Broadly applicable methods for the detection of artefacts in electroencephalography acquired simultaneously with hemodynamic recordings

Abstract

Electroencephalography (EEG) data, acquired simultaneously with magnetic resonance imaging (MRI), must be corrected for artefacts related to MR gradient switches (GS) and the cardioballistic (CB) effect. Canonical approaches require additional signal acquisition for artefact detection (e.g., MR volume onsets, ECG), without which the EEG data would be rendered uncleanable from these artefacts.•We present two broadly applicable methods for artefact detection based on peak detection combined with temporal constraints with respect to periodicity directly from the EEG data itself; no additional signals are required. We validated the performance of our methods versus the two canonical approaches for detection of GS/CB artefact, respectively, on 26 healthy human EEG-functional MRI resting-state datasets. Utilising various performance metrics, we found our methods to perform as well as - and sometimes better than - the canonical standard approaches. With as little as one EEG channel recording, our methods can be applied to detect GS/CB artefacts in EEG data acquired simultaneously with MRI in the absence of MR volume onsets and/or an ECG recording. The detected artefact onsets can then be fed into the standard artefact correction software.

Keywords: Artefact; Artifind; Ballistocardiac; Cardioballistic; ECG; EEG-fMRI; Gradient; Pulse.

Graphical abstract

Fig. 1

Our proposed methods for artefact detection. Left Panel: Gradient switch (GS) artefact detection. (A) An example segment of artefactual data from electrode C6 is shown. (B) An iterative threshold is applied, lowering the threshold on each iteration until a number of peaks equal to the expected number of artefacts can be automatically found and labelled as the artefact onsets (the red crosses as shown in (C)). These can then be entered into canonical software for artefact correction. Right Panel: CB artefact detection. (A) shows an example segment of artefactual data from electrode AF4. Two clearly discernible and consecutive peaks are first labelled. Using the information from these peaks, other peaks (i.e., artefact onsets) in the segment are automatically identified and a mean across all artefacts is taken (as shown in (B)). This mean artefact is convolved with the data segment (C) and peaks are automatically identified from the convolved signal (D). These are the artefact onsets that can then be inputted into canonical software for artefact correction.

Fig. 2

Distributions of estimated cardioballistic inter-artefact intervals: Canonical method versus our method. Per subject, a histogram of the first derivative of estimated cardioballistic artefact onsets is presented. In blue are those estimated by the canonical approach; in orange are those estimated by our approach. Across subjects there is a good overlap between the two approaches, showing a similar performance.