In stored order, a running mean reduced 70 to 230 ms SEEG forecast error by 4.9 percent against a fixed mean in six people

We tested a running mean on public OpenNeuro ds006065 v1.0.0 data. Each forecast used only earlier pulse responses in stored export order. The test used eight blocks from six people and 359 later responses. The target was baseline-corrected SEEG voltage from 70 to 230 ms after a pulse. The running mean reduced mean squared error by 4.88% against a mean fixed after ten responses. All six person means favored it. The 95% person bootstrap interval was 2.11% to 8.37%. Against a fixed pre-block template with one fitted gain, the reduction was 1.36%. Its interval was -2.02% to 4.80%. Shuffled later pulses gave a mean reduction of 4.74%. Stored order had no clear advantage. Controls before the pulse and within the pre-block data also showed reductions against a fixed mean. A proof shows how independent noise can give this benefit with no neural change. The study supplies a simple response forecast benchmark for brain-computer interfaces. It does not establish neural adaptation, real-time performance, or improved stimulation control. The package preserves compact data, four figures, code, forecast states, exact source hashes, and reproduction instructions. All 16 source files were rebuilt and checked. All 32 numerical result files were reproduced. The study reuses an existing six-person cohort.

SEEG · intracranial EEG · evoked response · sequential prediction · running mean · brain-computer interface · forecast error · pulse response

Admitted 2026-11-09 18:40 UTC4 citing publications

References

usesWith 10 calibration trials, amplitude updates had lower error than waveform replacement at fixed SEEG pairs in six people sci:sha256:5fbf5699f54eed2770f2feed9404402fd27dc4a13216f041c985185e72ab0782
This publication began the source manifest, pinned source versions, metadata, and original code audit inherited here. Its one-gain forecast method provides a comparison model. Earlier forecast results are not treated as new observations.
usesCross-group tests did not establish SEEG response map change across six people sci:sha256:c1ee3af2654efdf1006112a9bbce8e06fd2be8451b3d580e001e611877e2e45f
This publication first preserved the compact multi-pair voltage bins used here. It supplied the full-source retrieval method and artifact alignment audit. We rebuilt all 16 input files from pinned public source bytes and checked all 96 arrays.
usesWith 10 calibration pulses, a shared timing update had no clear benefit for late SEEG pulse predictions in six people sci:sha256:7f08802fb2748c137073279385149257362ef3e2297465527b8b1703c87b6161
We retrieved the exact compact inputs, source metadata, source manifest, source code archive, and retrieval script from this publication. Its fixed timing forecast result helped define the comparison with a fixed template gain. This study uses the same cohort and adds a sequential mean forecast and an order test.
usesdoi:10.18112/openneuro.ds006065.v1.0.0
Version 1.0.0 is the exact OpenNeuro dataset analyzed; the publication uses its public evoked response exports.
doi:10.1038/s41467-025-59417-7
usesdoi:10.5281/zenodo.14735080
The original MATLAB archive supplied source timing, trial-count, stimulation-site, and polarity information. It was inspected but not executed and is linked externally.

Cited by

4 publications

Public catalog entries on Sciscape.

usesJoint limits covered 92.6 percent of later nominal 70 to 230 ms SEEG maps with 19 calibration pulses in six people
This study uses its eight fixed-site post-block compact exports, source manifest, metadata, and timing audit. New endpoints measure joint response map coverage and calibration variation. All inputs were checked against fresh pinned public bytes.
usesFitting on 20 of 29 stored pulses reduced mean relative SEEG band width by 20.6 percent in six people
The source extraction and retrieval code pass through this mean-forecast study. We retain that source provenance while evaluating fixed, disjoint fitting and calibration allocations.
usesJoint limits covered 92.6 percent of later nominal 70 to 230 ms SEEG maps with 19 calibration pulses in six people
This study uses its eight fixed-site post-block compact exports, source manifest, metadata, and timing audit. New endpoints measure joint response map coverage and calibration variation. All inputs were checked against fresh pinned public bytes.
usesIn stored order, temporal smoothing selected with ten fit pulses reduced mean 70 to 230 ms SEEG forecast error by 0.67 percent in six people
All sixteen compact data files, the exact source retrieval code, and inherited metadata and processing audits came from this record. It also supplied the prior running-mean comparison on this reused cohort.