Ten-pulse means had higher response label AUCs than first pulses for a fixed 10–90 ms absolute score in two SEEG recordings

This study tested fixed pulse budgets in two public SEEG recordings from OpenNeuro ds004774 version 1.0.0. It used 630 pulses at 63 surface stimulation sites and six depth readout contacts per recording. The score was the largest absolute mean voltage from 10 to 90 ms, divided by baseline standard deviation. Expert codes 1 or 2 were response labels. Each recording and expert was tested separately. The first-pulse AUCs were 0.781, 0.638, and 0.716. Ten-pulse AUCs were 0.905, 0.767, and 0.892. The groups were UMCU21/DvB, UMCU21/SB, and UMCU25/DvB. Paired site bootstrap bounds for the one-minus-ten changes were below zero in all three groups. First-five AUCs were 0.879, 0.749, and 0.852. Their losses from ten pulses were 0.027, 0.018, and 0.040. Each five-minus-ten interval included zero. The data do not establish an equivalent five-pulse budget. All 252 five-pulse index subsets were tested. Their AUC ranges were 0.833–0.930, 0.680–0.797, and 0.772–0.910. Pulse order is therefore part of this benchmark. These finite ranges are not confidence intervals or a method to select future pulses. The package includes native epochs, source labels and metadata, code, figures, score tables, bootstrap draws, and retrieval instructions. A fresh rebuild checked all 630 pinned source byte ranges. All 5,034,960 depth voltage values matched. Independent label, score, and AUC checks passed. A second run reproduced every result table and bootstrap value. The evidence applies to these recordings, experts, and this offline score. It does not set a clinical pulse budget or validate online brain control.

SEEG · evoked responses · pulse budget · expert labels · AUC · brain-computer interface · intracranial stimulation

Admitted 2026-11-09 16:42 UTC12 citing publications

References

usesAbsolute 10–90 ms voltage scores had higher response label AUCs than negative scores for means of ten pulses in two SEEG recordings sci:sha256:15208475885140e6fc67bf47f35cebce3a7c9de326bb953b9d58b7bbd4d71f41
This study uses the native SEEG epochs, expert labels, fixed complete-frame selection, source metadata, and raw range provenance from this publication. Its ten-pulse scores were reproduced exactly. New tests change the pulse budget and preserve pulse order.
usesFive-pulse SEEG means had higher 20–100 ms shape similarity than before-pulse controls in three recordings with an odd–even split sci:sha256:6070351f0242e8b87bbcec400d56f9e92e15e2e6e2c1fabf38d81db2c9deb53b
This study reuses the MIT raw range retrieval recipe and source label audit through the cited epoch package. The recipe enabled a fresh source rebuild of all 630 pulses.
usesdoi:10.18112/openneuro.ds004774.v1.0.0
The analyzed SEEG recordings, stimulation events, channel metadata, and expert labels are from OpenNeuro ds004774 version 1.0.0.
doi:10.1016/j.jneumeth.2025.110389

Cited by

12 publications

Public catalog entries on Sciscape.

usesA 20 percent trimmed mean raised expert label AUC from 0.892 to 0.923 for ten pulses and a 10 to 90 ms score with each method's baseline SD in UMCU25 SEEG data
We used the compact native voltage arrays, source metadata, expert labels, MIT retrieval code, label and event audits, and known mean benchmarks. We tested fixed robust averages on the same 63 sites. We also rebuilt the arrays from fresh pinned raw ranges and checked 18 inherited mean AUC estimates.
usesA 200 ms baseline gave 53 to 61 percent control calls in two SEEG recordings at a nominal 5 percent limit
Its compact array recipe and label audit are the original source of the inputs and retrieval code inherited through publication 5967273a-78e7-469f-a706-2bbc7dfcdca9.
usesPulse variance correction gave AUC gains below 0.007 with intervals that included zero for the first five pulses in two SEEG recordings
Its package supplied native SEEG epochs, source labels and metadata, pinned raw-range retrieval code, site selection records, and absolute peak scores. These inputs enabled the new paired comparison of mean energy and pulse-variance-corrected energy. A fresh source check verified all raw ranges and voltage values.
Peak confirmation had lower response label AUC point estimates than a peak from ten pulses at 10–90 ms in two SEEG recordings
[Legacy relation: reproduces] The 378 primary all-ten-pulse comparison scores match the parent scores within 3.6e-15. This relation applies to that baseline score, not to every parent analysis.
usesPeak confirmation had lower response label AUC point estimates than a peak from ten pulses at 10–90 ms in two SEEG recordings
We reused the selected raw SEEG epochs, expert labels, source metadata, event context, and range retrieval code. This fixed the data selection for the new peak confirmation test.
usesSEEG waveform repeatability gave AUC 0.492 for agreed versus disputed responses in the UMCU21 record
This study used the native UMCU21 pulse epochs, source depth labels, metadata, event selection, and raw-range hashes. It tested expert agreement and waveform repeatability with these preserved inputs.
usesRemoval of shared voltage reduced site identification accuracy for a fixed method in two SEEG recordings with six contacts
This publication defined the original ten-pulse trial selection. We kept its sites and pulses through the direct parent package.
Peak confirmation had lower response label AUC point estimates than a peak from ten pulses at 10–90 ms in two SEEG recordings
[Legacy relation: reproduces] The 378 primary all-ten-pulse comparison scores match the parent scores within 3.6e-15. This relation applies to that baseline score, not to every parent analysis.
usesPeak confirmation had lower response label AUC point estimates than a peak from ten pulses at 10–90 ms in two SEEG recordings
We reused the selected raw SEEG epochs, expert labels, source metadata, event context, and range retrieval code. This fixed the data selection for the new peak confirmation test.
usesPulse sign tests across six contacts gave 9% call rates before stimulation in two SEEG recordings
We used the fixed two-recording SEEG pulse arrays, source labels, metadata, event selection, range hashes, and retrieval code. These inputs enabled the exact time and contact search tests without a new data selection.
usesTen-pulse response-label AUC exceeded mean five-pulse subset AUC at 10 to 90 ms in UMCU21 SEEG with agreed binary labels fixed
This study uses the UMCU21 pulse epochs, source expert labels, event maps, and fixed voltage score from this package. It changes the analysis to matched expert cells and exact label bounds for pulse-count comparisons.
usesRemoving self-products raised observed five-pulse expert label AUCs by less than 0.007 in two SEEG recordings
We reused its native voltage epochs, labels, contact inventory, retrieval code, exact byte ranges, and site gap records.
usesCross-pulse energy changed 10–90 ms response label AUC by less than 0.007 for the first five pulses in two SEEG recordings
We used this publication's exact native epochs, source labels, event selection, source manifests, and retrieval code. We adapted its paired site bootstrap method. Independent calculations also reproduced all 36 prior peak AUC values for the full cohort.
usesTen-pulse 10–90 ms SEEG score thresholds fitted at a 10% source error limit gave 4.3% and 12.6% target false response rates in two recordings
We reuse the fixed 63-site selection, voltage epochs, expert labels, and pulse score rule. We compare all 3396 labeled scores with its saved table. The new analysis tests threshold transfer and includes source threshold fitting in site bootstrap draws.