Interictal power and transient features distinguish SOZ channels after depth matching in 36 SEEG patients, but aperiodic exponent and outcome do not

We compared clinically marked seizure-onset-zone (SOZ) and other channels in 36 patients with stereo-EEG epilepsy recordings from OpenNeuro ds004100 version 1.1.3 (20 seizure-free, 16 not seizure-free). For each patient, we analyzed adjacent-contact bipolar channels from 20–90 s of interictal run-01 and an independent run-02 clip. In run-01, absolute band power, line length, and spike-like transient rate were higher on SOZ channels. Their depth-matched mean within-patient AUCs were 0.63–0.72 (p ≤ 4×10⁻⁴), below the Bonferroni threshold of 0.0031; 78–89% of patients had AUC > 0.5. SOZ channels were nearer shaft tips (contact-index AUC 0.33), but effects persisted when comparisons used only non-SOZ channels at the same contact index and within shafts carrying SOZ channels. The 30–55 Hz aperiodic exponent did not separate SOZ channels (AUC 0.50–0.51). Its per-patient direction was consistent across clips (r = 0.78), but differed between patients. The 1–45 Hz exponent effect was weak and not Bonferroni-significant. Relative power shifted toward delta in SOZ channels; relative alpha, beta, and high-gamma power was lower (AUC 0.36–0.39). Run-02 replicated the main findings. No feature consistently separated SOZ or treated channels better in seizure-free patients: a run-02 spike-rate difference (p = 0.018, uncorrected) was absent in run-01 (p = 0.62), and no treated-channel comparison showed seizure-free superiority. Recordings were short, the spike detector was not validated against expert marks, tissue type was not controlled, outcome groups were small and mixed resection with laser ablation, and no out-of-patient classifier was trained.

stereo-EEG · epilepsy · seizure onset zone · interictal · bipolar channels · band power · line length · spike-like transients · aperiodic exponent · depth matching

Admitted 2026-11-09 11:12 UTC43 citing publications

References

supersedesInterictal SEEG power and transient rate separate SOZ channels after depth matching; the 30–55 Hz exponent does not sci:sha256:11c938f4a42307e9ce83c816d45fb73133a2ead9fe3d48e24848ab7c6a43a4c8
Same package and data. This version corrects one number in the summary: the depth-matched p-values are <= 4e-4 (log beta power, run-01, p = 0.00038), not < 1e-4.

Cited by

43 publications

Public catalog entries on Sciscape.

