A 20–100 ms energy update gave mixed forecast results in eight reused SEEG sessions with five calibration pulses

We tested whether early voltage energy improves a forecast of later voltage energy after a stimulation pulse. The source was public OpenNeuro ds004696 version 1.0.1. We used 85 selected readouts in eight reused epilepsy sessions. Five pulses at each site set local means. The other seven people supplied the fitted slopes. The final five pulses at each site supplied 425 test pulses. Adding 20–100 ms energy to energy before the pulse reduced mean error across people for 100–500 ms log energy by 0.56%. Error decreased in five people and increased in three. The mean change in squared log error was -0.00244. Its descriptive 95% person interval was -0.01057 to 0.00617. The combined model had more mean error than a constant five-pulse mean. The test does not establish a useful fast feedback readout. The package preserves exact input windows, source pins, code, pulse forecasts, three figures, and replay steps. A fresh public source download matched all 61 pulse windows in one person. This was an offline test of recorded voltage energy. It did not test a live interface or neural causation.

stereoelectroencephalography · electrical stimulation · energy forecasting · brain-computer interface · ridge regression

Admitted 2026-11-09 20:27 UTC2 citing publications

References

usesEarly SEEG increased late forecast error in eight patient sessions with five calibration pulses and ridge strength 1 sci:sha256:84398bf3c7feb1c56467f8999e13d20cee476064678d534e4fc33c2a7dc7d90d
This earlier early-to-late waveform forecast test helped define the scalar energy question. Our target, shared slopes, before-pulse comparison, and selected readout set differ. The people are reused, so this is not independent replication.
usesA fixed model gave no clear 4–30 Hz power forecast gain from input ending 20 ms versus 500 ms before a pulse for targets at 100–500 ms after the pulse in eight SEEG sessions sci:sha256:7def033d65e62e11ff56502733ae76b99a84f7764072fd1b8d669d98536e3b22
This study reported the sub-08 duplicate source event at rows 49 and 74 and tested power inputs before stimulation. We excluded that entire site in advance of our new calculations. Our main input is energy after stimulation.
usesFive calibration pulses gave mean later signed template AUC of 0.939 compared with 0.859 for one pulse in eight SEEG sessions sci:sha256:1d2aa103275ad7400befec4c899740ede8f2eb59a09897bdc3f4d4e190e54418
This work supplied the inherited five-pulse choice of one recording pair per site. The present test applies to those selected readouts and does not repeat contact selection.
usesSelection of time bins from five pulses reduced storage and increased mean later SEEG error at 86 sites in eight reused sessions sci:sha256:aec9768a6512c81c46ebb7344e1874916eb2830d71d902e1f8d3b2def2c35d2a
This publication supplied the exact eight-person pulse windows, original event maps, pinned source records, and range download code. We checked all 16 input file hashes. The new analysis forecasts one energy value and uses slopes from other people.
usesdoi:10.18112/openneuro.ds004696.v1.0.1
Source dataset: OpenNeuro ds004696 version 1.0.1, used for the recorded voltage arrays and metadata.
doi:10.1523/JNEUROSCI.2201-22.2023

Cited by

2 publications

Public catalog entries on Sciscape.

usesEarly energy with slopes from other people reduced late SEEG log energy forecast error by 11.1 percent in eight people with reversed test contact order
This work supplies the other-person log energy forecast design, fixed windows, and ridge penalties. We apply that rule to the fixed STReEF sample and retain its primary penalty of 0.1.
usesA before-pulse energy model gave 1.1 percent more SEEG error forecast loss after five fit pulses at 85 selected sites in eight people
Its donor model design and audit of duplicate pulse times informed our fixed model and exclusion. Our target is individual waveform template error, with a donor intercept comparison.