Rejecting deep brain stimulation artefacts from MEG data using ICA and mutual information

Abstract

Background: Recording brain activity during deep brain stimulation (DBS) using magnetoencephalography (MEG) can potentially help clarifying the neurophysiological mechanism of DBS. The DBS artefact, however, distorts MEG data significantly. We present an artefact rejection approach to remove the DBS artefact from MEG data.

New methods: We developed an approach consisting of four consecutive steps: (i) independent component analysis was used to decompose MEG data to independent components (ICs); (ii) mutual information (MI) between stimulation signal and all ICs was calculated; (iii) artefactual ICs were identified by means of an MI threshold; and (iv) the MEG signal was reconstructed using only non-artefactual ICs. This approach was applied to MEG data from five Parkinson's disease patients with implanted DBS stimulators. MEG was recorded with DBS ON (unilateral stimulation of the subthalamic nucleus) and DBS OFF during two experimental conditions: a visual attention task and alternating right and left median nerve stimulation.

Results: With the presented approach most of the artefact could be removed. The signal of interest could be retrieved in both conditions.

Comparison with existing methods: In contrast to existing artefact rejection methods for MEG-DBS data (tSSS and S(3)P), the proposed method uses the actual artefact source, i.e. the stimulation signal, as reference signal.

Conclusions: Using the presented method, the DBS artefact can be significantly rejected and the physiological data can be restored. This will facilitate research addressing the impact of DBS on brain activity during rest and various tasks.

Keywords: Artefact; Deep brain stimulation; Mutual information; Neuromodulation; Oscillatory activity; Parkinson’s disease.