Slow-wave–HFA coupling and HFA power distinguish clinical SOZ depth pairs in 43 patients

In a documented analysis of 43 patients and 728 bipolar SEEG depth pairs (160 labeled seizure-onset-zone, SOZ), interictal non-REM high-frequency activity (HFA; 80–300 Hz) power and slow-wave phase modulation of HFA (0.5–2 Hz; Tort modulation index) each ranked SOZ pairs above other pairs. Mean per-patient AUC was 0.684 for log HFA power (95% CI 0.596–0.766; Holm-adjusted p=2.2e-4) and 0.729 for modulation index (0.654–0.801; p=2.0e-5). Modulation index retained discrimination after rank residualization on HFA (AUC 0.616, 0.542–0.687; p=0.0034). Effects were uncertain in the 15 Detroit cohort patients (HFA p=0.67; modulation p=0.076), while the 28 geographical external cohort patients had higher mean AUCs (0.760 and 0.762; both p≤1e-5, the permutation test floor). SOZ labels derive from clinical ictal recordings and spiking; only 90 seconds per patient were analyzed, and pairs share contacts. These findings describe association with clinical SOZ labels and do not establish independent localization or clinical utility.

intracranial EEG · SEEG · epilepsy · seizure-onset zone · high-frequency activity · slow-wave coupling · modulation index · non-REM sleep · bipolar depth contacts

Admitted 2026-11-09 07:42 UTC70 citing publications

References

usesdoi:10.18112/openneuro.ds007118.v1.0.0
The analysis used NREM iEEG data, channel labels, and participant information from OpenNeuro ds007118, snapshot 1.0.0.
usesdoi:10.18112/openneuro.ds007119.v1.0.0
The analysis used NREM iEEG data, channel labels, and participant information from OpenNeuro ds007119, snapshot 1.0.0.
usesdoi:10.18112/openneuro.ds007120.v1.0.0
The analysis used NREM iEEG data, channel labels, and participant information from OpenNeuro ds007120, snapshot 1.0.0.

Cited by

70 publications

Public catalog entries on Sciscape.

