In 9 mesial temporal epilepsy patients, fast-ripple maps stabilized after about 2–5 NREM intervals, but more data did not improve resection agreement

We used the published HFO detector markings in OpenNeuro ds003498 (20 patients, 450 intervals of 5 min of NREM sleep, bipolar channels). We did not use the raw signals. We measured how many intervals a channel map of HFO rate needs to become reproducible. We also measured if more intervals increase the agreement of the map with the resected zone (RZ). In the 9 temporal-lobe patients with mesial depth electrodes, the median split-half Spearman rho with 1 interval was 0.91 for ripples, 0.67 for fast ripples that co-occur with ripples (FRandR) and 0.47 for all fast ripples (FR). The Spearman-Brown estimate of the intervals necessary for rho 0.8 was 0.4 for ripples, 2.0 for FRandR and 4.6 for FR (range 2.1-13.2). From night to night, the median rho was 0.96 (ripple), 0.83 (FRandR) and 0.63 (FR). The AUC of the FR map for RZ vs. non-RZ channels was 0.62 with 1 interval and 0.62 with all intervals (median paired change +0.04, Wilcoxon p = 0.74, n = 8). For FRandR, the AUC was 0.63 vs. 0.65 (p = 0.25). Ripples did not find the RZ (AUC about 0.5). In one seizure-free patient, more data made a contralateral FR map more stable, but not more correct (AUC 0.19). In the 11 extratemporal patients with grids and strips, FR needed 3.7 intervals and the AUC increased a small amount (0.69 to 0.72, p = 0.037). In all 20 patients, FR and FRandR found the RZ better in seizure-free patients (FRandR: Mann-Whitney p = 0.003). In the depth group alone, this difference was not significant (p = 0.13). Thus, about 10-25 min of NREM sleep gives a reliable FR map. More recording time does not fix the localization errors. Limits: small groups, RZ as the reference, and detector markings only. The package contains the code, an input manifest with SHA-256, the result tables and 4 figures.

high-frequency oscillations · fast ripples · epilepsy · intracranial EEG · resection · reliability