Dynamic Modeling and Classification of Epileptic EEG Data

Abstract

Brain functional connectivity has been used to investigate the interaction between brain regions. It provides important information related to brain diseases, injuries, and high level cognitive functions. Statistical methods have been widely used to model brain functional connectivity based upon which insights of brain function are expected to be revealed. Most statistical approaches were developed based upon an assumption that connectivity patterns are static during the recording. This is not true because the connectivity changes over time. A dynamical modeling of connectivity patterns allows to characterize these variations. In this work, a simplified dynamic Bayesian modeling approach, parallel Hidden Markov Model (PaHMM), was investigated by characterizing temporal variations of cortical functional connectivity patterns computed using epileptic electroencephalogram (EEG) data. The performance of the PaHMM was evaluated based on an experimental study of epilepsy detection and classification, where multisubject epileptic EEG data from Temple University Hospital EEG Data Corpus were used. Experimental results show that an accuracy of 93.5% was obtained for the epilepsy detection, and an overall accuracy above 81% was achieved for the seizure type classification. This indicates that the method can efficiently capture temporal variations of functional connectivity patterns, and is potentially applicable in clinical settings to detect epilepsy and differentiate seizure types.