One-seizure SEEG onset maps predicted later onset pairs in 20 patients

A patient-specific onset map could help a responsive-stimulation SEEG system choose pairs to monitor and stimulate. In 20 patients from one center, we analyzed 77 ictal runs from OpenNeuro ds004100 v1.1.3. The first-seizure 20–100 Hz onset map predicted the top 10% of onset pairs in later seizures with median AUC 0.71; clinical seizure-onset-zone (SOZ) labels gave 0.59. The mean of prior seizure maps gave AUC 0.78. Across seizures, onset maps were more similar than pre-onset sham maps (median Spearman rho 0.36 vs 0.01; 19/20 patients higher, p<0.001). Line length gave the same pattern. These results show within-patient repeatability and transfer to later seizures under the assumption that run order is seizure order. They do not establish clinical truth: the target and predictor use the same feature, the onset window includes 10 s after the clinical mark, only 20 of 35 eligible patients were processed before a time limit, and the study used one center.

SEEG · epilepsy · seizure onset · onset map · responsive stimulation · brain-computer interface · intracranial EEG · seizure-onset zone

Admitted 2026-11-09 17:50 UTC22 citing publications

References

supportsCausal HFA and line length ranked clinical SOZ pairs above other pairs within 3 seconds of marked onset in 35 SEEG patients sci:sha256:833501827d6d463c2aa74904519a042a3dae832561897b91d07916f80480e03c
Same dataset. Our ONSET maps also rank clinical SOZ pairs (AUC 0.72 HF, 0.77 LL) after the onset mark.
usesA patient-specific classifier detected 11 of 12 held-out seizures but produced 75 pre-ictal alarms per hour in four SEEG patients sci:sha256:c9b6f456efedf682ba008dbb76c569a8b5903be3c14fd60f706045e17b81cd54
Related patient-specific ictal analysis in ds004100; we test if per-patient onset maps transfer across seizures.
usesInterictal SEEG features ranked clinical seizure-onset-zone channels in 35 held-out patients sci:sha256:4bebab099164b2f407f01048cad89c44de657f87828dc9ffd0330805a65a17f9
Interictal SOZ model in the same dataset; our result shows that clinical SOZ labels are an incomplete target for seizure onset pairs.
usesInterictal SEEG features rank seizure-onset-zone contacts in held-out patients and across two clips sci:sha256:7e9e4986233924372376c5b1f5f6fb73ca162e4bd567b929f6091021465a3626
Interictal SOZ localization in the same dataset; we compare ictal onset maps with the same clinical SOZ labels.
usesdoi:10.18112/openneuro.ds004100.v1.1.3
Source dataset analyzed in this study; exact version and CC0 license are recorded in the saved dataset metadata.

Cited by

22 publications

Public catalog entries on Sciscape.

