Dynamic causal modelling of electrographic seizure activity using Bayesian belief updating

Area of Science:

• Neuroscience
• Computational Neuroscience
• Epilepsy Research

Background:

• Seizure activity in EEG recordings exhibits complex spatiotemporal dynamics.
• Analyzing large datasets is crucial for understanding seizure evolution.
• Standard Dynamic Causal Modelling (DCM) for seizure analysis is computationally intensive.

Purpose of the Study:

• To develop an efficient Bayesian belief updating procedure for DCM within the existing framework.
• To enable faster and more accessible analysis of seizure activity from EEG/ECoG data.
• To investigate the spatiotemporal evolution of seizure dynamics.

Main Methods:

• Implemented a Bayesian belief updating scheme for DCM.
• Tested the scheme on simulated and empirical seizure data (invasive and non-invasive EEG/ECoG).
• Compared the updating method with standard Bayesian inversion procedures.

Main Results:

• The Bayesian belief updating scheme achieved similar accuracy to standard methods.
• Variance explained by the model showed minimal difference (<5%).
• The updating method was substantially more efficient, reducing analysis time from hours to minutes (5-10 min vs. 1-2 h).

Conclusions:

• The efficient DCM inversion method accurately characterizes seizure spatiotemporal evolution.
• This method facilitates the investigation of neuronal activity across different timescales.
• Enables deeper understanding of seizure generation mechanisms.