Moving average template subtraction to remove stimulation artefacts in EEGs and LFPs recorded during deep brain stimulation

Deep brain stimulation (DBS) is a well established therapy to treat movement disorders such as Parkinson's disease. More recently it has also been discussed as a therapy for certain psychiatric diseases. However, during active DBS the recordings of local field potentials (LFP) and the electroencephalogram (EEG) can be corrupted by substantial spike-type artefacts which need to be removed before any analysis. Here, we present a new method that we term ' Moving Average Subtraction' (MAS) that removes DBS artefacts by subtracting adaptive DBS artefact templates from the artefact contaminated data. In particular we have developed a resampling technique which is more efficient than upsampling for a precise reconstruction of the artefact shape without the need to oversample the EEGs. By applying this method we can derive undistorted signals even in case of the low sampling frequencies that are usual in clinical recordings. We applied the new technique to 12 data sets recorded at the surface and in various brain structures [subthalamic nucleus (STN), pedunculo pontine nucleus (PPN), Globus pallidus internus (GPi)] with 7 patients. Our results demonstrate the suppression of artefact related activity at the basic and harmonic frequencies of DBS. The new technique outperforms the non-adaptive template subtraction technique for the removal of high frequency artefact residuals without producing the spectral dips that occur with notch filter approaches. The new technique facilitates the analysis of higher frequency bands (Gamma activity) in LFPs and EEGs recorded during active DBS.