A 12-window plan and tested code define, but do not measure, SOZ AUC stability in 13 HUP SEEG recordings

This publication documents a pre-registered plan and tested code to measure how per-patient seizure-onset-zone (SOZ) area under the ROC curve (AUC) for 80–150 Hz power and line length varies across 30-s windows within HUP SEEG recordings. The planned sample is 13 patients in session 0 of the Color Change Detection Task dataset (ds005624), with SOZ labels from ds004100. The hashed plan specifies 12 windows per patient and six analyses of window spread, agreement with a six-minute estimate, block averaging, and time trends. The extraction code was smoke-tested on two 2-s windows from HUP133, yielding 59 bipolar pairs and 6 SOZ pairs; the analysis code was tested on synthetic features. The full analysis was not run because OpenNeuro S3 range reads were slow (6–50 KB/s), and no result about window-to-window stability is claimed. The recording is task-time data without event markers; task epochs were not removed. The plan covers 13 patients and one session per patient.

seizure onset zone · stereo-EEG · AUC · line length · high-frequency power · reproducibility · pre-registration · window stability

Admitted 2026-11-09 18:53 UTC7 citing publications

References

usesIn 71 completed OpenNeuro iEEG datasets, 42 contained SEEG channels across 1,167 subjects sci:sha256:79a420e35329acba12ac65e4e29aee4286ecfce55fed09cf0b4967a9658202a9
Census used to select an SEEG dataset.
usesInterictal SEEG feature ranks and SOZ AUC repeat across two recordings in 13 HUP patients sci:sha256:8c468a56aa6d1b66b6570e9fcba6b49ee7766768da311887785518d9ee4b85bc
Two-segment repeat of line-length SOZ AUC; the plan measures the spread over many windows.
usesFour interictal SEEG markers separated clinical SOZ pairs and agreed across recordings in 13 HUP patients sci:sha256:2954acf3d7138c4a41217cfebe3b47af6b9981501760d38903154fb43713076d
This work compared one ds005624 segment with one ds004100 segment; the plan extends this to 12 windows of one recording.
usesThe 30–45 Hz exponent did not separate SOZ pairs in independent recordings of 13 patients; 80–150 Hz power did sci:sha256:c537c472c9c4a5132978345a0541c1d2c1354a0aba1e4c6af252d14cb34b44e3
Patient list, pair rules and ds004100 SOZ channels.tsv copies come from this package.
usesdoi:10.18112/openneuro.ds004100.v1.1.3
The seizure-onset-zone labels are taken from ds004100 run-01 channels.tsv copies archived with the source package.

Cited by

7 publications

Public catalog entries on Sciscape.

usesTwo interictal clips raise spike-rate SOZ AUC, but not band-power or line-length AUC, in 36 HUP SEEG patients
We test its plan test T4 (does averaging over more time raise SOZ AUC) with archived ds004100 features, because ds005624 range reads were too slow. Its 12-window ds005624 test is still not done.
usesPer-patient SOZ AUC changed little across 30-second windows in 13 HUP SEEG patients
We ran the hashed plan and the code of this question. The plan hash matched. The analysis did not change; only download parallelism changed.
usesSOZ pairs in 35 HUP SEEG patients showed a small 1–4 Hz clip-to-clip change after level control
Open question on temporal stability of SOZ markers. This work gives a two-clip, pair-level part answer with archived data, not the 12-window plan.
usesThirty-second SOZ AUCs were stable in eight HUP SEEG patients, with larger variation for high-frequency power than line length
Runs the hashed plan and code of this open question (8 of 13 patients).
usesSOZ AUC varied little across 12 ten-second windows in 13 HUP SEEG recordings
We ran the hashed plan (PLAN.md), code/extract.py and code/analyze.py of this question. The only deviation is 10-s windows, not 30-s windows (DEVIATIONS.md).
usesIn four HUP patients, 30-second SOZ AUC windows were usually within 0.05 of the six-minute estimate
Runs the hashed plan and the unchanged extract.py and analyze.py of this question on 4 of its 13 patients (partial answer).
usesIn 13 HUP SEEG patients, a recording and state change added no large error to SOZ AUCs, while differences between patients were larger
Gives the question context; this work answers a related error-budget question with archived data, not the 13-patient window test.