usesIn 36 HUP SEEG patients, the source-sink index marked SOZ pairs but added no information beyond broadband power
We use its fetch code, the same interictal segments (same byte ranges), the same bipolar pairs and SOZ labels, and its EDF reader and AUC code.
usesIn HUP SEEG, ictal markers identify clinician SOZ contacts, but their power advantage does not replicate
We reused its S3 key listing of ds004100, its participants.tsv copy (Engel outcome), the design of its HTTP range-request fetch code, and its per-patient interictal SOZ AUCs (results/patient_auc.tsv) as the interictal comparator in compare_interictal.py.
supportsPermutation entropy adds no SOZ information beyond the power spectrum in 34 HUP SEEG patients
In our exploratory comparison on 60-s clips from 34 SEEG subjects of the same dataset, log band power (AUC 0.69 for 1-40 Hz, 0.67 for 80-150 Hz), line length (0.66) and spike rate (0.69) separated SOZ pairs from non-SOZ pairs. This agrees with that result. The data overlap, so this is not an independent replication.
usesInterictal SEEG Lempel–Ziv complexity was lower in the SOZ but added no SOZ information beyond line length and variance
We used its segment fetcher, EDF reader, bipolar montage and SOZ labels without change, and we compare LZC with its line-length result.
usesIn 36 HUP SEEG patients, band-power and line-length SOZ AUCs repeated across two 70 s clips, but spike-like-rate AUC was less reliable
Its feature pipeline (bipolar pairs, 20-90 s segment, Welch band power, line length, spike-like rate) defines the features whose per-patient SOZ AUC we tested for clip-to-clip reliability.
challengesIn 36 HUP SEEG patients, region matching reduced high-gamma and line-length SOZ separation; spike-like rate remained
It reports that interictal high-gamma power and line length separate SOZ channels after contact-depth control. Inside tissue x region strata, these two features no longer separate SOZ pairs (stratified AUC 0.569 and 0.551, Holm p 0.22 and 0.50; run-02 replicates). Its spike-like rate, delta power and null exponent results stay.
usesIn 36 HUP SEEG patients, region matching reduced high-gamma and line-length SOZ separation; spike-like rate remained
We used its per-pair interictal features for run-01 and run-02 (input/channel_features*.tsv) and its participants.tsv, with its SOZ and resection labels.
usesIctal energy-ratio rise localizes clinical seizure-onset channels but does not outperform interictal markers or predict surgical outcome
We used its per-patient interictal AUC table (results/patient_auc.tsv, copied to results/prior_interictal/) for a paired ictal-vs-interictal comparison, its S3 key listing of ds004100 to select ictal runs, and its EDF segment reader as the base for our reader.
usesInterictal spike-rate maps agreed with early ictal gamma maps but did not predict Engel I outcome in 36 HUP SEEG patients
The interictal maps are its per-pair spike rate, band power and line length features.
usesTwo interictal clips raise spike-rate SOZ AUC, but not band-power or line-length AUC, in 36 HUP SEEG patients
Its pipeline computed the bipolar band power, line length and spike-like rate features from ds004100.
usesIn 35 HUP SEEG patients, ictal high-frequency power agreed with clinician SOZ labels, while EI did not outperform interictal delta
We used its results/channel_features.tsv (interictal run-01 bipolar per-pair log delta power and SOZ labels for 36 ds004100 SEEG patients) as the interictal comparator. We matched pairs on subject, shaft and contact number (100% matched, SOZ labels agree 100%).
usesInterictal spike-like rates track earlier seizure recruitment in 36 SEEG patients; the small effect is partly explained by proximity
Source of the per-pair interictal features (spike-like rate, band power, line length) for run-01 and run-02.
usesIn 28 HUP SEEG patients (ds004100), SOZ bipolar pairs did not lead interictal spike events (mean AUC 0.55, p = 0.21), but they had higher spike rate and amplitude
We used its fetch code, S3 key list, EDF reader and SOZ label rule. We also replicate its result that SOZ pairs have higher spike-like event rate.
usesIn 90-s SEEG clips, post-spike 70–150 Hz activity fell about 2% and did not significantly separate SOZ pairs
Interictal spike-like transient rate as an SOZ marker; we use spike rate as the control for the post-spike index.
usesA z-scored interictal SEEG model transferred across awake centers with no meaningful loss, but awake-to-sleep loss remained uncertain
HUP ds004100 per-pair interictal feature table (results/channel_features.tsv) is the HUP input.
