Modeling the Complex Dynamics and Changing Correlations of Epileptic Events

Area of Science:

• Computational Neuroscience
• Biostatistics
• Medical Signal Processing

Background:

• Epilepsy patients exhibit both clinical seizures and sub-clinical epileptic bursts.
• Quantitative analysis of the relationship between these events is lacking.
• Intracranial EEG (iEEG) data presents challenges due to variable electrode configurations.

Purpose of the Study:

• To develop a quantitative method for parsing complex epileptic events into distinct dynamic regimes.
• To investigate the relationship between sub-clinical bursts and clinical seizures.
• To address the variability in iEEG electrode number and placement.

Main Methods:

• Developed a Bayesian nonparametric Markov switching process.
• Incorporated shared dynamic regimes across a variable number of channels.
• Utilized a Markov-switching Gaussian graphical model for channel dependencies.
• Enabled asynchronous regime-switching and an unknown dictionary of dynamic regimes.

Main Results:

• The model successfully parses iEEG data into distinct dynamic regimes.
• Demonstrated the model's effectiveness in out-of-sample predictions.
• Generated intuitive state assignments for iEEG data.
• Enabled comparison between sub-clinical bursts and clinical seizures.

Conclusions:

• The novel Bayesian approach effectively analyzes complex epileptic events in iEEG.
• The method automates clinical seizure analysis and facilitates comparison of different epileptic event types.
• This work provides new insights into seizure dynamics and patient-specific iEEG patterns.