Automatic Sleep and Wake State Detection in Rats Using EEG

Epilepsy is one of the leading neurological disorders characterized by recurrent seizures. These epileptic seizures are often associated with sleep stages. The longitudinal EEG studies in animal model of epilepsy have voluminous data, which is difficult to mark sleep stages manually. This demands an automated way to detect sleep and active states. We used continuous video-EEG recordings as the gold standard to characterize sleep-wake states in animals. We implemented frequency band power-based thresholding approach to detect sleep stages in 600 hours of 5 epileptic and naïve animal data. We observed that the conventional delta and theta band powers were prominently higher than the other frequency bands in the sleep-wake states however not efficient enough to detect sleep. We considered a set of novel frequency bands for their ability to robustly differentiate individual sleep states by using EEG only. We implemented k-nearest neighbor, logistic regression and brute force threshold approach. The KNN was the best predictor of sleep and wake states with 94% accuracy where the brute force was significantly similar accuracy (92%). We further characterized the sleep in epileptic animals and found that the epileptic animals sleep longer compared to naïve animals and also individual sleep durations are significantly longer than the naive animals. Such efficient algorithms can improvise and expedite sleep studies and understanding relationship between and epilepsy.