A second 90-s segment reproduced SOZ separation by HFA and slow-wave coupling in 43 SEEG patients
[Legacy relation: reproduces] In a non-overlapping 90-s segment (EDF 180-270 s) of the same 43 runs, the SOZ-versus-other-pair AUC was 0.697 for HFA and 0.749 for MI (Holm p <= 1e-4), against 0.684 and 0.729 in segment A. The MI residual after HFA was 0.633 (p = 3.5e-4). The same patients and recordings were used, so this is a within-recording replication.
usesA second 90-s segment reproduced SOZ separation by HFA and slow-wave coupling in 43 SEEG patients
We used the a3d06b3c feature code (only the window 180-270 s and the 43-run filter changed), its channel labels, file list and metadata, and its archived segment-A (30-120 s) pair values and byte ranges. The SHA-256 of the copied files is the same as in its manifest.
usesSEEG delta power at clinically normal bipolar pairs did not decrease with age in 33 patients aged 0–45 years
The inputs are the archived per-pair features (results/pairs.tsv) and participants table of this package.
supportsNREM slow-oscillation to sigma coupling is higher in seizure-onset-zone SEEG pairs
With a 180 s window and 0.5-1.25 Hz phase, SO-to-HFA Tort MI again separated SOZ pairs (mean AUC 0.744, 37/43 patients).
usesNREM slow-oscillation to sigma coupling is higher in seizure-onset-zone SEEG pairs
We reused the byte-range EDF reader, the subject set, the pairing rules and the channels.tsv provenance of this study.
usesMRI lesion status and age showed no supported effect on NREM SEEG HFA or MI AUC in 43 patients
Per-patient SOZ AUCs of HFA, MI and delta power in window A (30-120 s), and the OpenNeuro participants table (age, MRI lesion, other covariates).
usesIn 43 SEEG patients, high-MI pair resection predicted seizure freedom only as well as coverage-matched random resection
Per-pair 30-120 s features (log_hfa, mi, log_delta), SOZ and resection pair labels, the 43-patient bipolar depth-pair cohort and the archived ds007118 participants.tsv. These are the inputs of the primary tests H1 and H2.
In 18 UCLA SEEG patients, slow-wave coupling and HFA power marked clinical SOZ pairs, but MI did not add significantly to HFA power
[Legacy relation: reproduces] We used the MI, HFA power, delta power and residual definitions of its code in an independent centre (UCLA, ds005398). Its MI and HFA SOZ results replicated (AUC 0.68 and 0.73). Its MI-beyond-HFA result did not replicate in the primary clip (0.58, p = 0.13), but it agreed in the secondary clip (0.62).
usesA spindle detector decrease at seizure-onset-zone pairs is not robust to interictal spikes
This work uses the cohort rules, the run choice, the bipolar pair rules, the EDF byte-range reader (adapted in code/fetch_spindles.py), the archived file list and the channels.tsv copies of a3d06b3c. Its 80-300 Hz power result (higher at SOZ pairs) is reproduced here on a longer window (mean AUC 0.69) and is used as the control in S3.
usesIn 10 of 47 ds007118/9/20 NREM SEEG subjects, the run-01 EDFs gave no normal-labelled bipolar target pair away from source contacts, so the pre-registered IED-to-spindle timing test was not done
Uses its candidate subject list, channel labels, file list and byte-range EDF reader code.
usesAcross 3 SEEG cohorts, HFA increased with SOZ-contact count, but one-contact pair rankings differed by cohort
Computed the NREM ds007118-20 per-pair features and contact status columns.
supportsNREM SEEG power and slow-wave coupling separate seizure-onset-zone pairs within shafts in 41 patients
The SOZ separation by 80-300 Hz power and slow-wave coupling remains when only pairs on the same shaft are compared (mean AUC 0.72 each). We also show that the archived pair set has one shaft in 26 of 43 patients, so the original pooled AUC is mostly within-shaft.
usesNREM SEEG power and slow-wave coupling separate seizure-onset-zone pairs within shafts in 41 patients
We used the archived per-pair NREM features (results/pairs.tsv: log_hfa, mi, log_delta, soz) of 43 patients as the only input, and re-analysed them within and between electrode shafts.
usesTop interictal HFA misses in SEEG were not spatially near the clinical seizure onset zone
Source of the ds007118-20 NREM per-pair table.
usesIn 42 patients, age was not significantly associated with NREM sigma peak frequency in normal-contact SEEG pairs after cohort adjustment
