Abstracts

Rationale: Traumatic brain injury (TBI) usually disrupts circadian rhythms and sleep. The ability to track changes in the microstructure of sleep in the post-traumatic period could help assess the effect of intervention and perhaps provide clues about the likelihood of epileptogenesis. However, convenient metrics that track sleep-wake dynamics over time beyond simplistic measures such as the percent time spent in each state (Wake, REM sleep, and non-REM sleep) or mean bout duration are lacking. Here, a methodology based on hidden Markov models (HMMs), estimated from physiological measurements in mice, is proposed for characterizing transient sleep-wake dynamics. Methods: With IACUC approval, adult C57BL/6J mice were implanted with EEG/EMG preamplifiers and monitored round-the-clock within hours of surgery for 3-4 weeks with brief weekly interruptions for cage cleaning. The HMM, an unsupervised probabilistic model, was used to sequence time series of EEG/EMG features in 4s epochs into REM, non-REM, and Wake states. The features supplied to the HMM were the delta/theta power ratio of the EEG and the root-mean-squared EMG. The HMM is parameterized by a 3 3 state transition matrix (STM) representing the probabilities of different Markov state transitions. HMMs were re-estimated from EEG/EMG feature sets every four hours. The output of the HMM was validated for sample recordings against manual scores. Trends in HMM properties were inspected to characterize progressive changes in behavior following surgery. Results: : HMMs were found to track instantaneous sleep-wake state with over 90% accuracy from continuous EEG/EMG measurements (92% sensitivity and 95% specificity; n = 4 mice) when compared to manual scores. Two parameters extracted from the HMM, the probability of Wake (P_w) and the trace (T_r) of the STM, were used as measures of sleep quality and the persistence of any ongoing state respectively. Immediately after implantation, P_w was abnormally low, as expected after general anesthesia and mild head trauma associated with the EEG surgery. Although low P_w indicates increased somnolence, T_r was low as well, which suggests more fragmented sleep. P_w and T_r took 7-10 days to reach stable levels consistent with full recovery, with patterns characteristic of normal sleep-wake cycles and circadian rhythms.Conclusions: Preliminary results suggest that HMMs estimated from physiological measurements could provide quantitative markers of transient behavior and recovery from brain injury. We intend to use this approach to track physiological and behavioral changes in a model of post-traumatic epilepsy. Acknowledgement: This work was supported in part by grants from the National Institutes of Health (NS065451) and the Kentucky Spinal Cord and Head Injury Research Trust (KSCHIRT; 10-5A).