Neighboring stimulation pairs reduced offline SEEG prediction error in five patients, but site order confounded spatial transfer

We tested whether two neighboring stimulation pairs could predict voltage at a held-out pair. The test used public SEEG data from OpenNeuro ds004457 version 1.0.2. It used 70 targets in five patients, 6 mA stimulation, two site folds, four calibration pulses per donor, and bipolar voltage from 30 to 500 ms. Neighbor predictions gave a mean 27.8% reduction in error against the exact expectation for matched random pairs. All five patient scores were positive. The two-sided patient sign-flip p value was 0.0625. Only 32 of 70 predictions had lower error than zero. In a later timing check, the spatial and time-nearest donor pairs were the same at 67 targets. The time predictor gave a mean 27.9% reduction. Only one target had both neighbor calibrations before its test pulses. Thus these data cannot separate spatial proximity from acquisition time or establish prospective calibration savings. The package preserves all source-derived inputs, predictions, score tables, code, three figures, source hashes, and a raw retrieval recipe. All input arrays were independently rebuilt from verified raw MEF files and matched exactly.

stimulation response prediction · SEEG · intracranial EEG · spatial transfer · acquisition time · voltage prediction

Admitted 2026-11-09 16:36 UTC5 citing publications

References

usesSEEG response shapes at 15–100 ms gave higher site identification accuracy than shapes at 100–500 ms within each of five patient sessions sci:sha256:aa5b71d76d214f861e99964287f9da01f3f9fc6c707a4dcfe24807283918ab40
We reused its five fixed recording pairs, trial signals, event metadata, source selection rules, raw retrieval code, and extraction code. We independently rebuilt all input arrays and used a new held-site voltage prediction test.
usesdoi:10.18112/openneuro.ds004457.v1.0.2
The analysis used OpenNeuro ds004457 version 1.0.2; its pinned source paths and hashes are preserved.
doi:10.1523/JNEUROSCI.1325-22.2023

Cited by

5 publications

Public catalog entries on Sciscape.

usesPulse time alone gave 74.8% mean site accuracy in five offline SEEG tests with shared calibration and test blocks
Its finding about site order in a held-site voltage prediction task informed this schedule audit. Our new task predicts labels at calibrated sites. We used none of its target predictions as data.
usesA distance model gave 1.1% higher mean relative 20–100 ms SEEG response error on whole held-out shafts in four patients
Its adjacent-site prediction result and acquisition-time confound motivated a different test that excludes whole stimulation shafts and compares anatomical distance with acquisition time. The present analysis uses the same source cohort and does not refute the earlier neighbor result.
usesPulse time alone gave 74.8% mean site accuracy in five offline SEEG tests with shared calibration and test blocks
Its finding about site order in a held-site voltage prediction task informed this schedule audit. Our new task predicts labels at calibrated sites. We used none of its target predictions as data.
usesA fixed average of local and earlier neighbor pulses increased 20–100 ms SEEG forecast error in three of five sessions
Its spatial-transfer result and acquisition-time audit motivated the requirement that donor responses end before the target first pulse. We tested a different forecast against local calibration on the same source cohort.
usesSEEG data after a pulse gave higher site accuracy than data before a pulse for neighboring stimulation pairs in five offline sessions
The earlier site-order confound informed the calibration-time and candidate-block separation audits. These audits limit the new local discrimination result to an offline scope.