Adaptive stimulation design for the treatment of epilepsy

Dr. Joelle Pineau, School of Computer Science, McGill. Electrical stimulation has recently emerged as a promising therapy for patients with medically intractable epilepsy, but little is known about the best stimulation patterns to use in order to get maximal seizure reduction while minimizing long-term damage to the brain. This project's overall goal is to automatically optimize a closed-loop strategy for the control of deep brain stimulation using reinforcement learning methods. I will present some of the key methodological challenges that must be addressed in order to automatically learn adaptive treatment strategies in this context. In particular, I'll discuss (1) the use of ensemble methods to automatically detect seizures, (2) the design of a computational model of epilepsy to provide synthetic training data for the reinforcement learning agent, (3) initial results of applying reinforcement learning using in vitro recordings. This is joint work with Massimo Avoli, Aaron Courville, Arthur Guez, Philip de Guzman and Robert Vincent.