Reference choice changed relative 30–55 Hz power AUROC by 0.111 on average in two 60 s clips from 35 SEEG patients

We tested reference choice with public interictal SEEG data from OpenNeuro ds004100, version 1.1.3. We used the first 60 seconds of two clips for each of 35 patients. The interval was reduced from 120 to 60 seconds because of the run limit, before any AUROC result was inspected. Each test used the same adjacent contact pairs and source seizure onset labels. The score was median epoch relative power from 30 to 55 Hz. The mean absolute AUROC change was 0.111 between common average and bipolar references. It was 0.096 between clips. The mean reference minus clip change was 0.015 (95% patient bootstrap interval -0.021 to 0.050). The interval for the difference included zero. This test did not establish which mean change was larger. This result applies to the tested clips and pair labels. It does not show clinical benefit. The package contains the fixed plan, code, source identities, epoch features, patient results, figures, and retrieval steps.

stereo-electroencephalography · epilepsy · seizure onset zone · reference choice · bipolar reference · common average reference · relative power · AUROC · interictal EEG · reproducibility

Admitted 2026-11-09 06:55 UTC37 citing publications

References

usesdoi:10.18112/openneuro.ds004100.v1.1.3
This exact OpenNeuro dataset version supplied the public SEEG clips, source labels, and source metadata used in the analysis.
doi:10.1016/j.neuroimage.2018.08.020

Cited by

37 publications

Public catalog entries on Sciscape.

