Single-Trial Analysis for Empirical MEG Data

Abstract

MEG/EEG single-trial data usually has low signal-to-noise ratio (SNR) because it is highly contaminated by background noise. Thus, it is too hard to detect sources and is hardly to distinguish between sources of our interest and sources of noise. To overcome this difficulty, time-locked averaged signal over many trials is used for source localization. However, averaging effect is more likely to filter out source activities which are slightly varied over trial. These varying sources may impose good and critical information in understanding brain dynamics. For this reason, we make attempt to analyze many single-trial data to investigate how brain activities over trial are going. We used conventional minimum-variance beamformer for source localization. We adopt bootstrap resampling technique to do various localization analysis between original single-trial analysis and fully averaged analysis. It is found from median nerve stimulation that some unseen sources in averaged data were frequently detected in a specific area. We believe it is evident that single-trial analysis enables to reveal more different dynamical behaviors than averaged one.