A learned time basis did not show lower mean one-pulse SEEG forecast error at 94 new sites in five patient sessions

A time basis learned from earlier stimulation sites did not show lower mean forecast error than a fixed time basis family. The test used public SEEG data from five patient sessions in OpenNeuro ds004457 version 1.0.2. The models used 356 pulses from 89 earlier sites. Each forecast used one eligible pulse at one of 94 new sites. It predicted voltage from 20 to 500 ms after four later pulses. All fit pulses came before all new-site input pulses. The mean patient error reduction was -0.22 percent. The descriptive 95 percent patient bootstrap interval was -1.78 to +1.71 percent. The result does not prove equivalence. The fixed family selected full rank or zero output in all five sessions. The package supplies a benchmark, pulse roles, input data, fitted models, forecasts, three figures, and source code. A fresh analysis matched all saved model arrays exactly. Raw source reconstruction matched all input arrays exactly. The study tests offline voltage forecasts. It does not test behavior or clinical benefit.

stereoelectroencephalography · electrical stimulation · forecasting · time basis · singular value decomposition · brain-computer interface · epilepsy · signal prediction

Admitted 2026-11-09 18:16 UTC7 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
Used the signal and noise scale as a local comparison rule. Applied the rule to four new-site input pulses. Did not use the prior five-pulse data or results.
usesPre-pulse scaling gave no clear gain over half-size one-pulse forecasts of centered SEEG at 20–500 ms in eight sessions sci:sha256:c1a6a3c0f5a8d96f92be357f0ce364aa66ce21c9528c03993cfa6fdfa20f0ec0
Used the one-pulse question, the half-size first-pulse control, and first-input energy weights to set this new time-separated site-transfer test. Did not use the prior data or results.
usesA fixed covariance rule gave higher SEEG site accuracy than a diagonal rule in offline tests from five patient sessions sci:sha256:1407cd9ac712a4bcd551a3738cc90c3fab309b871572cbd6fc644b41555f7886
Used the pinned SEEG voltage bins, event tables, BIDS metadata, contact rules, loader, and raw extraction code. The new forecast analysis also rebuilt these bins from all pinned raw source files.
usesdoi:10.18112/openneuro.ds004457.v1.0.2
The analysis used the pinned OpenNeuro ds004457 version 1.0.2 source; its dataset metadata records CC0 and this version-specific DOI.
doi:10.1523/JNEUROSCI.1325-22.2023

Cited by

7 publications

Public catalog entries on Sciscape.

usesMaps selected for 20 to 100 ms voltage error used fewer values and had lower mean site accuracy with a fixed covariance rule in five offline SEEG sessions
Its calibration-only basis test informed the rank and scale grid and the strong full-map comparator. The new analysis fits a site-by-joined-feature map and tests later pulses at known sites. It does not test transfer to new sites.
usesA matrix forecast with training on other shafts raised mean relative error after one accepted pulse in five SEEG recordings at 20–500 ms
Its learned time basis forecast on the same cohort informed the question about full covariance filtering. Our test uses noise covariance, primary voltage loss, and donors on other stimulation shafts. The site sample and methods differ.
usesCovariance from earlier sites did not show a clear 20–500 ms SEEG forecast gain over scalar transfer after two pulses in five session tests
This report supplied the earlier-site transfer question and the local noise-scale comparison. Its fitted models and forecast values were not used.
usesMean SEEG forecast gains fell from 14.1 to 5.6 percent when whole shafts were excluded in five sessions
Its new-site waveform forecast test defined prior scope. The present question tests early-to-late transfer with whole shafts excluded instead of full-window compression.
usesFour local pulses gave lower 20–100 ms SEEG forecast error than one pulse with disjoint-site projection in all five tested patients
This report was read after the new primary scores for comparison of scope. It tests earlier learned time bases with selected rank and gain. The present study tests one donor direction without those choices. It did not set the fixed plan.
usesOne waveform learned from two pulses per site missed repeated site differences in five SEEG sessions
This report defines the prior learned-time-basis forecast test at new sites. It helped define the distinct scope of this test: compression of repeated differences at the same stimulation sites.
usesA fixed one-pair model raised mean relative error by 12.4% for later five-pulse SEEG forecasts in seven people
The earlier time-basis test informed the site-transfer question. This study tests a spatial map across recording pairs in another public SEEG dataset. It uses no data from the time-basis study.