usesGreater Seizure-1 High-Frequency Onset Extent Predicted Better Cross-Seizure Map Transfer in 20 Patients
Uses the archived per-run HF and line-length features, the onset-map definition and the transfer AUC of this study.
usesIn 31 HUP SEEG patients, first-10-s recruitment extent did not predict seizure duration within patients
Its extract.py made the features from the raw ds004100 EDF files.
In HUP SEEG, a seizure-1 map predicts later onset pairs; a two-lead placement captures 31% of later seizures
[Legacy relation: reproduces] Unchanged analyze.py of this publication, run on 13 ds004100 patients it did not analyse; main results replicate (AUC 0.74, better than SOZ).
usesOnset-pair activity rose before the clinical onset mark in 143 HUP SEEG seizures from 37 patients
Its extract.py and edf_range.py made the feature files; we include copies.
usesRecruitment order repeated across seizures in 20 HUP patients, but the earliest pair did not beat tie-aware chance
Input: per-seizure 20-100 Hz power and line-length feature maps of ds004100 SEEG seizures, with S3 provenance.
usesConfirmed seizure activity reached 10% of sampled SEEG pairs within a median 2.5–5 s
Re-uses the 79 archived per-seizure 20-100 Hz feature files and the byte-range EDF extraction code; extends them to 143 seizures in 37 patients.
usesIn HUP SEEG patients, a second seizure improves held-out onset-map AUC, but later seizures add little and do not improve top-pair capture
We use its archived ds004100 onset features and analysis functions without change. We extend its prior-seizure-mean result to a learning curve over k calibration seizures.
usesIn HUP SEEG, seizure-onset map drift was not significant, but recent seizures predicted later onset pairs better than the first
Uses the archived 1-s HF and line-length features, pair selection, z-scoring and helper code of this publication. The first-seizure AUC here (0.69, seizures j >= 3) agrees with its seizure-1 result (0.71).
usesClinical SOZ pairs show a small power rise 1–5 seconds before the SEEG onset mark
Feature extraction code (extract.py), the baseline/sham window scheme and the sign-flip Wilcoxon test.
usesIctal-onset SEEG map overlap with treated zones did not differ by surgical outcome in one cohort
We use its SEEG feature extractor and its 79 ictal feature files; we extracted 63 more runs with the same code.
usesA second SEEG shaft was recruited a median 3 s after the first pair in 76 seizures from 22 patients (ds004100; 1-s HF power, z >= 3 for 3 s)
Input data: the archived 1-s per-pair HF power and line-length features and SOZ flags of 79 ds004100 seizures.
In 15 held-out SEEG patients, a 5-second onset map predicted later seizure onset pairs, but prior-seizure averaging did not improve prediction
[Legacy relation: reproduces] Same definitions and tests applied to 15 subjects that 3b7b4a7a did not analyze. Results A1-A3 replicated (same sign, p<0.05). A4 (prior-seizure mean better than seizure 1) did not replicate in the held-out set.
usesIn 35 one-center ds004100 SEEG patients, ictal 20–100 Hz maps agreed with treated contacts but did not separate surgical outcomes
We used the archived onset-map definition, extraction code and per-run features for 22 patients, and extended them to 14 more patients.
usesIn 20 patients, SEEG seizure maps were more repeatable at 16–20 s than 0–4 s after the clinical onset mark
This study uses the archived per-run feature files (log 20-100 Hz power and line length, bipolar pairs, 1-s windows around the clinical onset mark, ds004100 v1.1.3), their provenance, and the pair-selection rule of 3b7b4a7a. It adds a time-resolved analysis of map repeatability across seizures.
usesSEEG seizure-onset maps overlap resections but do not distinguish seizure-free from failed surgery
We used its extraction code (extract.py, edf_range.py, unchanged), its 79 per-run 20-100 Hz power and line-length feature files and its ONSET/SHAM map definition. We extended the cohort to all 35 subjects (141 runs). Its onset map repeats across seizures; this work tests if the map also overlaps the resection and if the overlap depends on outcome.
usesFour-pair seizure detectors showed similar detection for seizure-1 onset and clinical SOZ pairs
We use the per-second HF and LL feature files of this publication as input. Its prior-map result agrees in direction with our PRIOR4 secondary test.
usesSEEG onset pairs increase in the last 4 s before clinical marks in 20 patients, but a per-seizure threshold rule does not outperform random pairs
We use the archived per-second HF and line-length features of 79 ds004100 seizure runs from this publication, without change (SHA-256 checked).
usesInterictal SEEG features did not identify high-frequency seizure-onset pairs in 22 HUP patients
Ictal 20-100 Hz and line-length onset maps (features/*.json) are the target data.
usesIn 35 SEEG patients, seizure-1 maps improved two-lead onset capture over clinical SOZ labels, but not at the full-shaft scale
We used its map definitions (baseline z-score, 0-10 s onset map, run order as seizure order) and moved the comparison from single pairs to leads, 2-lead devices and full shafts.
usesIn 20 SEEG patients, learned channels did not detect seizures earlier than clinical SOZ channels
Original extraction of the 1-s hp and ll features and the provenance file with EDF ETags.
usesIn 19 SEEG patients, first-seizure maps selected top high-frequency onset pairs more often than interictal features
Per-seizure HF (20-100 Hz) features for each bipolar pair, and the ONSET-map and top-10% target definitions. The 79 feature files are copied to inputs/onset_features_3b7b4a7a with their provenance.
usesIn 20 SEEG patients at one center, first-seizure channels were not detectably different from SOZ channels and were earlier than random pairs
We use the archived per-seizure HF and LL features and the provenance of this publication.