| Zugriffsnummer | 51026 |
| Dokumenttyp | Zeitschriftenartikel |
| Peer Review | mit Peer Review |
| Sprache | Englisch |
| Titel | How to avoid measurement of spurious inter-regional functional connectivity from EEG - a simulation study |
| Autor(in); Institution |
Pellegrini, F.; Charité – Universitätsmedizin Berlin, Berlin Center for Advanced Neuroimaging (BCAN), Berlin, GERMANY
Nikulin, V.; National Research University Higher School of Economics, Moscow, RUSSIAN FEDERATION; Max Planck Institute for Human Cognitive and Brain Sciences, Leipzig, GERMANY
Haufe, Stefan; 8.4, Mathematische Modellierung und Datenanalyse, PTB-Berlin
|
| Quelle/Jahr | Clinical Neurophysiology: 137 (2022), e60 - e61 |
| Artikelnummer | P 80 |
| ISSN | 1388-2457 (print) ; 1872-8952 (online) |
| Verlag | Amsterdam: Elsevier |
| Zusammenfassung | Introduction: Aggregating statistical dependencies between multivariate time series is important to characterise functional connectivity (FC) between brain regions. However, it is still unclear how to reliably detect true FC from source-reconstructed M/EEG data. This study generates ground truth data and compares the performances of various pipelines that estimate directed FC (Time-Reversed Granger Causality TRGC1) and undirected linear FC (imaginary part of coherency, absolute part of coherency and multivariate interaction measure MIM2) between regions. Material and Methods: To simulate EEG-like sensor signals, we proceeded as follows: 1. Ground truth source activity was generated as white noise time series filtered in the alpha band. Between one and five pairs of sources interacted with certain time delays. 2. After adding individual 1/f noise, the source signals were projected to sensor space. 3. White sensor noise was added. Afterwards, sensor data were projected to source level by applying one of four tested inverse solutions: Dynamic imaging of coherent sources (DICS), Linearly Constrained Minimum Variance source projection (LCMV), eLORETA, and Champagne. All tested connectivity analysis pipelines calculate one connectivity score for every region combination. The pipelines‘ ability to detect the true interactions was evaluated by percentile rank. Results: The best-performing FC pipeline consists of the following steps: 1. Source projection with a beamformer inverse solution. 2. Principal component analysis within every region. 3. Selection of a fixed number of strongest principal components as basis for further analysis. 4. Calculation of the MIM for every region pair in case of undirected FC and calculation of the TRGC in case of directed FC. Worst performance was obtained with pipelines estimating undirected FC with the absolute value of coherency. DICS source projection resulted in good detection of undirected FC, but failure to detect the direction of FC with TRGC. |
