Multimodal detection of head-movement artefacts in EEG

Artefacts arising from head movements have been a considerable obstacle in the deployment of automatic event detection systems in ambulatory EEG. Recently, gyroscopes have been identified as a useful modality for providing complementary information to the head movement artefact detection task. In this work, a comprehensive data fusion analysis is conducted to investigate how EEG and gyroscope signals can be most effectively combined to provide a more accurate detection of head-movement artefacts in the EEG. To this end, several methods of combining these physiological and physical signals at the feature, decision and score fusion levels are examined. Results show that combination at the feature, score and decision levels is successful in improving classifier performance when compared to individual EEG or gyroscope classifiers, thus confirming that EEG and gyroscope signals carry complementary information regarding the detection of head-movement artefacts in the EEG. Feature fusion and the score fusion using the sum-rule provided the greatest improvement in artefact detection. By extending multimodal head-movement artefact detection to the score and decision fusion domains, it is possible to implement multimodal artefact detection in environments where gyroscope signals are intermittently available.