. We further regressed out the 6 motion parameters (3 translations, 3 rotations) and the BOLD signal averaged across the WM and CSF masks that were created during preprocessing of the T1-weighted images. We did not apply global signal regression since it can artificially introduce anti-correlations in the BOLD signal (Murphy et al., 2009; Murphy and Fox, 2016).