One shared temporal noise matrix gave lower mean SEEG site accuracy than full covariance with two fit pulses per site in offline tests of five sessions

We tested whether recording pairs can share one temporal noise matrix in a reader of SEEG responses to stimulation. We used OpenNeuro ds004457 version 1.0.2 and a fixed cohort of five sessions. The first two eligible pulses per known site fitted the model. The final four tested it. The signal window was 20 to 100 ms after each pulse. All methods used the same site means and feature variances. Across 183 sites and 732 test pulses, mean accuracy was 32.45% with sharing and 35.00% with full covariance. Each patient had equal weight. The observed difference was -2.55 percentage points. The descriptive patient bootstrap interval was -5.51 to 0.00 points. The exact sign-flip value was 0.25. This small sample does not establish a population cost or equal performance. Sharing also scored below separate temporal matrices in all four sessions with multiple recording pairs. Shared metrics stored 88.27% fewer numeric values. Site means plus metric stored 40.53% fewer values. All methods can compile to linear readers of the same numeric size. The benchmark is retrospective within sessions. We checked 110 raw source files, rebuilt all inputs exactly, and repeated the analysis. The package preserves three figures, code, inputs, predictions, fitted models, source hashes, and reproduction instructions.

stereoelectroencephalography · brain stimulation · site identification · covariance · temporal correlation · brain-computer interface · linear discriminant analysis · offline benchmark

Admitted 2026-11-09 17:56 UTC4 citing publications

References

usesA joint SEEG shape decoder did not show a clear mean gain over one selected pair in four patient sessions sci:sha256:81f77d4f52365d5b14fed7d3d6db7b430a2cf478b9a6a6062540c4e90bbd8e7c
The cohort reader, contact rules, raw retrieval, and raw extraction derive from this publication through the cited input publication. These methods supplied a fixed source selection and a checked route from raw MEF recordings to 5 ms voltage bins.
usesA fixed covariance rule gave higher SEEG site accuracy than a diagonal rule in offline tests from five patient sessions sci:sha256:1407cd9ac712a4bcd551a3738cc90c3fab309b871572cbd6fc644b41555f7886
This publication supplied the fixed five-session cohort, voltage inputs, and full covariance rule. We used its inputs as a reference for a fresh raw replay. Its four-pulse comparator counts matched our counts in all 30 checks. The new test compares a shared temporal matrix with fixed methods using two fit pulses per site. These are reused patients and pulses.
usesdoi:10.18112/openneuro.ds004457.v1.0.2
Source recordings and metadata for the five-session SEEG benchmark.
doi:10.1523/JNEUROSCI.1325-22.2023

Cited by

4 publications

Public catalog entries on Sciscape.

usesSeparate state sums preserved SEEG site decisions during known invertible reference switches in five offline sessions
This work uses the contact means, event tables, cohort rules, and raw-data code from this package. This run repeated raw extraction and checked every input array and event table against the cited archive.
usesBefore-pulse covariance gave both higher and lower SEEG site accuracy than response covariance with two fit pulses per site in five offline sessions
Use its fixed cohort, raw reader, voltage bins, and event rules. Reproduce its two- and four-pulse response counts before comparing before-pulse covariance.
usesA check of two scores gave 27 false guarantees for 10 µV contact errors in 20–30 ms bins in five offline SEEG readers
This publication supplied the fixed five-patient cohort, source contact selection, extraction code, and full covariance reader. We rebuilt its inputs from raw files and matched all 732 nominal predictions. We then tested the fixed readers under a new contact error model.
usesEight learned shapes increased mean relative SEEG prediction error with two fit pulses in five offline sessions at 20–500 ms
Supplied an independent raw reconstruction record for the same archived bins. This record is preserved with its publication metadata.