Research PaperRemoving eye blink artefacts from EEG-A single-channel physiology-based method

- Eye blink removal method superior to popular independent component analysis methods (SOBI, Infomax and FastICA).
- Success rate of over 90% in terms of recovering the variance of the original EEG.
- Requires only a single channel at Fp1.
- Can be easily implemented and used for real-time processing in brain-computer-interfaces and neurofeedback training.

BackgroundEEG signals are often contaminated with artefacts, particularly with large signals generated by eye blinks. Deletion of artefact can lose valuable data. Current methods of removing the eye blink component to leave residual EEG, such as blind source component removal, require multichannel recording, are computationally intensive, and can alter the original EEG signal.New methodHere we describe a novel single-channel method using a model based on the ballistic physiological components of the eye blink. This removes the blink component, leaving uncontaminated EEG largely unchanged. Processing time allows its use in real-time applications such as neurofeedback training.ResultsBlink removal had a success rate of over 90% recovered variance of original EEG when removing synthesised eye blink components. Fronto-lateral sites were poorer (∼80%) than most other sites (92-96%), with poor fronto-polar results (67%).Comparisons with existing methodsWhen compared with three popular independent component analysis (ICA) methods, our method was only slightly (1%) better at frontal midline sites but significantly (>20%) better at lateral sites with an overall advantage of ∼10%.ConclusionsWith few recording channels and real-time processing, our method shows clear advantages over ICA for removing eye blinks. It should be particularly suited for use in portable brain-computer-interfaces and in neurofeedback training.