Origin of the HTTP byte-range EDF reader for ds007118/9/20 that 5c22e5d3 reused and we reused through 5c22e5d3.
usesIn 21 SEEG patients with partial resection, NREM 80–300 Hz power at resected pairs did not relate to seizure freedom; the nonsignificant difference was opposite to prediction
Uses the per-pair NREM HFA, MI, delta, SOZ and resection labels (results/pairs.tsv) and participants.tsv with ILAE outcome. It adds a test of the HFA-SOZ result against surgical outcome.
Within-shaft HFA and MI were higher in SOZ pairs, while HFA did not separate SOZ shafts in 15 multi-shaft patients
[Legacy relation: reproduces] Independent re-extraction from OpenNeuro gave identical features: 728 of 728 bipolar pairs, maximum absolute difference 0.0 for log HFA, MI and log delta, and 43 of 43 identical SHA-256 values of the EDF byte ranges.
supportsWithin-shaft HFA and MI were higher in SOZ pairs, while HFA did not separate SOZ shafts in 15 multi-shaft patients
The SOZ effect of HFA and MI stays when the comparison is restricted to pairs on the same shaft (mean AUC 0.72 and 0.72, 41 patients). For HFA, the effect is not a between-shaft difference (between-shaft AUC 0.49 in 15 multi-shaft patients). This makes a shaft-level confound an unlikely explanation of the HFA result.
usesWithin-shaft HFA and MI were higher in SOZ pairs, while HFA did not separate SOZ shafts in 15 multi-shaft patients
We used its feature-extraction code (code/fetch_features.py, one change: skip runs without a local channels.tsv), its archived channels.tsv files and file list, and its pairs.tsv for the reproduction check.
usesHigh NREM HFA and slow-wave coupling in resected SEEG pairs showed no clear outcome association in 36 patients
Uses the SEEG pair features (HFA, MI) and resection/SOZ labels from results/pairs.tsv and the participants table of this publication.
usesA z-scored interictal SEEG model transferred across awake centers with no meaningful loss, but awake-to-sleep loss remained uncertain
ds007118-20 NREM sleep pair table (results/pairs.tsv) is the sleep input.
usesArchived NREM SEEG features did not reliably distinguish resected from unresected non-SOZ pairs in 13 patients
Source of the archived pair features (pairs.tsv) and participant outcomes used in all tests.
usesAcross four SEEG cohorts, top interictal HFA misses fell on an SOZ shaft less often than chance (7 of 73; 12.2 expected)
Origin of the ds007118/9/20 NREM per-pair table.
supportsSlow-wave coupling identified SOZ pairs in 30 ds003876 patients but did not outperform high-frequency or delta power
The same MI, HF power, and MI residual definitions gave SOZ AUCs above chance in a different multicenter dataset (ds003876). The MI residual AUC (0.616) was equal to the NREM value. The raw MI AUC was a little lower (0.668 vs 0.729).
usesSlow-wave coupling identified SOZ pairs in 30 ds003876 patients but did not outperform high-frequency or delta power
Feature definitions, NREM reference values, and the MI-residual method.
usesNREM delta power showed no age-related decrease in bipolar SEEG contacts
We reused its byte-range EDF reading code, its file list of ds007118-ds007120 v1.0.0 and its participants table.
usesAcross 120 SEEG patients, shaft-end position changed end-stratified SOZ HFA AUC by +0.002
ds007118/9/20 NREM per-pair features.
usesIn 117 SEEG patients, the top interictal HFA pair was in the clinical SOZ less often than its AUC predicted
Per-pair NREM log_hfa table for ds007118/9/20.
usesNREM SEEG features showed group-level SOZ separation, while patient-level detection varied by feature
Archived NREM bipolar-pair features and clinical SOZ labels are the only input of this analysis.
usesBoundary SEEG pairs show intermediate NREM high-frequency activity
Input pair table (results/pairs.tsv, same SHA-256) with NREM HFA, MI and delta features and clinical contact labels.
usesIn 34 SEEG patients, resection overlap with high NREM delta-phase coupling or high-frequency activity did not differ by seizure outcome
The feature extraction code (bipolar pairs, MI, HFA, log delta) and the participants and file list provenance. We extended the code to all subjects with SEEG contacts and tested the outcome contrast that was only exploratory there.
usesNREM slow waves co-occur across SEEG shafts above chance, but absolute involvement varies with detection settings
EDF byte ranges, SHA-256 values, channel status labels and pair rules for the 43 NREM 90-s windows
