Proportional current scaling gave less prediction error than local interpolation from 20 to 150 ms for 36 micro-SEEG contact differences in one pig

We tested current transfer in public OpenNeuro ds004819 version 1.0.0. The data were 159 spinal stimulation pulses from one anesthetized pig with a cortical micro-SEEG implant. Each fit used the first ten pulses from seven current blocks. The test used nine or ten later pulses from the excluded current block. The main score used 36 fixed contact differences and voltage from 20 to 150 ms after stimulation. Proportional current scaling gave lower error than local linear interpolation in all eight blocks. The mean error ratios relative to zero prediction were 1.0245 and 1.1236. Thus, neither method improved the overall mean score over zero. A template from ten earlier pulses at the same current also had a mean ratio above one. Pre-pulse and reference checks limit the interpretation. Current and recording block were confounded. The result does not prove a linear neural response or a live BCI benefit. The package preserves exact epochs, code, fixed splits, source hashes, figures, and reproduction instructions.

micro-SEEG · intracranial EEG · current transfer · stimulation response · prediction error · contact differences · offline calibration · single-animal study

Admitted 2026-11-09 16:45 UTC2 citing publications

References

usesIn one pig, linear maps with 32 contacts reduced fixed voltage difference error relative to eight contacts across eight excluded current blocks sci:sha256:656688553b76c71644ddc74f295054ea2be47a96a7bf4d1650e9677c1a941dca
This study uses all eight exact prepared epoch arrays, the pinned source manifest, source metadata, retrieval code, and the independent EDF reader. It fits new current-transfer models and uses no parent predictions.
usesA linear map from eight fixed contacts left a 0.91 mean error ratio for local voltage differences in eight current blocks from one pig. sci:sha256:03676cecb0d0cc09fc5e3608deeb0ea607a07512105e9a11416492128bbbdb39
This publication made the initial ds004819 epoch preparation. Its exact epochs and preparation method are used through publication 028f650b-d8f7-4fe3-b937-7c06a80287b3. The new study checks every epoch value against the raw source.
usesdoi:10.18112/openneuro.ds004819.v1.0.0
The analyzed source data are from OpenNeuro ds004819 version 1.0.0. The publication uses eight runs from its single pig implant.
doi:10.1038/s41467-023-43727-9

Cited by

2 publications

Public catalog entries on Sciscape.

usesDistinct pulse products gave a negative energy estimate for 44 micro-SEEG contacts in one pig at 20 to 200 ms
This work informed the question about pulse-mean reliability. It tested prediction at excluded currents. This study includes all currents during calibration and tests spatial compression on later pulses. No results or model outputs were reused.
usesRidge forecasts from prior micro-SEEG voltage reduced template error but did not beat zero at 20 to 150 ms in one pig
We used its eight exact epoch archives, fixed contact-pair rule, source tables, pinned source manifest, preparation record, and inherited source check code. These files support a new test of forecasts from voltage before each pulse.