EEG artefact detection and removal
- Liam Marnane
- Simon O'Regan
Electrical signals which arise from sources other than the brain that are detected in EEG are known as EEG artefacts. EEG activity can be severely contaminated by artefacts such as eye movement, ECG, electrode pop, EMG, among others. These artefacts constitute a serious problem for the interpretation and analysis of the EEG signal by a clinician, and generally reduce the accuracy of seizure detection algorithms.
The aim of this project is to develop real-time EEG artefact detection and removal algorithms.
These algorithms may prove useful in the following applications:
- EEG monitoring by a clinician
- automatic seizure detection in traditional EEG monitoring system
- automatic seizure detection in ambulatory EEG system
The lack of availability of an extensive labelled EEG artefact database, on which to train and test artefact removal algorithms, is a major obstacle in the development of automatic artefact detection techniques. An artefact generation protocol was drawn up which instructed the participants to perform repetitions of over 32 separate actions that result in disturbances in the EEG (e.g. eye blink, head movement, chewing). Particular focus was placed on movement artefacts which have received very little attention in the literature and which will play a major part in an ambulatory EEG system. A database of over 300 minutes of EEG artefact data, using 20 participants has been collected. The collection process was video-recorded in order to improve the EEG annotation accuracy.
Current and Future Work
At present, optimum epoch length for each type of artefact is being investigated. Additionally, annotation of the EEG artefact database is continuing.
Future work will aim to develop automatic artefact detection and removal algorithms, using Support Vector Machines. These algorithms will then be used independently to flag artefacts in EEG examined by a clinician, or alternatively as a pre-processing stage in a seizure detection system.