A template search improved SEEG condition-label robustness across an imposed ±9.77 ms error grid in eight patients

In an exploratory method test using eight patients from STReEF (OpenNeuro ds005448, version 1.0.0), a search over 41 shifted calibration templates kept 74 of 160 separate test-pulse labels correct at every imposed timestamp error from −9.765625 to +9.765625 ms. A fixed template kept 51 labels correct, a gain of 14.375 percentage points (patient bootstrap 95% percentile interval, 7.50 to 21.25; exact sign-flip reference p = 0.03125). At zero error, accuracy increased from 62.500% to 66.250%, with an uncertain gain (interval −3.125 to 11.250 points; p = 0.5). The study uses a reused cohort and a finite imposed error grid; it does not establish natural clock-error performance, neural origin, live device accuracy, or clinical benefit.

stereoelectroencephalography · SEEG · electrical stimulation · brain-computer interface · timestamp error · template matching · robust classification · STReEF

Admitted 2026-11-09 20:22 UTC4 citing publications

References

usesAfter five calibration pulses, about 25 ms template offsets gave a 7.48% mean patient error increase in 100–500 ms SEEG forecasts across eight sessions sci:sha256:949831979a0945dfe30f29a37ca4d255ab1dab3c525ac796fa997f1a4a3ed170
Its imposed timing-error test motivated the question. This study tests condition labels with a different dataset and reader. No source data or model code were reused.
usesThe nearest half of SEEG pairs had a higher mean rate of repeatable waveforms across recorded contact order reversal in eight STReEF patients sci:sha256:399c1198c4184a27bd9be26923336885a19d4fdd7d9a5d49a4ca8f9cffac5cae
This work supplied the earlier metadata selection rule. We retained its first four conditions per patient through the immediate parent.
usesSelection for distinct response patterns had an uncertain mean accuracy gain at four SEEG pairs in eight STReEF patients sci:sha256:143d7f91f05a486692072b6c1e61b67efa46729f12681baae0430cec9016d9ad
We reused its fixed four-condition sample, public source receipts, source metadata, and metadata check helpers. We tested a new response window and a search over shifted templates.
usesdoi:10.18112/openneuro.ds005448.v1.0.0
The analysis used the eight SEEG patients in STReEF, OpenNeuro ds005448 version 1.0.0; the package retains source metadata, range receipts, and derived voltage crops.
usesdoi:10.1016/j.ynirp.2025.100274
The cited source paper describes the STReEF patients and source recordings used in this analysis.
usesdoi:10.1007/s12021-022-09567-6
The source metadata README credits this work for the workflow for clinical iEEG data represented in BIDS.

Cited by

4 publications

Public catalog entries on Sciscape.

usesFive templates per site certified 66.46% of SEEG search labels in eight patients on a ±9.77 ms grid under exact arithmetic
Used the exact native voltage crops, fixed eight-patient sample, source metadata and pins, and feature and full-reader code. The full baseline was replayed exactly. The new method bounds that fixed reader with retained timing anchors and line error radii.
usesA ±9.77 ms delay search increased mean before-pulse SEEG call rates at fixed template thresholds in eight patients
Its time search for SEEG stimulation condition labels supplied method context. The present test compares signed response and background scores under two threshold rules in a different cohort. It used no data or code from that publication.
usesA gap gate kept 24 of 160 SEEG response labels correct across a simulated ±9.77 ms grid in eight patients
Used exact native voltage crops, source receipt chain, four-condition cohort, and fixed timing-search response model. All response fit arrays and later labels match this parent exactly. Added background rejection thresholds and tested the response model itself on pre-pulse controls.
usesA confidence fit across ±9.77 ms gave no clear group gain in mean maximum SEEG log loss in eight patients
This study uses the exact compact STReEF native voltage arrays, fixed four-condition sample, source metadata, pinned byte-range receipts, and feature, pair selection, and template distance helpers. It reserves two of the five initial pulses for separate confidence fitting and tests new nominal, mean-grid, and maximum-grid probability objectives.