Semi-synthetic EEG Data for the Evaluation of Linear EEG Cleaning Methods

Abstract

Electroencephalography (EEG) data recordings can be contaminated by artefacts that reduce the quality and make analysis difficult, and therefore cleaning methods are essential for accurate analysis of EEG data. It is not yet well established how to measure performance based on measured contaminated data since there is no established benchmark for comparison. Here we use “clean” EEG data synthetically contaminated by electrocardiography (ECG), electrooculography (EOG) and electromyography (EMG). This introduces fewer assumptions to the comparison between various cleaning methods, providing a clear datum for comparison. Further contamination is controlled, adding artefacts individually and also as a combination of artefacts. The results show that signal to noise ratio (SNR) of the simulated artefacts was within the same ranges as found with measured artefacts from literature. Popular linear cleaning methods were evaluated on the dataset, showing similar results to those in the literature, further validating the usefulness and accuracy of the semi-synthetic dataset. The semi-synthetic dataset showed comparable characteristics to real measured EEG data and proved useful in the assessment of EEG cleaning methods. The cleaning methods showed varied results when performance was evaluated on individual artefacts.