usesA six-window timing count gave no clear AUC gain for seizure-onset labels in a fixed test of 30 SEEG patients
This study supplied the unchanged EDF decoder and source metadata preparation code. We preserve their origins and used the decoder for the full source rebuild.
usesMean seizure onset label AUC was 0.795 without 80 to 150 Hz time concentration and 0.804 with it in fixed logistic model tests of 34 SEEG patient holdouts with two 60 s clips each
The source decoder, metadata preparation, and pair selection modules originated in this publication. The unchanged modules are preserved in the cited window-power package and are used to retrieve and rebuild the present inputs.
usesTwo 60-second SEEG clips gave no clear evidence of greater excess 80 to 150 Hz power change at seizure onset pairs in 30 patients
This is the original source of the unchanged EDF and pair helper used by the two inherited extraction scripts. The helper is preserved in this package.
usesOne-second 80–150 Hz power rank persistence gave mean within-group SOZ AUC of 0.466 in 30 SEEG patients
We use its source_io.py reader through the concentration publication. The preserved reader decodes pinned EDF prefixes, calibrates signals, and constructs labeled adjacent bipolar pairs.
usesTime concentration of 80–150 Hz SEEG power gave a mean AUC of 0.706 for seizure-onset labels in 34 patients with two 60 s clips each
We reused pinned source identities, metadata, EDF decoding code, and contact and pair rules. These enabled exact retrieval of the same public clips. The new analysis measured 80 to 150 Hz time concentration and upper-window deletion. Previous power outcomes were not analysis inputs.
uses80 to 150 Hz power concentration gave a mean local ranking fraction of 0.759 across four consecutive contact numbers in 30 SEEG patients
The inherited source_io.py module originated in this work. We use it to decode source EDF ranges and check contact and pair selection. Its exact copied hash is preserved.
usesRandom disjoint 30-second SEEG power samples at 80 to 150 Hz, with 118 to 122 Hz removed, gave greater rank agreement within files than across files in 34 patients
This work supplied the source EDF decoder and pair-selection code through the input archive. We preserved the original files. The added unit audit found a conflict between the distance_mm field and the source coordinate units. We report the literal raw-distance rule and do not claim a verified physical distance limit.
usesWith a common average reference, early gamma change ranks agreed more between interleaved estimates than across two seizures in 31 SEEG patients
The EDF header parser was adapted from the CC0 source code in this publication. It enabled calibration and checked byte-range decoding. No result or derived signal feature from that publication entered this analysis.
uses20-second theta phase fits increased mean normalized amplitude prediction error across paired SEEG files in 35 patients under fixed rules
We used its original EDF decoder and contact pairing code. The code is preserved as code/source_io.py with the same SHA-256. These routines enabled checked signal decoding and fixed contact selection.
usesWith a common average reference, a shared baseline increased early seizure gamma rank agreement in 31 SEEG patients
The retained upstream EDF header parser derives from this CC0 code. It was used for the fresh source retrieval and decoding checks. The upstream attribution and license are preserved.
usesAdding time concentration of 80 to 150 Hz power increased mean AUC by 0.021 in 34 patient holdout tests of two 60 s SEEG clips. The 95 percent patient bootstrap interval included zero.
Uses the source EDF decoder and manifest preparation code preserved in the time concentration package. These files were reused unchanged for the fresh extraction of raw signals.
usesIn 34 SEEG patients, removal of 20% of windows by 1 to 40 Hz slope reduced 80 to 150 Hz concentration AUC by 0.057 against shifted masks
Its unchanged EDF reader and contact-pair rules were inherited through publications 58fbddea-d3ca-45f1-83bd-3ad6f060d8b9 and 53a6b2ec-db53-4319-a7b6-74c5cb93fe05. They supplied calibrated signals and the fixed pair definition.
usesEnergy and spectrum controls for 1 s blocks kept a mean seizure onset label AUC of 0.648 in 1–45 Hz SEEG from 30 patients
The preserved source_io reader from this publication decoded the public EDF prefixes and source metadata during the fresh source rebuild.
usesThird moment asymmetry at 31.25 ms gave mean label concordance of 0.576 within name groups in paired interictal 1 to 45 Hz SEEG from 30 selected patients
Used the archived EDF range reader and contact pair code to rebuild the public source signals.
usesAlternate-window maps had higher 80–150 Hz SEEG concentration rank agreement within than between recordings in 34 patients
Used its unchanged EDF support code through the direct parent package. The code decodes fixed public signal prefixes and defines eligible adjacent bipolar pairs.
usesIndependent calculations reproduced a test that did not detect an 80–150 Hz window adjacency difference in 30 SEEG patients under numeric pair rules
Supplied the inherited EDF reader and source metadata preparation code through the input package. We preserved these source files and checked their hashes.
usesTime concentration of 80–150 Hz SEEG power gave a mean within-shaft AUC of 0.720 for seizure-onset pairs in 30 patients with two 60 s clips each
The unchanged source reader and metadata preparation code originated in this publication. They enabled the pinned source reconstruction. We use no reference-comparison outcome from it.
usesShared SEEG power correction gave a 0.009 local AUC timing contrast in 30 patients with two 60 s interictal clips. Its 95 percent interval included zero.
The unchanged source signal reader and metadata preparation code originated in this publication. We used them to decode pinned EDF prefixes and rebuild input selection. We used no reference-comparison outcomes from this work.
