Fast-ripple patterns were reproducible across recording halves, but outcome prediction was limited in 20 patients

We assessed test-retest stability and outcome prediction using published interictal high-frequency oscillation (HFO) events from OpenNeuro ds003498: 385 slow-wave-sleep intervals from 20 patients (13 seizure free, 7 not seizure free). No HFOs were redetected and raw signals were not used. Fast-ripple (FR) channel rates had a mean first-half versus second-half Spearman rho of 0.73 in 18 patients with at least two runs (exact sign-flip p < 1e-4); correlations between single runs averaged 0.55. FR rate discriminated resection-zone from other pairs (mean AUC 0.67, p = 0.016), with AUC 0.75 in seizure-free patients (p = 0.007, n = 13). A simplified rule based on pairs at or above 50% of the maximum rate was correct for 13/20 patients using pooled runs; across single runs, mean patient accuracy was 64%, and the fraction predicting seizure freedom had AUC 0.80 (permutation p = 0.029). This outcome signal came from extratemporal patients. In nine temporal-lobe epilepsy depth-electrode patients, the outcome AUC was 0.50 and rule accuracy was 4/9. Ripple patterns were more stable (half-split rho 0.83; single-run rho 0.82), but did not identify the resection zone (AUC 0.51, p = 0.88) and the outcome rule accuracy was 35%. This exploratory analysis is limited by the small sample, uncorrected multiple tests, unknown run-to-night mapping, a simplified outcome rule, and omission of pairs with zero events across all runs. The included count table, code, per-patient results, input file hashes, and download recipe document the analysis.

high-frequency oscillations · fast ripples · epilepsy surgery · test-retest reliability · resection zone · seizure outcome · intracranial EEG · OpenNeuro ds003498