Removal of artifacts contaminated records from MEG records using a mixture of factor analyzers

Contamination of magnetoencephalogram (MEG) records by various artifacts seriously influences the estimation accuracy of magnetic sources. Although some methods have been proposed to reject contaminated records [1][2], most of them are ineffective when artifacts models are not available. In this study, we proposed to classify MEG records into contaminated records and others by mixture of factor analyzers (MFA). MFA provides a classification method which groups records into classes which are described by the model of factor analysis[3]. It is known that MFA is superior to conventional methods in its discrimination ability in a noisy environment [4]. It should be ideal in the classification of evoked responses on which large spontaneous waves are superimposed.We applied the proposed method to visual evoked fields (VEF) records.