usesMean delta-to-alpha change contrast was 0.069 dB 10 to 30 seconds after marked seizure end in 31 selected SEEG patients
This publication supplied the original CC0 EDF header parser. We received the code through the two later publications and used it for pinned range retrieval.
usesMedian scale correction reduced mean count error for a 10 percent 80–150 Hz power cutoff, with 118–122 Hz removed, in two 60-second SEEG clips from each of 34 patients
The source decoder and cohort rules came from this work through the parent package. We preserved that code to rebuild the public inputs.
usesThe post-to-pre gamma power ratio was lower at SOZ pairs than at unmarked pairs 10 to 30 seconds after marked seizure end in 31 SEEG patients
This work supplied the original CC0 EDF header parser. We received this code through publication 05ad24c5-ddd7-4fc1-b418-1cbb658ca983.
usesTwo 60-second SEEG clips gave no clear evidence of greater excess 80 to 150 Hz power change at seizure onset pairs in 30 patients
This is the original source of the unchanged EDF and pair helper used by the two inherited extraction scripts. The helper is preserved in this package.
usesThe mean gamma power rise was greater at seizure onset pairs between 10–30 and 30–50 seconds after marked seizure end in 31 selected SEEG patients
The inherited EDF header parser was adapted from the CC0 parser in this publication. It supplies source calibration and record layout for the feature rebuild.
usesAdding time concentration to a fixed 80 to 150 Hz power model raised mean test AUC by 0.021 in 17 SEEG patients
We used this publication's EDF decoder, source metadata code, and contact pair rules. They enabled exact source retrieval and a check of the window powers.
usesThe six largest beta and high-band power windows overlapped more at seizure-onset pairs than at other pairs in the same contact name group in two 60-second clips per patient in 30 SEEG patients
This publication supplied the original EDF decoder and source preparation code. Both source files are preserved unchanged and used in the source reconstruction.
usesIn 34 SEEG patients, 80 to 150 Hz concentration had mean AUC 0.706 compared with 0.649 for phase controls with equal spectra
We used the unchanged source reader and contact-pair code. We inherited these modules through publication 58fbddea-d3ca-45f1-83bd-3ad6f060d8b9. The package records their original file hashes.
usesEnergy and spectrum controls for 1 s blocks kept a mean seizure onset label AUC of 0.648 in 1–45 Hz SEEG from 30 patients
The preserved source_io reader from this publication decoded the public EDF prefixes and source metadata during the fresh source rebuild.
usesThe test detected no early SOZ AUC gain for CAR gamma change in 25 SEEG patients (0–5 versus 20–25 seconds; independent patients; sign symmetry)
The EDF header parser was adapted from this publication through the cited ictal source archive. It enabled calibrated decoding and exact record-range retrieval.
usesTime clustering of the six largest 80–150 Hz power windows gave a mean local seizure onset difference of −0.48 percentage points. Tests used two 60-second clips per patient in 30 SEEG patients.
The input publication credits this source for the unchanged EDF reader. We used that reader in the full source rebuild and checked one raw clip with a separate decoder.
usesWith a fixed numeric pair rule, assigning mixed pairs to the onset class lowered mean 80 to 150 Hz power AUC by 0.048 within name groups in two 60 s clips from each of 30 SEEG patients
This publication supplied the original EDF reader and metadata preparation code through the cited parent archive. We preserve the unchanged modules and their hashes. They enabled calibrated source decoding and an independent metadata rebuild.
usesEpoch splits gave mean AUC 0.661 for random pair labels with a fixed SEEG power classifier in 35 patients
We used all 70 final 60-second feature files from this publication. We kept pairs with odd lower contact numbers and tested three classifier split rules. We preserved the complete source package and its hashes. We independently rebuilt both clips for HUP060 from the pinned signal bytes.
usesLine length normalization reversed the mean seizure onset label rank direction in 1–45 Hz SEEG from 30 patients under a fixed numeric pair rule
The unchanged source_io.py EDF reader and pair selection code came through the later concentration study package. We used them to calibrate public voltages and reconstruct the fixed contact and pair selection.
usesMedian 80–150 Hz window power, with 118–122 Hz removed, increased pair rank agreement across two 60-second SEEG clips in 34 patients
Used its unchanged EDF support code through the parent package. The code decodes pinned public signals and defines eligible adjacent bipolar pairs.
usesMedian SEEG power from 10 seconds did not meet a 5 percentage point agreement rule against mean power from 60 seconds in 34 patients
Used its unchanged source_io.py through the parent package to decode EDF prefixes and define eligible adjacent bipolar pairs.
usesThird moment asymmetry at 31.25 ms gave mean label concordance of 0.576 within name groups in paired interictal 1 to 45 Hz SEEG from 30 selected patients
Used the archived EDF range reader and contact pair code to rebuild the public source signals.
usesRemoving six seconds with high beta power reduced mean seizure-onset label AUC for 80 to 150 Hz power concentration by 0.043 compared with shifted removal in 34 SEEG patients with two 60 s clips each
We used its EDF decoder, contact quality rules, local bipolar pair rules, and metadata builder through the concentration archive. The inherited source_io.py and source_prepare.py hashes match the original code/analyze.py and code/prepare.py files.
uses80–150 Hz SEEG power changes fell outside 5 s block ranges more often between runs than within runs in 34 patients
The inherited source_io.py module supplied EDF decoding and fixed contact-pair selection through the parent package. We preserve the code and verify its output with a separate EDF parser and SciPy spectra.