State-Dependent Stimulation for Seizure Termination in a Morris-Lecar Network: A Reinforcement Learning Approach

In a sparsely connected network of 100 extended three-dimensional Morris-Lecar neurons, two learned state-dependent stimulation policies markedly reduce median seizure duration versus trigger-only and state-independent controls.

• Why it matters: Neurostimulation success depends on the epileptic network’s dynamical state, but effective timing rules that adapt online to that state are difficult to design.
• What they did: The authors build a dynamical-systems-informed semi-gradient Q-learning agent with tile coding, using a two-dimensional time-delay embedding from either population-averaged membrane potential or a derivative-based extracellular potential proxy to choose stimulation or none.
• The result: Across 30 independently generated test networks, both learned policies markedly shorten seizure-like activity, and their comparable performance suggests the extracellular potential proxy preserves enough dynamical information for control in this model.

AI summary of the abstract · may contain errors · not medical advice · read the paper.