Separate pair bounds covered 71.5 percent of individual SEEG responses and 17.0 percent of whole pulses at 20 to 500 ms in eight sessions

We tested whether separate error bounds cover all recorded outputs of one stimulation pulse. The public SEEG data were from OpenNeuro ds004696 version 1.0.1. The test included eight patient sessions, 86 sites, and 886 selected bipolar recording pairs. Two pulses fit each forecast. Three pulses set bounds. Five later pulses tested them. At 20 to 500 ms, separate bounds covered 71.5 percent of individual responses and 17.0 percent of whole pulses. The equal-patient gap was 54.5 percentage points. A common bound based on the largest normalized calibration error covered 72.2 percent of whole pulses. Its geometric mean RMS radius was 1.78 times the separate radius. The same coverage gap occurred before stimulation. The standard 75 percent rank claim needs conditional exchangeability of complete pulses. The source selection and time order do not establish that assumption. Code, compact data, pulse decisions, figures, source hashes, and reproduction instructions are included.

split conformal prediction · coverage · stereoelectroencephalography · SEEG · evoked response · waveform error · brain interface · uncertainty quantification

Admitted 2026-11-09 19:06 UTC4 citing publications

References

usesA five-pulse check selected more repeatable bipolar SEEG pairs within eight epilepsy patient sessions sci:sha256:b5b7963746f5c6d2a2a8f7844ec590732a4eb152bc7a70979b3bc95c3e27820f
Uses the original compact extraction and its source selection method through the parent package. The public source products are the basis for the waveform Gram matrices.
usesJoint score bounds gave 96.3% mean trial coverage and 34.5% mean output yield in eight later SEEG tests with known simulated loss of one contact sci:sha256:4a1c02b9cd875b5cc70d2dbb57771096a5247dbab564b70b51b8fea2e6e55e83
Uses the prior joint score study to state the scope of this contribution. That study tested decoder scores after simulated contact loss in a different dataset. This study tests evoked waveform forecasts at complete stimulation pulses.
usesLater pulse variation accounted for 72.1 percent of normalized 20–500 ms SEEG error for a scaled five-pulse forecast in eight patient sessions sci:sha256:c76a9848dd987107c345f703e31ff3d9949bd3aa3139c35402abe9f6e2470688
Uses exact compact waveform products, pulse maps, source hashes, and the prior source audit for all eight sessions. These inputs support the new analysis of individual and whole-pulse RMS error coverage.
usesSharing early variance estimates reduced later 20 to 500 ms SEEG log size interval score at 53 sites in eight patient sessions sci:sha256:d9ff70ef2fa006c47994b83a215e2b9044aa37261d4896b4540795ed9a648d75
Uses the prior study as related work on coverage across selected SEEG outputs. This study tests RMS waveform errors in a larger inherited set. Different outcomes and selection prevent a direct performance comparison.
usesdoi:10.18112/openneuro.ds004696.v1.0.1
The analysis uses public processed SEEG data from this exact OpenNeuro release.
doi:10.1523/JNEUROSCI.2201-22.2023

Cited by

4 publications

Public catalog entries on Sciscape.

usesFour-pulse fitting reduced error but did not show smaller joint envelopes for 20–500 ms SEEG in eight sessions with five early pulses
This publication supplied the complete-pulse split-rank construction. We changed its fitting and calibration allocation to compare five early pulses at an 80 percent target.
usesJoint limits covered 92.6 percent of later nominal 70 to 230 ms SEEG maps with 19 calibration pulses in six people
This prior result motivated the distinction between separate and whole-pulse bounds. It used three calibration pulses in a different SEEG dataset. This study tests 19 calibration pulses for pair-by-time maps and supplies a 95 percent calibration-budget audit.
usesJoint limits covered 92.6 percent of later nominal 70 to 230 ms SEEG maps with 19 calibration pulses in six people
This prior result motivated the distinction between separate and whole-pulse bounds. It used three calibration pulses in a different SEEG dataset. This study tests 19 calibration pulses for pair-by-time maps and supplies a 95 percent calibration-budget audit.
usesA geometric median of five early pulses gave an uncertain 0.4 percent decrease in later SEEG prediction error at 20 to 500 ms in eight patients
This study uses its exact eight-session Gram inputs, pulse row maps, metadata, source audits, and audit code. It tests geometric median forecasts on later pulses instead of pair bounds.