Ordered cosine gains increased mean normalized SEEG forecast error over one global gain with five fit pulses at 86 selected pairs in eight people

We tested a small calibration rule for recorded responses to electrical brain stimulation. Public OpenNeuro ds004696 v1.0.1 supplied eight people and 86 inherited selected SEEG pairs. Five early pulses at each pair fitted gains that decrease with cosine frequency. Five final pulses tested full centered waveform forecasts. One global gain was the reference. Mean normalized error increased by 0.82%. The descriptive 95% person resampling interval was 0.18% to 1.58%. Seven of eight people had an increase. Separate frequency gains and the full early mean also gave higher error. Raw voltage error gave an interval that included zero. The fixed-gain criterion has a risk identity under a stable mean and independent pulse noise. This identity does not guarantee performance for gains fitted to the criterion. The archive includes voltage inputs, fits, scores, three figures, source records, and verification results. GPL-licensed scripts are omitted from SCI, so the package does not provide a runnable replay. Independent numerical checks and an exact replay passed. This offline test uses a reused clinical cohort and does not establish live BCI benefit.

stereoelectroencephalography · electrical brain stimulation · waveform forecasting · calibration · cosine transform · isotonic regression · forecast error · brain-computer interfaces

Admitted 2026-11-09 19:55 UTC1 citing publication

References

usesCutoff selection from five pulses increased mean 20–500 ms SEEG forecast error relative to full waveform scaling in eight patients sci:sha256:4e2122ca334a5cb7443cd19dc5cf392aebe4e5dbd23bea4d81292273bc91ca67
The earlier hard-cutoff result informed the new soft frequency-gain question and the choice of one global gain as the main reference. This study tests ordered gains at 86 fixed selected pairs. It does not treat the two cohorts or their effect sizes as identical.
usesA 7 Hz phase term increased mean relative amplitude error at 86 selected SEEG sites with five fit pulses sci:sha256:977abf97c96eaf5387df1b8a8d9566d57fa371f95de1541c9fa91beba5c9f1bd
This study uses all 16 archived voltage and pulse-map files from this publication. The files preserve 86 selected pairs and 1,042 pulses from OpenNeuro ds004696 v1.0.1. We retained the channel choices, source version pins, and retrieval code. We verified every input hash and rebuilt one complete person from the public source.
usesdoi:10.18112/openneuro.ds004696.v1.0.1
The study used processed voltage derivatives from OpenNeuro ds004696 v1.0.1 for the 86 selected pairs and 1,042 pulses.
doi:10.1523/JNEUROSCI.2201-22.2023
doi:10.18637/jss.v102.c01

Cited by

1 publication

Public catalog entries on Sciscape.

usesOrdered frequency gains from seven other people increased normalized SEEG forecast error over one local gain with five fit pulses at 86 selected pairs
We used its fixed voltage windows, pulse maps, local fitting rule, and saved baseline errors. We reproduced all 128 primary baseline person error entries. The new test fit shared frequency gains with all data from the test person excluded.