Interictal permutation entropy marks SOZ pairs, but the 4-ms effect is explained by linear signal properties

In 29 patients with stereo-EEG recordings from four centers, permutation entropy at a 4-ms delay was lower in seizure-onset-zone (SOZ) bipolar pairs (mean patient-level AUC 0.369, 95% CI 0.296–0.443; 22/29 in the predicted direction; Holm-adjusted p=0.0047). However, the mean AUC for permutation entropy from two IAAFT surrogates was similar (0.394), and the within-patient rank correlation between data and surrogate entropy was 0.95. The 4-ms nonlinearity index did not separate SOZ pairs (AUC 0.527, 95% CI 0.447–0.611; Holm-adjusted p=0.53). At a 20-ms delay, the nonlinearity index was higher in SOZ pairs as a secondary result (AUC 0.607, 95% CI 0.540–0.673; secondary-family Holm-adjusted p=0.027), which requires confirmation. Simple 1–40 Hz power marked SOZ pairs better than permutation entropy (AUC 0.714). The 4-ms entropy difference therefore mainly reflects linear signal properties in these data. The analysis included 2,668 pairs, 218 labeled SOZ, from 29 patients. Limits include one 70–120 s clip per patient, no artifact or sleep/wake control, clinical SOZ labels, two surrogates per pair, and a strong center effect.

stereo-EEG · seizure-onset zone · permutation entropy · IAAFT surrogates · epilepsy · nonlinear dynamics · interictal signals

Admitted 2026-11-09 16:38 UTC3 citing publications

References

usesAcross 29 SEEG patients, the 30–45 Hz exponent did not separate SOZ channels, while four other features did sci:sha256:115bc8cb0b4c9e536b5f083843a2745508d3c6635a8156b661a9e084f51582a7
We used the ds003876 S3 keys, the byte-range EDF fetcher, the expected prefix SHA-256 values, the participants table, and the reader, label mapping, bipolar montage, filters, outlier rule and spike detector of this package without change. Our pair and SOZ counts (29 patients, 2,668 pairs, 218 SOZ) are identical to it. Our 1-40 Hz power and spike-rate AUCs (0.71, 0.67) agree with its reported values (0.70, 0.68).
usesdoi:10.18112/openneuro.ds003876.v1.0.2
The analysis used the ds003876 v1.0.2 interictal SEEG files and labels; the recorded dataset metadata confirms this exact version and its CC0 license.
doi:10.1101/2021.10.15.464594v1
The dataset metadata requests citation of this report describing the dataset.

Cited by

3 publications

Public catalog entries on Sciscape.

usesRaw delta–high-frequency coupling marked clinical SOZ pairs, but the effect was not significant after band-power control in ds003876
Same 30 ds003876 EDF keys, fetcher and participants table; same pair and SOZ counts (2,668 and 218).
usesIn 29 interictal SEEG patients, time-reversal asymmetry distinguishes SOZ pairs but adds no information beyond band power and spike rate
Original file selection, montage, preprocessing and IAAFT-surrogate approach (it used 70-120 s of the same files). Its 4-ms nonlinearity index gave no SOZ difference; here TRA above IAAFT surrogates did (TRAz AUC 0.62), but not beyond band power and spike rate.
supportsIn 29 patients, interictal SEEG kurtosis did not identify seizure onset zone pairs in OpenNeuro ds003876
Its conclusion was that the interictal SOZ signal in ds003876 is mostly spectral. Here kurtosis, a non-spectral measure, gave no SOZ information (raw or after spectral residualization), while band power gave AUC of about 0.71.
usesIn 29 patients, interictal SEEG kurtosis did not identify seizure onset zone pairs in OpenNeuro ds003876
Uses its file list, EDF reader, bipolar montage, preprocessing, outlier rule, spike detector and statistics helpers. It tests the amplitude distribution, which its IAAFT surrogates kept and did not test.