32. Spatio-temporo-spectral patterns of eeg and brain hemodynamics using bayesian approach

Abstract

Introduction BrO [1] introduced a Parallel factor analysis method (Parafac) which could blindly separate 3D data into components. This method could be applied to 3D spectrogram of EEG data to get natural spatio-temporo-spectral patterns of EEG spectrum (STSp) without any apriori spectral or spatial constrains. Temporal features of STSp could be then used to find brain regions with correlated brain hemodynamics in simultaneously acquired fMRI data. In this work we show the comparison between two methods for Parafac estimation, so far used Alternating least square (ALS) and our proposed method based on Variational Bayesian statistic (VB). The VB method is then applied to real EEG data from healthy controls to get natural STSp and brain regions with correlated hemodynamics. Methods We used a large set of simulated 3D data which were subjected to the Parafac estimation using either ALS or VB. The similarity between true and estimated components was computed for each dataset and both methods were compared. The simultaneous EEG-fMRI data were acquired from 52 healthy subjects. The Parafac was then applied using VB method to 3D spectrogram of EEG data concatenated along time dimension to get group-specific STSp. The temporal features of resulting STSp were then used to get brain regions with correlated brain hemodynamics. Results The similarity between true and estimated components was significantly higher for VB than for ALS. Using the real data from healthy subjects we found several STSp in all frequency bands and corresponding brain regions with significantly correlated hemodynamics. Conclusion The VB based Parafac is a powerful blind decomposition method which could draw natural STSp from EEG data with better performance than ALS. It could add valuable contribution to the studying of physiological relationship between EEG and hemodynamics. Acknowledgements The research was supported by the Grant GACR P304/11/1318.