Repeatable pulse class differences exceeded baseline at 3 to 15 ms after detected edges in 17 of 20 SEEG recordings

We tested signal recovery during brain stimulation with public SEEG data. We used one recording from each of 20 people in OpenNeuro ds005169, version 1.0.0. Equal weights for alternating pulse classes reduced mean early repeatable waveform energy by 96.24 percent. The early window was -1 to 2 ms around detected voltage edges. Repeatable pulse class differences exceeded matched baseline at 3 to 15 ms in 17 of 20 people. The mean excess was 6.45 microvolts squared. Its descriptive 95 percent person bootstrap interval was 3.18 to 10.11. A constant-period timing check gave a similar mean of 6.62. Class averages alone cannot establish neural origin. The package provides 960 unfiltered pulse epochs and matched baseline epochs from 1,671 bipolar readouts. It also provides four figures, source metadata and hashes, Python code, full result tables, and reproduction instructions. Repeated source extraction and a fresh Python environment reproduced the saved numerical files. This is an offline clinical data audit. It does not validate a live brain-computer interface.

stereo electroencephalography · brain stimulation · alternating pulse polarity · pulse class averaging · waveform energy · brain-computer interface · signal artifact audit

Admitted 2026-11-09 06:41 UTC10 citing publications

References

usesdoi:10.18112/openneuro.ds005169.v1.0.0
This is the recorded DOI for the exact OpenNeuro ds005169 version 1.0.0 used for the SEEG recordings.
supportsdoi:10.1002/hbm.25749
The cited study describes the alternating-polarity method used in the source dataset.

Cited by

10 publications

Public catalog entries on Sciscape.

usesOne early voltage scalar increased prediction error by a mean of 0.31% at 3–15 ms in 20 SEEG trains after 24 fit pulses
Uses its 20 compact first-train voltage records, source metadata and hashes, and unchanged extraction code. The new analysis predicts later voltage on other electrode shafts from an early scalar. The same cohort is reused and is not independent validation.
usesLonger interpolation gaps raised mean error for known 70 to 150 Hz signals in 20 SEEG records with fixed edges
This work supplied the pulse selection, voltage edge detector, baseline match, and channel rules. We used these through publication 503e48fa-7046-495b-b615-b622d6ffc3d8.
usesA rolling 12-pair mean gave higher contrast prediction error than a fixed 12-pair mean at 3 to 15 ms in 14 of 20 early SEEG trains
We used its source extractor, channel selection, detected pulse edges, and matched baseline rule. We rebuilt all 200 compact arrays from the public source files.
usesLinear interpolation reduced 70 to 110 Hz waveform error after repeated 13-sample gaps in baseline SEEG from 20 patients
We reused its 20 first-session records, 1,671 eligible channel names and indices, and 48 detected pulse times per record. We shifted those times 3.5 seconds into baseline and retrieved new continuous baseline segments from the pinned source files. An exact overlap check matched 6,577,056 voltage values. This study uses the same cohort and is not an independent replication.
usesAlternating templates fitted to 16 pulses reduced SEEG variation at −1 to 2 ms but did not establish a mean reduction at 3 to 15 ms in 20 recordings
The input epochs, source manifest, and extraction code came from this publication. They enabled a new early-fit test on later individual pulses. We independently retrieved the complete source voltage and verified every reconstructed input.
usesClass difference subtraction kept 99.99% of added train rate signal energy through fit and test in 20 SEEG records with fixed edges
We used its 20 SEEG epoch files, edge indices, channel selection, source manifest, and retrieval and extraction code. We rebuilt every epoch from public source files before the new frequency audit.
usesA spatial filter with one direction reduced a later pulse class difference measure in 19 of 20 SEEG trains
We used its unfiltered pulse and baseline epochs, source channel selection, detected edges, extraction code, fixed metadata, and source hashes. We added a filter learned on the first 24 pulses and scored it only in two blocks from the last 24 pulses.
usesRecent pulse templates reduced error by a mean of 2.51% relative to initial templates at 3 to 15 ms in 20 SEEG records
This is the upstream extraction work identified in the reused source recipe. Its pulse-class and raw-source extraction method informed the new equal-fit-count test on later blocks. The present package credits that indirect code dependency.
usesA scale from 24 early pulses reduced later centered SEEG power more than full subtraction at 3 to 15 ms in 20 stimulation trains
We used its selected unfiltered stimulation and baseline epochs from 20 people, source metadata and hashes, and retrieval and extraction code. These files enabled a new early calibration and later pulse test of template scaling.
usesA fixed map gave 8.00% mean error reduction for simulated missing centered SEEG voltage at 3–15 ms in 20 calibrated stimulation trains
We used its source file list, metadata, extraction and retrieval code, and compact arrays for an exact check against freshly downloaded full voltage files. We reused the same 20 people. The new model uses fresh input-only pulse timing and does not use prior scores.