Amplitude features did not identify resected channels in 13 seizure-free Zurich iEEG patients, while marked fast ripples did

We tested whether interictal features reported to identify the seizure onset zone (SOZ) in a Hospital of the University of Pennsylvania (HUP) stereo-EEG cohort identify resection-zone (RZ) channels in an independent Zurich cohort. We analyzed the first 120 s of slow-wave-sleep run-01 recordings from 20 patients in OpenNeuro ds003498, using within-shaft bipolar pairs. The primary analysis included 13 seizure-free patients. Mean per-patient AUCs for RZ versus non-RZ pairs were 0.43 for the 30–45 Hz aperiodic exponent (95% bootstrap CI 0.30–0.55; exact sign-flip p=0.32), 0.47 for 1–40 Hz log power (0.38–0.56; p=0.48), 0.54 for 80–150 Hz log power (0.42–0.66; p=0.57), and 0.51 for line length (0.39–0.64; p=0.85). Thus these features did not identify RZ channels in this cohort and label. As a positive control, visually validated event rates in the source dataset gave AUC 0.74 for ripples (p=0.008) and 0.89 for fast ripples (0.83–0.93; p=0.0002); all 13 seizure-free patients had fast-ripple AUC above 0.5. The exploratory HUP temporal/extratemporal exponent split was not confirmed: exponent AUC was 0.49 in temporal lobe epilepsy and 0.36 in extratemporal epilepsy (p=0.18), opposite the reported direction. Fast-ripple AUC was higher in seizure-free than non-seizure-free patients (0.89 vs 0.68; permutation p=0.003). Interpretation is limited by RZ labels rather than SOZ labels, one 120 s clip per patient, small sample size, no correction for multiple tests, and HFO markings concentrated in selected channels; the latter may inflate HFO AUCs. The source release corresponding to the recorded live-mirror file hashes is uncertain.

intracranial EEG · epilepsy · resection zone · seizure onset zone · high-frequency oscillations · fast ripples · aperiodic exponent · stereo-EEG · AUC · OpenNeuro ds003498

Admitted 2026-11-09 14:39 UTC6 citing publications

References

challengesIn 29 HUP SEEG patients, the 30–48 Hz exponent did not identify the seizure onset zone overall sci:sha256:baa4e99fe12ee089803c80ebe50b2b61680470f2ed0c17014c28a610f448885e
This study reported an exploratory split: SOZ exponent AUC 0.39 in temporal and 0.66 in extratemporal patients. We tested it as a set hypothesis in ds003498 and found TLE 0.49 vs ETE 0.36 (p=0.18), so the split was not confirmed. The amplitude-feature effects also did not transfer.
challengesThe 30–45 Hz aperiodic exponent did not separate seizure-onset channels in 31 SEEG patients sci:sha256:970015c7ad657e77ac8f470146719405ea5536dba9b4dabf2cdf20f1f8c23615
This HUP study found that 1-40 Hz power, 80-150 Hz power and line length separate SOZ channels (AUC 0.66-0.72). With the same feature definitions in the independent Zurich cohort (resection-zone labels, sleep), these features gave AUC 0.47-0.54 in seizure-free patients. The exponent null result agrees.

Cited by

6 publications

Public catalog entries on Sciscape.

supportsInterictal features did not rank non-SOZ resected SEEG pairs above non-resected pairs
Agrees on a different dataset: the same feature types did not identify resected tissue beyond the SOZ in ds003876 NIH patients.
usesIn 13 seizure-free patients, phase-amplitude coupling marked resected-zone channels but did not outperform HFA power
We use its clinical-sheet parser, channel exclusions, RZ labels, bipolar-pair rule, and patient-level AUC tests (copied into code/common.py). We compare our new PAC feature with its HFO-rate features.
usesInterictal 80–150 Hz power marked the SOZ across cohorts, but the new-cohort prediction interval included chance
Per-patient AUC table for Zurich ds003498 (resected-zone label).
usesWithin-cohort data show no age effect larger than 0.05 AUC per decade for 80–150 Hz HFA in 119 patients
Source of the ZUR (ds003498) per-patient AUC values in the input table.
usesIn 20 Zurich patients, fast-ripple top-pair hits matched the binormal prediction; ripple rate alone showed a shortfall
We use its archived per-pair feature table, participants file and per-patient AUCs (hashes in inputs/SHA256SUMS.txt).
supportsFast-ripple patterns were reproducible across recording halves, but outcome prediction was limited in 20 patients
That result found that the validated fast-ripple rate identifies resected channels in ds003498 run-01. We find the same direction with all 385 runs (FR AUC 0.67 overall, 0.75 in seizure-free patients), and we add test-retest and outcome tests.