Abstracts

Rationale: It has been proposed that finding increased rates of high-frequency oscillations (HFOs), used in conjunction with other methods, could be used to localize epileptic tissue for surgical resection.  However previous research has shown that the rate of HFOs is not stable over the duration of recordings. The rate of HFOs increases during periods of slow-wave sleep, and the rate may trend up or down within each sleep stage (von Ellenrieder et al., 2017). In new work we seek to further characterize the changing rate of HFOs during long-term intracranial recordings by defining the temporal dynamics of the rate at both slow and fast time scales. Methods: High-frequency oscillations (HFOs) were recorded with intracranial electrodes (iEEG) interictally from both the irritative and non-irritative cortical tissue in patients with intractable epilepsy (n=5) who were candidates for cortical resection. HFOs were automatically detected in the ripple band (80 – 250 Hz bandpass) using a validated algorithm (Charupanit & Lopour, 2017). The time course of HFOs was then coded as counts per half minute over multiple hours (maximum of 69 hours, minimum of 19 hours). Simulations of dynamic Poisson processes with sinusoidal fluctuations in HFO rates confirmed that temporal dynamics of HFO rates can be recovered by the Hilbert transform of the time-course of HFO counts per half minute. Hierarchical Bayesian modeling was used to characterize the relationship of patient-recorded HFO rates across electrodes and time, which included hierarchical parameters to summarize rates across different subjects and brain regions (i.e. irritative and non-irritative cortical tissue). Patient sleep stages were defined using the delta (1-4 Hz) power of intracranial electrodes, in order to be related to HFO rate increases. At shorter time scales, coefficient of variation (CV) statistics (i.e. the ratio of the standard deviation over the mean) of inter-HFO times were calculated in order to test whether HFO onsets can accurately be characterized as Poisson processes. Results: We show evidence that the rate of HFOs can be characterized by a changing Poisson process over long time scales, with fixed rates in different sleep stages. Evidence was found that ripple-band HFOs rates do not fluctuate over time within distinct sleep periods on the minute time scale. However some evidence was found that HFO rates are generated from Poisson-like processes that allow clumping on the scale of seconds (i.e. the likelihood of a HFO occurring immediate after another is increased) as indicated by large CV statistics in some electrodes. Conclusions: This work helps inform clinicians and researchers about the multi-scale temporal dynamics of HFOs for the diagnosis and treatment of epilepsy. Furthermore, these statistical modeling procedures allow clinicians and researchers to accurately measure HFO rates with modest amounts of data and almost no visual inspection. Future work seeks to classify pathological and physiological brain regions using Poisson rates of HFOs in different sleep states and trends in HFO rate recovered by hierarchical Bayesian models. ReferencesCharupanit K, Lopour BA. A simple statistical method for the automatic detection of ripples in human intracranial EEG. Brain Topography. 2017;30(6):724-738.von Ellenrieder N, Dubeau F, Gotman J, Frauscher B. Physiological and pathological high-frequency oscillations have distinct sleep-homeostatic properties. NeuroImage: Clinical. 2017;14:566-573. Funding: Supported by a grant from the American Epilepsy Society