Journal article Open Access

HArtMuT - Modeling eye and muscle contributors in neuroelectric imaging

Harmening, Nils; Klug, Marius; Gramann, Klaus; Miklody, Daniel

Electroencephalography (EEG) and Magnetoencephalograhy (MEG) data are mostly mixed with activity stemming from the eyes and muscles. Neuroscientific researchers decrease the effect of these non-brain sources on the recorded data using artefact cleaning methods operating in sensor space (e.g. filtering, subtracting signals from additional eye or neck electrodes) or by removal of automatically or manually identified independent or principal components (ICs/PCs). The subsequent source localization techniques are commonly applied to data that was cleaned from such physiological sources and other non-physiological artefacts. However, this chronological procedure leads to distorted EEG time series due to imperfect artefact reduction routines,  introduces a bias into the subsequent source localization and excludes potentially interesting additional information.
Therefore, we propose adding sources that are usually considered artefactual to the head model in a similar way as it is done for cortical sources. Treating muscles and eyes as proper contributors to EEG potentials allows for a more precise identification of these sources that can then be ignored or excluded automatically for further data analyses.

We developed a head artefact model using tripoles (HArtMuT) - a volume conduction head model with cortical dipole sources enhanced by symmetric dipoles for the eyes and tripoles for face and neck muscles. HArtMuT can be used for modeling eye and muscle contributors to the EEG signal, cortical and artefactual source reconstruction and for evaluating or constructing algorithms.

We compared different artefact modelling approaches using physiologically motivated dipolar and tripolar source models. Their performance was evaluated with respect to source localization accuracy using as ground truth both simulated HD-FEM data and ICA patterns from EEG recordings of 19 subjects completing different body rotation tasks.

The best model for neural sources were found to be the standard equivalent dipole, while the eyes were better approximated by symmetric dipolar models. The muscular contributors were also well in accordance with a dipolar model but a better approximation of source location was achieved with tripoles. Preliminary results show a possible usage as classifiers of the resulting source locations for identifying neural and artefactual sources.
Compared to using standard EEGLAB 3-shell models on real ICA-decomposed data, the residual variance was reduced from median 0.25 to 0.11 for muscle sources and 0.18 to 0.09 for eye sources.
68% of the ICA patterns were detected as muscle sources based on source location in the 4-shell HArtMuT, while 30% were brain and around 2% symmetric eye sources.

this is a preprint to be published, soon.
Files (9.2 MB)
Name Size
9.2 MB Download
All versions This version
Views 103103
Downloads 105105
Data volume 985.4 MB985.4 MB
Unique views 9393
Unique downloads 9494


Cite as