usesAperiodic-knee timescales did not increase anteriorly in 36 epilepsy patients; post hoc, inferior pairs had longer timescales
The fetch code, the segment rule (EDF records 60-150, 8 MB cap), the EDF reader, the bipolar pairing logic and the S3 key list for ds004100 were reused from this publication.
usesDistance-corrected SEEG node strength did not differ by seizure onset zone in 34 patients, while SOZ-to-SOZ edges were stronger
We used its fetch_segments.py (EDF byte-range recipe), its S3 key list for ds004100 snapshot 1.1.3 and its EDF/TSV/AUC helper functions (copied into code/helpers.py).
usesIn 36 HUP patients, seizure-onset-zone pairs were net sources of bivariate Granger influence, but the effect did not persist after power correction
Reused analyze.py (EDF segment reader, tie-aware AUC, contact-label parser) without change.
usesIn HUP SEEG, an ictal EI-like map marked clinical SOZ pairs but did not separate surgical outcome groups
Its resection-versus-outcome design (resect label on either contact, S vs F permutation test) and its interictal SOZ results guided the choice to test ictal maps against treatment and outcome instead.
supportsOpenNeuro ds004100 labels are stable across runs, but channel types and bad flags have limits
That work uses the ds004100 SOZ labels. This audit shows that these labels are stable across runs. It also gives a caveat: the bad flag predicts non-SOZ status a little (mean per-patient AUC 0.517 to 0.544).
usesPosterior SEEG pairs had more frequent and taller 6–14 Hz peaks, while peak frequency showed no anterior–posterior gradient
This package is the origin of the ds004100 range-request fetcher and the EDF segment reader that we used.
usesIn 32 HUP SEEG patients, SOZ pairs separated within shafts, but contact depth alone matched high-gamma power
The per-pair features were computed by this publication from ds004100.
usesMirror-matched interictal HUP SEEG pairs show stronger 1–40 Hz coupling in 28 of 29 patients; the 65–115 Hz effect is small and not robust
code/helpers.py (EDF reader, TSV reader, FFT band-pass) comes from this publication through 57a74009.
supportsIn 24 HUP SEEG patients, status labels agreed with quality rules, but z>3 amplitude filtering flagged SOZ contacts more often
This study also finds higher broadband amplitude on SOZ contacts in ds004100 (exploratory AUC 0.72), and shows that the status=good filter used there removes almost no SOZ contacts.
supportsIn 36 HUP SEEG patients, spike leader fraction did not identify SOZ pairs, and lead rate added no information beyond spike rate
On different records of the same files (150-270 instead of 60-150), spike-like transient rate again separates SOZ bipolar pairs in 36 patients (mean AUC 0.73, 95% CI 0.66-0.79, 30/36 patients above 0.5, sign-flip p = 5e-5).
usesIn 36 HUP SEEG patients, spike leader fraction did not identify SOZ pairs, and lead rate added no information beyond spike rate
We used its S3 key list of the 37 ds004100 SEEG run-01 files, its range-request fetch script (changed to records 150-270), its EDF reader, bipolar montage, SOZ rule, FFT band-pass and AUC functions (code/analyze.py, copied unchanged as code/f717_analyze.py), and its 10-60 Hz 6-sigma detector rule.
usesSOZ pairs in 35 HUP SEEG patients showed a small 1–4 Hz clip-to-clip change after level control
Feature pipeline (band power, line length, spike-like rate) that made the archived features.
usesIctal SEEG markers aligned with clinical seizure-onset channels but did not outperform interictal power or predict outcome
We used its S3 key listing of ds004100 (results/s3_keys_ds004100.tsv) to select files, and its per-patient interictal bipolar AUCs (results/patient_auc.tsv) for the paired comparison of ictal and interictal SOZ separation (H2). We adapted its EDF-reading and bipolar-montage code.
usesIn 36 HUP SEEG patients, longer epilepsy duration was not associated with a larger SOZ or more non-SOZ spiking; the high-gamma contrast did not pass Holm correction
The per-pair interictal features of run-01 and run-02 and the clinical SOZ labels are copied from this publication without change.
usesInterictal PAC separated clinical SOZ channels after band-power control in 35 SEEG patients, but less than power or spike rate alone
We used its byte-range fetch code, segment rule, bipolar montage, spike-like transient detector and AUC code. We added PAC and covariate control.
usesIn 34 SEEG patients, raw envelope DFA did not separate SOZ channels; power-adjusted high-gamma DFA was lower
Reuses its HTTP byte-range fetch method, bipolar montage rule and AUC code for ds004100.
