Time-evolving controllability of effective connectivity networks during seizure progression

Area of Science:

• Neuroscience
• Control Theory
• Computational Psychiatry

Background:

• Epilepsy affects 3 million people in the US, with over a third experiencing medication resistance.
• Responsive neurostimulation offers an alternative to resective surgery but requires personalized parameter optimization.
• Network neuroscience and control theory provide frameworks for understanding and controlling anomalous neural activity.

Purpose of the Study:

• To develop and apply a method for characterizing dynamic controllability across evolving effective connectivity networks during seizures.
• To investigate the relationship between average and modal controllability and seizure dynamics.
• To identify optimal timing and electrode targeting strategies for seizure suppression using neurostimulation.

Main Methods:

• Utilized regularized partial correlations from intracranial electrocorticography (ECoG) recordings to estimate effective connectivity (EC) networks.
• Employed the Graphical Least Absolute Shrinkage and Selection Operator (GLASSO) to derive adjacency matrices from 1-second time windows.
• Calculated average and modal controllability metrics from time-varying EC networks across seizure onset, propagation, and termination phases.

Main Results:

• Average controllability increased throughout seizures, while modal controllability decreased, showing an inverse relationship.
• The energy required to transition from an ictal to a seizure-free state was found to be lowest during seizure onset.
• Control energy application at seizure onset zones was not consistently energetically favorable, suggesting complex dynamics.

Conclusions:

• Time-varying controllability metrics offer insights into brain network dynamics during seizures.
• A low-complexity model of evolving controllability can inform the development of improved seizure suppression strategies.
• Personalized neurostimulation parameter selection can be enhanced by understanding dynamic network control properties.