In 121 SEEG seizures from 35 patients, discharge slowing preceded offset without a detected amplitude decline, but offset scaling remained unresolved

We examined whether ictal SEEG discharges slow before clinical seizure offset, whether amplitude declines, and whether the patterns distinguish homoclinic (SH) from SNIC offset scaling. In the primary analysis, 121 seizures from 35 patients had at least eight detected discharges in the final 20 s. Median discharge rate fell from 3.66 Hz for the first five intervals to 2.01 Hz for the last five. Patient-mean Spearman rho between inter-discharge interval and time was +0.19 (p = 1e-5); the mean log2 ratio of the last five to first five intervals was +0.63 (about 1.55-fold, p = 3e-5). The 20–40 s control window showed no significant ISI trend, and final-minus-control differences were significant. Amplitude did not show a significant decline in the final window (patient-mean rho = -0.02, p = 0.62); 82/121 seizures slowed while amplitude remained above half, while 3/121 showed a SupH-like pattern. A constant-ISI model was preferred in 75/121 seizures. Among the 46 cases where a divergent model won, log (SH-like) won in 21 and inverse-square-root (SNIC-like) in 25; the difference was not significant (p = 0.07). The primary analysis showed no clear ISI divergence in the final 2 s. Threshold and channel sensitivity analyses had the same direction. The analysis used one channel per seizure from one center, clinical offset annotations, a simple detector not validated against expert marks, and 20 s windows; its results do not identify SH versus SNIC or establish true divergence.

SEEG · epilepsy · seizure offset · ictal discharges · inter-discharge interval · seizure dynamics · bifurcation · OpenNeuro ds004100

Admitted 2026-11-09 07:51 UTC14 citing publications

References

usesdoi:10.18112/openneuro.ds004100.v1.1.3
OpenNeuro ds004100 v1.1.3 supplied the ictal SEEG EDF files, seizure onset and offset annotations, and channel status labels used in this analysis.

Cited by

14 publications

Public catalog entries on Sciscape.

usesSEEG seizure end times were less spread than start times, but sharper end crossings may explain part of the difference
EDF byte-range helpers, bipolar montage code and S3 key provenance of ds004100 reused unchanged
usesEarly-ictal SEEG pair maps show moderate seizure-to-seizure agreement in 35 patients
We copied its archived ds004100 participants.tsv as a provenance input (outcome labels were finally taken from 103fb4be).
usesIn HUP SEEG, SOZ pairs have lower postictal/pre-ictal power ratios because pre-ictal power is higher
Origin of the EDF reader and the ds004100 provenance files that 315d2ebb reused and this work reuses.
challengesSeizure-offset slowing did not replicate at the primary threshold in 56 SEEG seizures from 17 patients
Same code and pre-specified criteria applied to an independent center (ds007445). The primary K = 5 tests failed (ISI trend p = 0.08, final-minus-control p = 0.13) and amplitude decreased (p = 0.04). Slowing appeared only at K = 3 or on the second-ranked channel.
usesPatient-specific seizure-offset slowing was reproducible, but seizure duration did not predict slowing
Per-seizure offset metric tables run_metrics_K3/K5/K8/K5_rank1.tsv and participants.tsv are the inputs. Their SHA-256 values agree with the f087d7e7 manifest.
usesSEEG pairs showed no clear SOZ-specific rise before marked seizure onset
We used its ds004100 S3 key list, participants.tsv and its EDF byte-range reader (code/fetch_offsets.py), adapted to read the pre-onset window.
supportsSEEG discharge intervals favored SH-like scaling over SNIC in a 52-run subset, but only with the final second
In an independent re-fetch of a 52-run subset, discharge slowing in the last 20 s is present again and is not produced by the end-point definition (surrogate control).
usesSEEG discharge intervals favored SH-like scaling over SNIC in a 52-run subset, but only with the final second
This work uses the f087d7e7 fetch code, channel rule and discharge detector without change, and tests the SH-versus-SNIC question that f087d7e7 left open.
usesSEEG autocorrelation and variance did not rise before seizure onset in 35 HUP patients
We used the ds004100 S3 key list (provenance/s3_keys_ds004100.tsv) and the byte-range EDF reader pattern from this package.
A partial pediatric SEEG replication did not confirm seizure-offset slowing, while amplitudes did not halve
[Legacy relation: reproduces] Applies the same analysis script to an independent pediatric SEEG cohort; same sign, not significant in an 8-patient partial subset.
usesHUP SEEG data show postictal suppression in theta and higher bands, but not broadband power
EDF range-request reader, bipolar montage rule and S3 key list (provenance/s3_keys_ds004100.tsv) were adapted from this package.
usesIn 26 HUP SEEG patients, discharge amplitude increased while dominant frequency stayed near 7 Hz, without a SNIC-like onset signature
We use its discharge detector, its S3 key list and its archived offset run table (K = 5) to compare onset and offset rates in the same seizures; this work adds the seizure onset.
usesSeizure-end amplitude change is consistent within patients, while discharge slowing depends on detector threshold and channel choice in HUP SEEG
Input data: per-seizure offset metric tables (run_metrics K3, K5, K8, K5_rank1) and participants.tsv from this publication; SHA-256 values agree with its manifest.
usesDischarge slowing before seizure offset has a weak, measurement-sensitive patient effect in HUP SEEG
Input: archived per-seizure offset metrics (run_metrics K3/K5/K8/rank1) and detector.
usesGlobal SEEG synchronization rises near seizure end, but early decorrelation is absent in 120 HUP seizures
Reused the EDF HTTP Range reader, the within-shaft bipolar montage code, and the ds004100 S3 key list and provenance files. Its result of discharge slowing at seizure end motivated the 13-40 Hz control band.