Towards a Real-time Seizure Detection Algorithm for Closed-loop Optogenetic Modulation of an Animal Model of Epilepsy

Abstract

Epilepsy is a highly prevalent disease affecting 50 million people worldwide, and 3 million domestically in United States of America. About one-third of the epileptic population does not respond to pharmacological treatment and are classified as medically refractory. Surgical intervention is the alternative solution for this population, however it is not effective in the whole population and leaves 10-15% of the patients deprived of relief from seizures. Deep Brain Stimulation is a novel treatment that is being investigated for this disease. In order to fully understand this treatment, one needs to be informed about the neural circuitry and downstream effects of the stimulation. A powerful investigation needs to be done in closed-loop fashion to tie the stimulation with onset of the seizure, without otherwise affecting the brain. This study evaluates the proper metrics for a real-time algorithm with high detection sensitivity and low latency for a closed-loop setup to be used in the experimental setups of epilepsy research. The study first investigates the previous features used for seizure detection, and implements Line-Length (LLN), Mean Power Spectral Density (MPSD) in 12-25 Hz and Maximum Cross Correlation in its algorithm. Offline performance evaluation of candidates identified LLN and MPSD as powerful features with high sensitivity and low detection latency, which could be implemented in future online algorithms for closed-loop experimental setup.