Abstracts

Rationale: Detecting transient high frequency oscillations (HFOs) in the human electroencephalogram (EEG) in the ripple (80-200 Hz) and fast ripple (250-500 Hz) range has attracted increasing interest because HFOs are reported to localize to seizure onset zones, and the resection of brain tissue with high rates of HFO activity appears to be correlated with better seizure outcomes. These oscillations are brief, lasting on the order of tens of msecs. There are also normal brain oscillations in the ripple and fast ripple range, and it can be difficult to tell them apart. An automated computer algorithm that detects pathological HFOs quantitatively and reliably would be useful for scoring long-term EEG from patients undergoing pre-surgical localization.Methods: We have developed a novel high resolution pseudo-wavelet time-frequency analyzer called the damped-oscillator oscillator detector (DOOD). This method uses a set of mathematical oscillators, i.e., damped harmonic oscillators, to detect oscillations in EEG. Here we show that HFOs can be detected with DOOD. The DOOD spectral density S(f,t) is first Z-normalized by subtracting out the mean and dividing by the standard deviation. Events with S(f,t) greater than some threshold S0 are then automatically flagged as candidate HFOs. Test data consist of 120 s of 8-channel intracranial microwire EEG collected from a patient undergoing epilepsy surgery evaluation, 100 s of 16-channel intra-hippocampal EEG collected from a normal live rat at baseline, and 100 s collected from the same rat 6 weeks after kainate-induced status epilepticus. The DOOD computer algorithm was compared with visual identification of HFOs, where the EEG was inspected visually second-by-second at one-second intervals. Institutional review board and animal care restrictions were strictly adhered to.Results: A total of 1076 HFOs were visually identified in the human microwire data spread across the 8 channels, and 1191 HFOs were visually identified in the rat data spread across 16 channels. Sensitivities and positive predictive values of the DOOD HFO detection algorithm are shown in Figs. 1 (human data) and 2 (rat data). An added feature of the DOOD approach is its ability to reveal wide-band structure. HFOs appear to be associated with intricate low frequency time-frequency structure, not apparent by more traditional short time Fast Fourier Transform techniques. Conclusions: DOOD may be a useful way to quantify HFOs for use in long-term EEG analysis. In addition, DOOD time-frequency analysis reveals an intricate low frequency structure that accompanies HFOs that may be useful for wide-band characterization of HFOs. In future, such wide-band characterization may help differentiate pathological HFOs from normal brain oscillations. Funding: NIH R01-25020, NIH R01-NS63039-01.