Abstracts

Rationale: Recent studies show that high frequency oscillations (HFOs) can help delineate the epileptogenic zone in individuals with refractory epilepsy. However, HFO incidence is known to vary with vigilance state, which makes the choice of an appropriate diagnostic baseline for spatiotemporal analysis of HFO activity an important issue. Furthermore, since detection of individual HFOs requires a high data sample rate (> 2000 Hz) and computation at millisecond timescales, it is desirable to have a surrogate measure that is easier to compute and correlates with HFO incidence in both spatial and temporal domains. An obvious choice for such a measure is an estimator of the instantaneous power in the frequency range in which HFOs reside. In this study, we evaluated alternative measures of relative HFO activity and then examined the consequences of vigilance states on themMethods: We have previously shown how electrocorticographic signals (ECoG) can be decomposed using an unsupervised hidden Markov model (HMM) into a sequence of vigilance states that is in agreement with a concurrent scalp electroencephalogram (EEG). With IRB approval and informed consent, we acquired and analyzed ten good quality overnight interictal recordings (1000-2000 Hz sampling rate) from ECoG grid contacts in five patients with refractory epilepsy, and correlated high frequency (HF) power in the ripple (HF1: 80-250 Hz) and fast ripple (HF2: 250-500 Hz) bands with vigilance state scored by an HMM. We tested whether HF1 and HF2 varied significantly with vigilance state. Finally, we analyzed sample segments using an HFO detection algorithm from the literature to determine whether HF1 and HF2 power were correlated with the incidence of actual HFOs.Results: Each overnight ECoG recording was partitioned into five vigilance states in 30s epochs using an HMM based on ECoG power in the delta, theta, alpha, beta, and gamma bands. Then, the distributions of HF1 and HF2 signal power were compared for the five HMM states using a nonparametric one-way analysis of variance (Kruskal-Wallis test). Results for 10 overnight recording in five subjects showed that ripple and fast ripple band power varied significantly with vigilance state in all cases (p < 0.01). Inspection of sample segments from these recordings using an HFO detection algorithm suggested that vigilance states with greater HF band power were likely to be associated with larger numbers of HFO candidates. A comprehensive analysis of events in each recording is ongoing.Conclusions: HFOs show great promise as markers of epileptogenic tissue but analysis of their dynamics is complicated by changes in HFO activity with vigilance state and the lack of uniform guidelines for detecting and labeling them. Our results suggest that simple estimates of HF band power vary with vigilance state and may even reflect the relative numbers of HFO events in the underlying signal. If the correlation with true HFO activity holds, such measures could provide a more convenient means of studying and evaluating patterns of HFO activity for diagnostic applications.