Abstracts

Rationale: High Frequency Oscillations (HFOs), including Ripples (R, 80-250Hz) and Fast Ripples (FR, 250-500Hz), are recorded from intracranial macroelectrodes in patients with intractable epilepsy. HFOs and Spikes (Sp) are visually identified by experienced neurologists, which is a highly time consuming job. It takes around 10 hours to mark HFOs in a 10-channel 10min recording. The lack of a complete definition of HFOs results in poor agreement among neurologists. We implemented new methods to establish the duration when a steady state in the rates of HFOs is achieved and to assess and iteratively improve concordance between neurologists. Methods: Intracerebral EEGs from 40 patients were processed (filtered at 500Hz and sampled at 2000Hz). A)HFOs and Sp were identified in 10 patients during 10min of slow wave sleep (8-23 channels). Different durations and starting points were compared against the 10min “gold standard”. The mean and variance of the relative rate per minute of R, FR and Sp were computed. The information gained when increasing the interval duration was analyzed and we determined how the different durations affected the ranking of channels with respect to HFO rate. B)To determine concordance, 2 neurologists independently identified HFOs and Sp during 1min in 10 patients. Cohen’s Kappa coefficient (κ) was computed for this 1st minute. Channels with κ<0.5 (reflecting poor agreement) were reviewed, discussed and independently re-marked. Based on the new criteria, HFOs and Sp were independently identified in the other 30 patients. Results: A)In 8/10 patients less than 5min provided the same information as 10min (fig.1). For R, in 8/10 patients the ranking of channels with respect to HFO rates remains stable for durations longer than 5min. For FR and Sp, this was the case in 9/10 patients (fig.2). For 1 patient, 10min do not seem to be enough. In this patient the last minute of selection showed a decrease in delta band, suggesting a change in sleep stage. B)In the first 10 patients the κ was 0.6 for R, 0.56 for FR and 0.49 for Sp. For the second group (30 patients), κ was high: 0.7 for R, 0.7 for FR and 0.67 for Sp, showing a clear improvement especially for Sp. Conclusions: The methods presented provide neurologists with new information to improve the visual identification of HFOs in patients with epilepsy and reduce the time needed to mark these events. We propose that 5min provides in most cases the same information as a longer interval, and present a method to identify when this is not the case. The following procedure is suggested for each patient: 1min is marked by both neurologists and κ is calculated. Once there is good agreement (κ>0.5, channels with poor agreement are re-reviewed) in the first minute, the remaining 4min are marked by one of them. These methods are useful to control for consistency between neurologists and to evaluate if the selected interval provides consistent information, or if a longer one is needed. Moreover, for the selection of intervals to analyze it is important to consider a stable stage of sleep, as reflected by the spectra in delta band. Supported by NSERC PGSD, CIHR 10189.