Mean SEEG forecast gains fell from 14.1 to 5.6 percent when whole shafts were excluded in five sessions

The model used four 20–100 ms SEEG responses to forecast the mean 100–500 ms waveform of four later pulses. The test excluded whole stimulation shafts from fitting and used 169 sites in five patient sessions. The equal patient mean error reduction was 5.6 percent against a training mean. Excluding one site gave 14.1 percent. A later comparison found lower error from a local three-pulse late mean in four of five patients. This local mean was available before the fourth input pulse. Fits used public OpenNeuro ds004457 version 1.0.2 data. The result does not show an online BCI benefit. Five figures, source code, pinned voltage bins, source hashes, saved fits, and raw retrieval instructions permit inspection and reproduction.

stereoelectroencephalography · SEEG · electrical stimulation · waveform forecasting · brain-computer interfaces · ridge regression

Admitted 2026-11-09 19:32 UTC2 citing publications

References

usesEarly RMS did not lower mean late SEEG forecast error in fixed linear models with five calibration pulses per site in eight patient sessions sci:sha256:0a1332bac59dbb1f0933a47b90b106e503aad6a2bade7e7142173f46e206deb5
Its early-versus-late response-size test helped define the question and the need for local comparison rules. Its dataset and outcome differ from this test.
usesA learned time basis did not show lower mean one-pulse SEEG forecast error at 94 new sites in five patient sessions sci:sha256:7aa30d9799432d4754152e931d764d9f390ea9b971b9c6e176501180aabbeebf
Its new-site waveform forecast test defined prior scope. The present question tests early-to-late transfer with whole shafts excluded instead of full-window compression.
usesOne waveform learned from two pulses per site missed repeated site differences in five SEEG sessions sci:sha256:d08ca03b09b85e16bff0cce29a0e32d65988740ca9cc1eb935f43e2d66c2eadd
This publication defined the cohort and first-four/final-four pulse rules inherited through the input package. The new test keeps those rules.
usesMean estimated SEEG site map alignment was 0.28 for 20 to 100 ms and 100 to 500 ms in five sessions under stable means and independent group errors sci:sha256:3da61344038cc338985933849b56f820fd6f4dd48b236c6f72c5aa873d6fbe28
This publication supplied the fixed voltage bins, source metadata, selection code, and raw retrieval recipe. They made the shaft forecast test possible.
usesdoi:10.18112/openneuro.ds004457.v1.0.2
The analysis uses the pinned OpenNeuro ds004457 version 1.0.2 source data; its recorded license is CC0.
doi:10.1523/JNEUROSCI.1325-22.2023

Cited by

2 publications

Public catalog entries on Sciscape.

usesFixed distance weights gave 1.6% higher mean relative SEEG power error at excluded shafts in eight patients
Used this study to choose whole-shaft exclusion and a local target-site reference. Its early-to-late waveform forecasts differ from the present relative-power endpoint.
usesPre-pulse ridge updates had lower mean relative forecast error than early updates in five SEEG replays with earlier fits from other shafts
The exact binned data, source metadata, raw manifest, and licensed readout and extraction code came from this publication. Its shaft-exclusion result motivated the present fit restriction.