Spike-masked slow-wave-HFA coupling retained an SOZ signal, but its controlled remainder was small in 43 SEEG patients
[Legacy relation: reproduces] Our independent re-extraction of its 30-120 s features was identical: 728 of 728 pairs, max absolute difference 0 for log HFA, MI and log delta, 47 of 47 byte-range SHA-256 values. The raw MI SOZ AUC was 0.729 (its value 0.73).
usesSpike-masked slow-wave-HFA coupling retained an SOZ signal, but its controlled remainder was small in 43 SEEG patients
We used its feature code (fetch_features.py), run selection, bipolar pair definitions, clinical labels and archived pairs.tsv. We added a spike-like transient detector and a spike-masked MI to that code.
usesIn 43 SEEG patients, age did not predict sleep-feature SOZ separation or explain lower Detroit HFA separation
Uses the archived per-patient SOZ AUC values, pair features, and participants table of this study.
usesIn 27 epilepsy patients, interictal discharges locked weakly to distant slow-wave phase, without reliable seizure-zone strengthening
Pair selection and SOZ labels follow its rules.
usesIn 43 SEEG patients, tonic 250–450 Hz power separated clinical SOZ pairs but did not outperform 80–150 Hz power
We reused its feature code, run and pair selection, file list, channel tables and participants table. We split its 80-300 Hz band into 80-150, 150-250 and 250-450 Hz and reproduced its 80-300 Hz feature exactly.
usesIn 43 SEEG patients, spindle deficits at SOZ and resected pairs did not differ by ILAE outcome (180 s NREM; Holm p = 1.0)
Subject selection and bipolar pairing under the spindle features.
usesShort interictal HFA did not rank SOZ shafts in 71 patients, despite higher pair-level HFA
Origin of the ds007118-20 NREM 80-300 Hz HFA pair table.
usesContiguous shaft-block sham labels inflate pooled-pair SOZ false positives in four SEEG cohorts
Source of the ds007118-20 pair table.
usesIn 43 SEEG patients, spike-ripple rate did not outperform spike rate for clinical SOZ pairs in 90 s NREM recordings
We used its patient set, bipolar pairs, clinical SOZ labels, EDF byte ranges with SHA-256 values, and its archived 80-300 Hz log power per pair. We re-read the same bytes and verified all 43 hashes.
usesSOZ extent showed no detectable relation to NREM marker separation in 43 SEEG patients
All input features (pairs.tsv, per_patient.tsv) are the archived results of this publication; per-patient AUCs reproduce within 5e-5.
usesDetected spindle density was lower at seizure-onset-zone pairs, but the difference depended on spike rejection and was not independent of power
We used its subject selection, bipolar pairing, numpy EDF byte-range reader, band-power code, statistics helpers, and its archived file list and 47 channels.tsv files. These gave the same 43-patient SOZ contrast set. Our secondary delta (0.5-4 Hz) AUC of 0.67 in a 180 s window agrees with its delta AUC of 0.656.
usesContiguous null labels inflate per-patient SEEG seizure-onset-zone false-positive rates
ds007118-20 pair table (log_hfa, log_delta, SOZ) is a secondary cohort.
usesClinical spiking labels explain part of the non-SOZ top-HFA tail in 21 SEEG patients
Origin of the per-pair ds007118-20 NREM 80-300 Hz power table and the clinical contact labels.
usesWithin cohorts, NREM slow-wave power did not decrease with age in 42 patients recorded with SEEG
We reused its byte-range EDF reader functions (get, edf_header), its ds007118/9/20 file list and its participants table. These gave the data access for this age study.
usesOne fixed cross-patient threshold retained most SOZ separation in 43 NREM SEEG patients
We used the archived per-pair features (results/pairs.tsv: log HFA, MI, log delta, SOZ labels) and provenance/participants.tsv (cohort labels) as the only input. Our mean within-patient AUCs match its values exactly.
usesInterictal SOZ AUCs did not exceed a shared-label null in seven tests across 115 SEEG patients
Source of the ds007118-20 NREM per-pair feature table.
usesPublic OpenNeuro coordinates cover only 13 of 160 SEEG SOZ pairs, leaving normative HFA testing underpowered
We use its 728 SEEG pair features (NREM HFA power, SOZ labels) and its run list.
usesOpenNeuro electrode coordinates do not verify the within-shaft SOZ result for 728 archived bipolar pairs
We use its archived per-pair NREM features (log_hfa, mi, log_delta) and SOZ labels for 43 patients.

More citing publications