Variability in the location of high frequency oscillations during prolonged intracranial EEG recordings

Abstract

The rate of interictal high frequency oscillations (HFOs) is a promising biomarker of the seizure onset zone, though little is known about its consistency over hours to days. Here we test whether the highest HFO-rate channels are consistent across different 10-min segments of EEG during sleep. An automated HFO detector and blind source separation are applied to nearly 3000 total hours of data from 121 subjects, including 12 control subjects without epilepsy. Although interictal HFOs are significantly correlated with the seizure onset zone, the precise localization is consistent in only 22% of patients. The remaining patients either have one intermittent source (16%), different sources varying over time (45%), or insufficient HFOs (17%). Multiple HFO networks are found in patients with both one and multiple seizure foci. These results indicate that robust HFO interpretation requires prolonged analysis in context with other clinical data, rather than isolated review of short data segments.

Fig. 1

Example rates of HFOs. The HFO rate per epoch of NREM sleep is shown for four example patients. All NREM data are concatenated together, with white vertical lines indicating breaks between NREM periods. a A small group of channels have high HFO rates, though the highest channel varies over time. b One channel is high but alternates with arbitrary groups with much lower rate. c Many channels are high in an organized cluster and all appear well correlated. d Much more complex patterns are also observed, in which different channel groups dominate on different days. Channels in the resected volume (RV) and clinically determined seizure onset zone (SOZ) are ordered together, as are channels with similar temporal dynamics in HFO rates. Note that patients in a and b did not have a resection

Fig. 2

Distribution of HFO rate asymmetries. Results are shown for both the Mayo cohort (a–c) and the UM NREM cohort (d–f), and for three different HFO detectors: qHFO (a, d), fast ripple enhanced qHFO (qHFO-FR) (b, e), and a Hilbert transform based detector (c, f). Asymmetry is the difference between average HFO rate inside versus outside the SOZ, divided by the sum. Asymmetries for HFO rates are displayed in histogram form for all available 10-min epochs in the cohort (blue, green) and averaged over all 10-min epochs in each individual patient (gray, orange). The asymmetry was only computed for cases with at least one channel having 0.5 HFOs/min. HFOs were determined to be related to SOZ if the median rate was significantly positive (right-tailed Wilcoxon Sign-Rank test, p < 0.05), seen as a skew to the right in the plots. HFOs from all three detectors in the UM cohort, and from the qHFO detector in the Mayo cohort, were associated with SOZ. There was no significant difference between using just 10 min of data or all data together in any cohort or for any detector (Wilcoxon Rank-Sum test, 0.1 < p < 0.9). Note, several patients had an asymmetry of −1 when there were no HFOs in the SOZ and at least one channel having >0.5 HFOs per min outside the SOZ. This often occurs in data with low signal quality, and thus occurs much more frequently in patients with worse signal to noise ratios

Fig. 3

Blind source separation of HFO rates. Each of the HFO rates presented in Fig. 1 are decomposed into groups of channels. The associations of each channel with each group (a–d, left), denoted by W, and each time epoch (a–d, right), denoted by H, are shown for each example patient. Denoting the rates in Fig. 1 by the matrix R, the blind source separation satisfies the matrix product R = W × H. White stars indicate the channels most involved with each source, which are color coded in Fig. 4

Fig. 4

Comparison of HFO channel groups and brain regions. Diagrams are shown for the same example patients as in Figs 1 and 3. Electrodes associated with specific channel groups (white stars in Fig. 3) are color coded by group number, and seizure onset electrodes are denoted by colored stars adjacent to the electrode. In c, a lesion (not shown) was found under channels 27, 28, 35, and 36. Note, the patients in a and b did not have resections, but both c and d did and have had class I outcomes for >3 years. Data were not available for grayed out electrodes in a

Fig. 5

Time of seizures versus HFO rates. The effective HFO rates per channel group for patient UM-07 during interictal NREM periods (using the same scheme shown in Fig. 3d) are plotted at the time of day of each epoch to indicate relative activity of each group for a given time. Tricolor vertical bars indicate the three channel groups. The starting time of each seizure (including subclinical seizures) is also denoted as short vertical lines at the bottom of each row, color coded for seizure type. Colorbar: relative strength of each HFO channel group. Note that different HFO groups become active prior to changes in seizure type

Fig. 6

Occurrence rates for each category of spatiotemporal variability. Occurrence rates are presented separately for each cohort, separating the results from the NREM interictal data and 1–3 AM data for the UM cohort. a HFOs detected using standard 80–500 Hz bandpass. b HFOs detected using 200–500 Hz bandpass for Fast Ripples (FR). c Hilbert-based detector. No statistical difference was found in the percentage of subjects in a given category between any two cohorts (all p > 0.05, χ² test). However, in all cases, the category (a) did not account for all patients with measurable HFOs (non-category (d)) (χ² test, *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001). Error bars: 1 s.d

Fig. 7

Category of variability as a function of total recorded time. Categorizations are made using all NREM interictal epochs from the recording start up to each given hour. Additionally, the location of the NREM interictal epochs are shown. Some patients cease to change categories as new data are acquired after a few days, whereas other patients still change category even after many days