In connectivity studies using EEG data, it is imperative to work on reconstructed sources in order to obtain interpretable results. Due to the illposed nature of the EEG inverse problem, however, source estimates highly depend on reconstruction parameters. In order to establish source connectivity as a meaningful research tool, consistency across estimation pipelines would be mandatory. We investigated reproducibility of EEG source location and connectivity across common forward models (BEM, FEM, spherical harmonics) and inverse models (WMNE, LCMV and eLORETA) using Fieldtrip, Brainstorm, and our own toolbox. Several pipelines of source reconstruction were carried out using the ICBM template anatomy on resting state (eyesclosed) EEG data (100 subjects, 5 min, 49 electrodes). For each pipeline, we estimated the location of the sources by computing the alphaband SNR. We moreover employed the imaginary part of coherency (iCoh) and the phase slope index (PSI) as measures of functional and effective connectivity in a 5 Hz band around the alphapeak. All analyses were restricted to 10 ROIs identical to the brain lobes by averaging results within ROIs. Consistency of localization and connectivity was quantified by calculating pairwise correlations between the localization and connectivity vectors obtained from different pipelines on the same data, and averaging across subjects. We observed medium to high consistency between forward models and toolboxes (0.4 < r < 0.95). Connectivity results were more dependent on these factors than source localization. All inverse solutions led to similar localizations (r > 0.75). However, while WMNE and eLORETA yielded similar connectivity estimates (r > 0.5 for iCoh, r > 0.7 for PSI), substantially different results were obtained using LCMV. Our results demonstrate a considerable withinsubject variability across source estimation pipelines that must be taken into account when reporting and discussing source connectivity results.
Influence of source estimation parameters on EEG source location and connectivity
MAHJOORY, KEYVAN;FATO, MARCO MASSIMO;
2016-01-01
Abstract
In connectivity studies using EEG data, it is imperative to work on reconstructed sources in order to obtain interpretable results. Due to the illposed nature of the EEG inverse problem, however, source estimates highly depend on reconstruction parameters. In order to establish source connectivity as a meaningful research tool, consistency across estimation pipelines would be mandatory. We investigated reproducibility of EEG source location and connectivity across common forward models (BEM, FEM, spherical harmonics) and inverse models (WMNE, LCMV and eLORETA) using Fieldtrip, Brainstorm, and our own toolbox. Several pipelines of source reconstruction were carried out using the ICBM template anatomy on resting state (eyesclosed) EEG data (100 subjects, 5 min, 49 electrodes). For each pipeline, we estimated the location of the sources by computing the alphaband SNR. We moreover employed the imaginary part of coherency (iCoh) and the phase slope index (PSI) as measures of functional and effective connectivity in a 5 Hz band around the alphapeak. All analyses were restricted to 10 ROIs identical to the brain lobes by averaging results within ROIs. Consistency of localization and connectivity was quantified by calculating pairwise correlations between the localization and connectivity vectors obtained from different pipelines on the same data, and averaging across subjects. We observed medium to high consistency between forward models and toolboxes (0.4 < r < 0.95). Connectivity results were more dependent on these factors than source localization. All inverse solutions led to similar localizations (r > 0.75). However, while WMNE and eLORETA yielded similar connectivity estimates (r > 0.5 for iCoh, r > 0.7 for PSI), substantially different results were obtained using LCMV. Our results demonstrate a considerable withinsubject variability across source estimation pipelines that must be taken into account when reporting and discussing source connectivity results.
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
