An ocular artefacts correction method for discriminative EEG analysis based on logistic regression

Electrooculogram (EOG) contamination is a common critical issue in general EEG studies as well as in building highperformance brain computer interfaces (BCI). Existing regression or independent component analysis based artefacts correction methods are usually not applicable when EOG is not available or when there are very few EEG channels. In this paper, we propose a novel ocular artefacts correction method for processing EEG without using dedicated EOG channels. The method constructs estimate of ocular components through artefacts detection in EEG. Then, an optimization based on logistic regression is introduced to remove the components from EEG. Specifically, the optimization ensures that the discriminative information is maintained in the corrected EEG signals. The proposed method is offline evaluated with a large EEG data set containing 68 subjects. Experimental results show that, through the artefacts removal correction by the proposed method, EEG classification accuracy can be improved with statistical significance.