Optimizing neuromodulation for temporal lobe epilepsy treatment based on a surrogate neural state model

Abstract

Temporal lobe epilepsy is the most prevalent form of medication-resistant epilepsy, and current electrical stimulation therapy has not been able to accomplish the goal of seizure-freedom. This underscores the need for a new target and a different approach with more effective neuromodulation for epilepsy treatment. The projections from the medial septum (MS) and its regulatory role on the hippocampus make it an attractive neuromodulation target. Optogenetics enables selective excitation or inhibition of individual genetically-defined neuronal subpopulations, and thus provides a chance to find a better target among neuronal subpopulations for inducing a greater therapeutic effect. I have exhaustively explored the effect of exciting or inhibiting different neuronal subpopulations in the normal rat medial septum by using optogenetic stimulation. As a result, MS optogenetic stimulation using hSynapsin promoter in combination with Channelrhodopsin-2 was well suited for modulating electrophysiological activity of the hippocampus. The conventional approach for preclinical studies requires a large amount of time and resources to find effective stimulation parameters and often fails due to the inter-subject variability in stimulation effect. As an alternative, I presented a novel data-driven approach which can optimize the neuromodulation more effectively and efficiently by investigating the stimulation effect on the surrogate neural state model. For the new approach, I implemented and demonstrated a variety of machine learning techniques to explore the stimulation effect, to describe the pathological neural states and to optimize the stimulation parameters. Specifically, first, I built a data-driven neural state model to estimate a seizure susceptibility based on electrophysiological recordings. The output of the model played a surrogate role by providing a metric which was regulated via the MS optogenetic stimulation. Second, I further increased the effectiveness of the stimulation by implementing in vivo Bayesian optimization which quickly finds the subject-specific optimal stimulation parameters. Finally, I tested whether modulating the surrogate neural state model affected the symptom of epilepsy (i.e. seizure). The treatment efficacy of the data-driven surrogate approach was compared to the stimulation with an empirically selected parameter set. The stimulation parameters to maximize the hippocampal theta (4-10Hz) power, which was a surrogate of the epileptic symptom, was more effective than the empirically selected parameter (7Hz) for the seizure suppression.