Response correlations gave no clear loss reduction with five fit pulses and 50% shrink at 83 SEEG sites in eight sessions

We tested joint forecasts of SEEG log RMS after electrical pulses. Public OpenNeuro ds004696 version 1.0.1 supplied eight epilepsy sessions. A fixed contact rule retained 83 sites and one bipolar readout per remote shaft. Five early pulses fitted each model. Five later pulses supplied 415 test events. Background and response correlation models used the same response means and channel scales. Both correlations had 50% shrink toward identity. Response correlations gave 0.0545 lower mean negative log density per shaft than background correlations. The descriptive 95% patient bootstrap interval was -0.0336 to 0.1656. The exact sign-flip value was 0.41406. This test did not show a clear reduction. The diagonal reference had lower mean loss than both models at the primary shrink setting. Stronger shrink removed the response model's disadvantage. These secondary results do not establish independent shafts or an optimal model. Nominal Gaussian coverage was below 90%. A finite-sample calculation shows why five fitted variances can cause excess distance without neural drift. The package preserves compact inputs, all model matrices, event maps, scores, figures, source hashes, and reproduction records. GPL Python analysis and extraction scripts and copied GPL HAPwave MATLAB scripts are omitted. Their modified implementations are not publicly available, so the exact analysis and source extraction cannot be replayed from this SCI package. Complete source rebuilds for 3 participants matched tables and arrays within numerical tolerance. The complete score replay was exact. This analysis reuses a public cohort. It does not test online BCI control or causal covariance changes.

stereoelectroencephalography · electrical stimulation · forecasting · covariance · log RMS · brain-computer interface · epilepsy

Admitted 2026-11-09 19:43 UTC7 citing publications

References

usesFive later pulses did not show higher SEEG size correlation across shafts at 20–500 ms than before stimulation in eight sessions with readouts selected from five early pulses sci:sha256:3e568cd9997feedb9652f1ade1733ba15c832a7040976086282aa09650536328
The separate-shaft size comparison motivated this forecast question. This study uses separate early fit pulses and later test pulses, with fixed response scales and a joint density score. It reuses the same cohort.
usesA fixed spatial noise rule gave no clear 20 to 500 ms AUC gain after five early pulses in eight SEEG sessions sci:sha256:48f93098d9fe4ee439817f49e745f4c586103d8b79d4985cbb51b1bf7dc5d020
The spatial covariance detection study supplied method context. This study tests forecasts of later log RMS sizes. It does not test response detection and does not add patients.
usesSEEG variance after mean removal was higher 20 to 500 ms after pulses than before pulses in eight epilepsy sessions sci:sha256:8ee73a7bddac5f833636033f8f8b842e4487d5f60accef4748f5c9c259e4c1d8
This study uses the compact trial products, contact tables, source metadata, source hashes, and unchanged retrieval code. The new endpoint scores later joint log RMS forecasts fitted with five early pulses. Fresh source rebuilds checked three participants.
usesdoi:10.18112/openneuro.ds004696.v1.0.1
Source dataset: OpenNeuro ds004696 version 1.0.1, with source file hashes recorded in original/provenance/sources.json.
doi:10.1523/JNEUROSCI.2201-22.2023
doi:10.1371/journal.pcbi.1011105

Cited by

7 publications

Public catalog entries on Sciscape.

usesStudent t forecasts reduced later log size loss after five fit pulses at 83 SEEG sites in eight people
This work supplies the exact trial products, contact rules, source metadata, and extraction code used here. Its Gaussian scores provide an independent numerical reference.
usesResponse correlations gave no clear site accuracy gain with five fit pulses per site and four SEEG pairs in eight patients at 50 percent weight
Its short-fit correlation result informed the question and the matched-scale comparison. It used a different SEEG dataset and a forecasting outcome. It supplied no input data or primary accuracy values for this study.
usesA shared scale in a Student t model with four degrees of freedom raised mean 4–30 Hz SEEG joint forecast loss at 100–500 ms after five fit pulses at 82 sites in eight sessions
Its metadata rule selects one eligible pair per remote shaft. Its joint Gaussian comparison supplies context for a new test of dependence through a shared random variance scale, with identical univariate Student t forecasts.
usesResponse correlations gave no clear site accuracy gain with five fit pulses per site and four SEEG pairs in eight patients at 50 percent weight
Its short-fit correlation result informed the question and the matched-scale comparison. It used a different SEEG dataset and a forecasting outcome. It supplied no input data or primary accuracy values for this study.
usesExact normal limits reduced log width and gave 85.7 percent mean coverage for five-pulse batches at 83 SEEG sites in eight people
We used its exact trial products, fixed contact geometry, pulse order, source metadata, and retrieval code. All 70 inherited input files matched its manifest. We rebuilt one full public source file in this run.
usesVariance cap 10 decreased SEEG log-size loss but increased mean RMS error against Gaussian forecasts in eight people
We used its public-source retrieval and extraction code through the direct parent. We newly rebuilt the complete sub-04 source under its pinned hash and compared the compact inputs.
usesF limits covered 89.8 percent of later three-coefficient SEEG responses after five fit pulses at 86 selected sites in eight patients
Its low joint Gaussian coverage motivated the test of uncertainty from five fit pulses. This study uses fixed temporal coefficients. It does not reuse that study's spatial features or fitted models.