Modeling brain desynchronization by EEG sensor variance in epileptic patients
The use of electroencephalogram (EEG) for predictive purposes of seizures in epileptic patients has grown steadily with the access to greater computing power. Methods of seizure analysis to date have focused on modeling and computer aided machine learning to help increase sensitivity and specificity of seizure detection. Brain synchronization between various areas of the brain at the onset of seizures tends to be a common feature of seizures, followed by a resynchronization of the various brain areas at the end of seizures. While previous methods have looked at the cross-correlation or lag-correlation of only two areas the brain, most EEG data these days has a vast array of sensors that can easily exceed 15-20 areas of the brain. The goal of this research is to take a relatively new approach to statistical modeling of multivariate EEG data, by use of the variance of multiple sensors as an extended measure of brain desynchronization in a time series format. Use of the Children's Hospital Boston and Massachusetts Institute of Technology (CHB-MIT) scalp database from PhysioNet is used to demonstrate the potential effectiveness of modeling multivariate EEG data by assessing the overall variance between sensors for three young patients with intractable seizures by use of a DCC-GARCH model, Bayesian regime switching mixture model, and a Bayesian change-point model.
Craig Michael Krebsbach,
"Modeling brain desynchronization by EEG sensor variance in epileptic patients"
Dissertations and Master's Theses (Campus Access).