An offline two-pulse site rank did not show lower 20–500 ms SEEG forecast error at a fixed budget in eight sessions

We tested whether disagreement between two early pulses can assign a fixed SEEG calibration budget. The public sample was OpenNeuro ds004696 version 1.0.1, with eight patient sessions, 86 stimulation sites, and 886 stimulation-to-recording pairs. The rule gave three extra pulses to floor(half the sites) within each patient. It forecast each later pulse with the mean of two or five calibration pulses. The score used centered bipolar voltage from 20 to 500 ms and fixed scout-energy weights. The simulated budget was 298 calibration pulses, versus 430 for all-five calibration. Against the exact random-allocation expectation at the same budget, the mean patient error reduction was 0.72%. The 95% patient bootstrap interval was -1.17% to 2.38%; the exact patient sign-flip p value was 0.477. Six patients had lower error and two had higher error. A planned noise-scaled forecast control had mean gain -0.21%. The data did not establish an allocation benefit or equal performance. We prove the optimal known-noise rank under stable responses, independent pulse noise, and fixed weights. That proof does not establish that two scouts estimate a useful rank. The package includes exact input products, all decisions, result tables, three figures, code, source hashes, and reproduction instructions. Two complete public source derivatives were rebuilt and matched the archive exactly. This is an offline test. It does not establish a prospective schedule, cross-session transfer, or clinical benefit.

stereoelectroencephalography · SEEG · electrical stimulation · calibration · forecast error · brain-computer interfaces

Admitted 2026-11-09 17:43 UTC10 citing publications

References

usesFive-pulse noise scaling reduced later single-pulse error for 20–500 ms centered bipolar SEEG in eight patient sessions sci:sha256:fbc06ab469d48eac7a6f54dd97cd24ccd9c7c5ca4b0db4d64340536b3e81428a
We used all eight corrected exact trial-product files and their chronological pulse sets. We also used its calibration-only noise-scaling rule as a planned descriptive control.
usesA five-pulse check selected more repeatable bipolar SEEG pairs within eight epilepsy patient sessions sci:sha256:b5b7963746f5c6d2a2a8f7844ec590732a4eb152bc7a70979b3bc95c3e27820f
We reused the public derivative retrieval and extraction code, source metadata, anatomical pair rules, and same-block reference cancellation analysis. This enabled source rebuilds and the compact fixed-budget benchmark.
usesdoi:10.18112/openneuro.ds004696.v1.0.1
Source data: OpenNeuro ds004696, version 1.0.1.
doi:10.1523/JNEUROSCI.2201-22.2023
doi:10.1371/journal.pcbi.1011105

Cited by

10 publications

Public catalog entries on Sciscape.

usesScaled five-pulse forecasts had more skill for centered bipolar SEEG at 20–100 ms than 100–500 ms in eight patient sessions
We used its 25 exact input files for eight patients, 86 sites, and 886 pairs. Its products preserve the corrected chronological pulse selection. We added a fixed early-to-late skill contrast against zero voltage.
usesAn offline two-pulse rule reduced 20–500 ms SEEG map error in two of five people at an equal pulse count
This study used the two-pulse allocation question and the exact random-placement comparison. It tested a bounded integer rule against uniform counts in a different public SEEG cohort.
usesA two-pulse site rank did not show lower 20–100 ms SEEG forecast error than a spread budget in five sessions
We use its two-scout disagreement rank, site-level budget design, and exact random-allocation comparator. The new test uses ds004457 and a two-to-four budget. It adds a same-budget spread reference and the model condition for concentration to beat that reference.
usesAn allocation from two early pulses did not establish lower maximum 20 to 500 ms SEEG mean response error at a fixed budget in eight offline sessions
This prior study of an offline two-pulse rank in the same dataset motivated the question. The present test uses an exact maximum noise model, the largest measured site error, and separate finite optimization checks. The prior result is method context, not independent patient evidence.
usesA sequential pulse quota gave no clear error gain over random allocation in an offline test of 20 to 500 ms SEEG from eight sessions
Its offline site allocation informed the question. We test a rule that acts at the current site without future site waveforms from the test patient.
usesSite block pooling increased three-pulse SEEG coverage and mean bound score at 20–500 ms in eight patients
It supplied the exact centered trial Gram matrices inherited through the cited input archive. Those matrices let us calculate waveform RMS errors for all selected readouts and fixed windows without the large voltage files.
usesThree-pulse RMS bounds gave mean coverage of 72.2 percent for later SEEG pulses and 38.1 percent for five-pulse sets at 20–500 ms in eight sessions
This test uses the unchanged 24 patient files, corrected chronological pulse maps, exact Gram products, source metadata, and retrieval code. The new outcome is fixed forecast bound coverage across readouts and future pulses.
usesA flexible calibration rule reduced 20 to 500 ms SEEG forecast error against concentrated calibration at the same budget in eight recordings
We used its two-scout normalized noise priority, fixed person budget, and concentrated two-or-five calibration rule as the comparator. We added flexible marginal-value allocation, a balanced three-or-four comparator, and an exact count-versus-assignment decomposition.
usesSeven calibration pulses gave a 3.33% mean person error reduction over five for centered 20 to 500 ms SEEG at 86 selected readouts in eight sessions
Its earlier fixed budget allocation result informed the calibration cost question. We hold the count at five or seven for every site and do not claim a new allocation rule.
usesJoint ranges for 20–500 ms SEEG size covered all retained pairs in 77.2 percent of later pulses on average across eight sessions
We used all eight exact SEEG input products, the extraction recipe, the source audit code, and the source metadata. We checked input hashes and rebuilt two source files. These inputs enabled the new test of coverage at all retained recording pairs.