Deep learning and eye-tracking for accurate EOG rejection
The electroencephalogram (EEG) directly measures the electrical activity generated by the brain. Unfortunately, it is often contaminated by various artifacts, notably those caused by eye movements and blinks (EOG artifacts). Such artifacts are usually removed using an independent component analysis (ICA) or other blind source separation techniques. However, it is difficult to assess whether subtracting EOG components estimated through ICA removes some neurogenic activity. It is crucial to address this question to avoid biasing EEG analyses. Toward that objective, we developed a deep learning model for EOG artifact removal that exploits information about eye movements available through eye-tracking (ET). Using a multimodal EEG and ET open-access dataset, we trained within-subject a long short-term memory (LSTM) model to predict the component of EEG signals predictable from ET data. We further used this ET-informed evaluation of EOG artifacts to investigate the sensitivity and specificity of ICA. Our analysis indicates that although ICA is very sensitive to EOG, it has a comparatively low specificity. These results motivate further research on EEG artifact removal to develop approaches with higher EOG rejection specificity.