Recently an automatic EEG-assisted retrospective motion correction (aE-REMCOR) method that utilizes EEG data to correct for head movements in fMRI on a slice-by-slice basis was developed.^1,2 The aE-REMCOR was shown to be capable of substantially removing head movements in the fMRI dataset.¹ Here we examined the utility of this method to improve rs-fMRI connectivity analysis.

The aE-REMCOR employs independent component analysis on the preprocessed EEG data and automatically identifies independent components (ICs) related to head motions for fMRI data correction with AFNI³ 3dTfitter. The simultaneous EEG&fMRI were conducted on a GE MR750 3T MRI with an 8-channel head coil and 32ch MR-compatible EEG system (Brain Products GmbH). For fMRI, an EPI sequence was used (TR/TE=2000/30 ms, FOV/slice thickness/gap=240/2.9/0.5 mm, 34 axial slices, SENSE=2, and voxel =1.875×1.875×2.9 mm³, scan time 526 s). First 3 EPI volumes (6 s) were excluded to allow the fMRI signal to reach steady state.

All fMRI analysis was performed in AFNI. When aE-REMCOR was applied, it was employed before a series of standard fMRI preprocessing steps (slice-timing correction, volume registration, and registration to the Talairach coordinates⁴). Then the fMRI data was spatially blurred to 4-mm FWHM and temporally band-pass filtered (0.01-0.08Hz). Functional connectivity of the default mode network (DMN) in 86 resting scans of 16 subjects was examined using GLM-based correlation analysis. A spherical seed ROI with radius 5 mm was centered at the posterior cingulate cortex (PCC) (Talairach coordinate: (0,-51,22)⁵) (green cross-hairs in Figs.2-4). Nuisance covariates included the average signals of the cerebrospinal fluid, white matter, and 6 rigid body motion parameters. Correlation difference between the data with and without aE-REMCOR was evaluated. We also quantified the correlation changes at the nodes of the medial prefrontal cortex (mPFC: (0,49,2)), lateral parietal cortex (LatPar-L: (-45,-60,32), LatPar-R: (43,-60,29)), and hippocampal formation (HF-L: (-22,-19,-15), HF-R: (22,-19,-15)) of the DMN⁵.

Fig.1 depicts the distributions of the motion severity (f) and κ_f for each scan, where f = mean(|D''|), κ_f = kurtosis(D''), D is the maximum displacement of the voxels for each brain volume, and D'' is the second derivative of D with respect to TR (the average acceleration of the head motions during the scan). Prolonged periods of rapid head movements cause a more Gaussian distribution of D'', a large f and small κ_f. Occasional rapid head movements give a moderate f and a large κ_f. The absence of significant head movements gives a small f and a small κ_f. Fig.2 plots the single subject correlation map in a scan with some occasional rapid head movements (f =0.024, κ_f=57.5). In Fig.2a, stripes are observed in the correlation map without aE-REMCOR. These stripy features originate from the signal loss in neighboring imaging slices caused by the rapid head motions. When aE-REMCOR is applied, the contrast of the stripes reduces (Fig.2b). Fig.3 plots the group correlation difference calculated with and without the aE-REMCOR procedures with occasional (κ_f >40) and prolonged (f >0.10) rapid head movements (Fig.1). Similar motion-induced stripes can be observed. The stripes disappear when all the 86 resting scans are considered (Fig.4c). This shows the reduced significance of the motion artifacts on the DMN connectivity when sufficient samples are considered. For the motion group with κ_f >40, correlation at PCC was decreased by 0.017 (p=0.004), and correlation at HF-R was increased by 0.019 (p=0.018). For the motion group with κ_f >40 or f >0.10, correlation at PCC was decreased by 0.012 (p=0.0027). For the whole group analysis, correlations at PCC and LatPar-L were decreased by 0.003 (p=0.028) and 0.006 (p=0.024), and correlations at HF-L and HF-R were increased by 0.006 (p=0.029) and 0.008 (p=0.006). In all cases, a slight decrease was observed in the correlation at the seed ROI.The default mode network connectivity was analyzed with and without aE-REMCOR in 86 resting scans. The aE-REMCOR reduces the position-dependent error in the DMN connectivity measurement. The results show the importance of the slice-by-slice fMRI motion corrections to improve the accuracy of the functional connectivity analysis, especially when the entire group of subjects exhibits significant rapid head motions.