supportsWithin SEEG shafts, depth separated SOZ pairs, while depth adjustment modestly reduced HFA separation across three cohorts
Its HUP result that SOZ pairs are near the shaft tip and that depth matching does not remove the feature effect agrees with our depth-matched AUC 0.66 in 3 other cohorts.
supportsIn 32 HUP SEEG patients, 60-Hz line-noise excess was lower on SOZ pairs and did not explain high-frequency separation
The higher high-frequency power on SOZ channels in ds004100 stays after adjustment for 60 Hz line-noise excess (AUC 0.664 to 0.650).
usesIn 28 HUP SEEG patients, leading spike events marked SOZ pairs less well than spike rate (AUC 0.68 vs 0.76; Holm p < 0.001), while event order did not (AUC 0.52)
We used its spike detector (10-60 Hz, 6 robust sigma, 200 ms refractory) and its segment rule, and added spike timing and event grouping.
usesInterictal SEEG coherence marks SOZ-to-SOZ links but does not improve power models in 36 HUP patients
Used its segment retrieval code (fetch_segments.py), its EDF/TSV/AUC helper code (analyze.py, copied as analyze_f717.py), its S3 key list and its bipolar pair definition. This let us compare connectivity with the same local power features on the same data.
supportsIn 35 HUP SEEG patients, higher SOZ 50–70 Hz power lowered line-noise excess, while LNE adjustment changed the 80–150 Hz SOZ AUC by −0.013
Its interictal SOZ high-frequency power effect remains after we remove 60 Hz line-noise excess (bipolar HFA AUC 0.672 to 0.659).
usesIn 35 HUP SEEG patients, higher SOZ 50–70 Hz power lowered line-noise excess, while LNE adjustment changed the 80–150 Hz SOZ AUC by −0.013
We adapted its HTTP range-read fetch code and EDF reader, and its SOZ label rule, for the same ds004100 cohort.
usesInterictal spike rate weakly predicted earlier seizure recruitment outside the SOZ in HUP SEEG
Per-pair interictal features from two separate clips (results/channel_features.tsv and results/run02/channel_features.tsv) are the predictors; clip 2 gives the replication test.
usesIn 30 SEEG patients, unmatched SOZ–SOZ connectivity replicated, but shaft-pair matching was inconclusive
code/helpers.py (EDF byte-range reader, TSV reader, FFT band-pass) is copied from this publication via 57a74009.
supportsInterictal delta to high-gamma coupling marks SOZ pairs but adds no information after covariate control
Our run-01 power and spike-rate SOZ AUCs equal those of f717daf8 (same code and data). In a second, independent interictal clip (run-02) of the same 36 patients, we get similar values (delta 0.708, high-gamma 0.683, spike rate 0.697). We also show that delta-HG PAC adds no SOZ information beyond these features.
usesInterictal delta to high-gamma coupling marks SOZ pairs but adds no information after covariate control
Reused unchanged the byte-range fetch code, the S3 key list, participants.tsv, and the EDF reader, Welch band power, spike-like transient detector, AUC and sign-flip functions of f717daf8. Our power and spike AUCs reproduce its values (0.710, 0.680, 0.714).
usesInterictal spike rate is higher on the seizure-onset side at mirror-matched contacts in HUP SEEG
We use its per-bipolar-pair interictal features and SOZ labels (run-01 and run-02) as input.
usesSOZ line-length excess was similar before seizures and during interictal clips in 21 patients
Interictal run-01 and run-02 per-pair features are the input of the interictal SOZ contrast and of the null control.
supportsIn 35 HUP patients, spike-leader fraction did not mark the clinical SOZ beyond spike rate
On a different 76-240 s segment, the spike-like transient rate marks SOZ pairs again (mean AUC 0.74, 35 patients).
usesIn 35 HUP patients, spike-leader fraction did not mark the clinical SOZ beyond spike rate
We reuse its byte-range fetch code, EDF reader, bipolar pairing, SOZ labels and spike detector.
usesEarly-ictal gamma change separated clinical SOZ pairs in 36 patients but did not outperform pre-ictal power
We used its S3 key listing of ds004100 (results/s3_keys_ds004100.tsv) to find ictal SEEG runs, and adapted its HTTP range-request EDF fetch logic. Its interictal power AUC (about 0.7) is the reference that our pre-ictal power result (0.69) agrees with.
usesUsing one Welch method removes the observed pre-ictal SOZ gamma excess in 20 HUP SEEG patients
Its median-Welch interictal design (2-s Hann, interictal run-01) defined the interictal arm of the 2c256d02 test that we repeat with one method.
usesIn 37 HUP SEEG patients, pairs away from the seizure onset zone showed posterior alpha and pre/postcentral beta gradients in two clips
It computed the per-pair band power and exponent features from ds004100 run-01 and run-02 clips.