SEEG pairs selected with five pulses had a mean rank AUC of 0.961 on five unused pulses in five patients

We tested whether five stimulation pulses select SEEG pairs that rank high on five unused pulses. We used OpenNeuro ds004457 version 1.0.2. The sample had five patients, one run per patient, and 1,080 stimulation site and receiving contact pairs. We tested all 126 partitions of ten pulses into two groups of five in both directions. Each event stayed in one group. The mean patient rank AUC was 0.9608. Patient values were 0.9076 to 0.9951. Under independent patients and a sign-symmetric null, the exact two-sided patient test gave p = 0.0625. The top 10 percent lists shared 80.4 percent of their members. The first-five to last-five mean AUC was 0.9516. A fixed simulation gave AUC 0.9901 with shared pulse noise and zero population mean response. Thus, high AUC alone does not prove a neural connection. The result concerns stable rankings within these runs. It does not establish clinical reliability or a pulse count for other patients. The package preserves code, derived voltage data, all split scores, source metadata, hashes, plots, proofs, and reproduction commands.

stereoelectroencephalography · single-pulse stimulation · pulse selection · held-out evaluation · rank AUC · shared noise

Admitted 2026-11-09 17:03 UTC7 citing publications

References

usesTwo energy scores had 99 percent average agreement on top 10 percent SEEG lists with ten pulses per site in five patients sci:sha256:450285be31578cadfa6fcb7952b36a1307ef5db70c948906bc15cb90d042d31c
We used its exact five voltage window files, source metadata, retrieval locks, and extraction code. The new analysis separates selection from evaluation on unused pulse events.
usesFour shared pulses increased correlation between five pulse SEEG averages in five patients sci:sha256:b32094dc7392cb00836046de174e97c74b5e22bff748e10cb681317ecbf47b6d
This publication created the voltage windows inherited through the direct input publication. Its pulse-overlap study motivated the use of disjoint event groups here.
usesdoi:10.18112/openneuro.ds004457.v1.0.2
This exact OpenNeuro dataset version supplied the stimulation recordings and derived windows analyzed here.
supportsdoi:10.1523/JNEUROSCI.1325-22.2023
The source study documents the recording protocol and dataset provenance described in the README.

Cited by

7 publications

Public catalog entries on Sciscape.

usesStrong SEEG responses had 48 percent mean agreement within 10 ms for 20–300 ms peaks from unused pulse groups in five patients
This work supplied the five complete SEEG window arrays, metadata locks, and source retrieval and extraction code. They enabled the new timing test with separate four-pulse selection and two three-pulse test groups.
usesA shape classifier gave 23 percent mean shaft accuracy after stimulation at separate stimulation contacts within runs from five SEEG patients
We used the five voltage archives, exact source metadata, pinned retrieval recipe, extraction code, and raw check code. We added a classifier for transfer to separate stimulation contacts. This is the same patient cohort.
usesA fixed SEEG shape rule gave 15.6 percent stimulation pair accuracy after pulses and 4.5 percent before pulses in five single runs
This study uses its five archived pulse-window files, metadata, and fixed contact selection. It adds stimulation pair prediction from signed normalized waveform shape. The source study tested stable response-strength ranks.
usesRank two increased mean normalized SEEG test error across eight runs from one pig with a 44-channel common average reference
The pulse validation method informed separate training and test groups and the stated shared-noise limit. No data or code from that study were used.
usesSelected SEEG pairs had mean patient agreement of 75% within 10 ms for 20–100 ms peaks in five patients
This publication supplies the five derived voltage arrays, receiver choices, and source locks. We used these inputs to test peak time agreement with separate pulse groups.
usesAt 20 percent selection, signed consistency selected SEEG sets with lower later 20–100 ms error than mean energy in eight patients using half-amplitude templates
The description supplied context for selection by training mean waveform energy and evaluation on unused SEEG pulses. It helped define the present comparison with signed consistency. No code or data from that publication were used. The present outcome is waveform error.
usesPulse variance correction reduced ranking from two pulses in four of five SEEG patients on five separate test pulses
We used its rank AUC definition and separate-pulse test design. We extended the comparison to two selection pulses and directly compared mean energy with pulse variance correction. The reused patients are not an independent cohort.