Pulse time alone gave 74.8% mean site accuracy in five offline SEEG tests with shared calibration and test blocks

We tested whether pulse recording time could identify stimulation sites without voltage input. We used public SEEG data from OpenNeuro ds004457 version 1.0.2. The fixed offline split had 183 sites in five patients, with four calibration and four test pulses per site. The site of the nearest calibration pulse in time gave 74.8% equal-patient mean accuracy. A fixed post-pulse voltage rule gave 40.6%; a uniform guess gave 2.8%. Of 732 test pulses, 711 shared a source stimulation block with calibration pulses. A secondary rule that used the latest earlier calibration label gave 97.2% mean accuracy. It was correct for all 711 shared-block test pulses. Only five sites had four eligible pulses in each of two separate blocks. This result shows that the schedule can solve much of this split. It does not show what information the voltage rule used or establish performance in a new session. The package preserves three figures, all inputs and predictions, reusable time-control code, source hashes, and reproduction instructions. Raw input arrays and a fresh analysis replay matched exactly. This version corrects a count in the description. It changes no result.

stereoelectroencephalography · electrical brain stimulation · recording schedule · site classification · brain-computer interfaces · offline evaluation

Admitted 2026-11-09 18:26 UTC21 citing publications

References

supersedesPulse time alone gave 74.8% mean site accuracy in five offline SEEG tests with shared calibration and test blocks sci:sha256:63641775031f9633094608ce0556d0dad6d96747830c30b721bfec12d789f479
This version corrects the stated count of planned secondary rules from six to five. The source code, inputs, predictions, and numerical results are the same.
usesA fixed covariance rule gave higher SEEG site accuracy than a diagonal rule in offline tests from five patient sessions sci:sha256:1407cd9ac712a4bcd551a3738cc90c3fab309b871572cbd6fc644b41555f7886
We reused its source-derived voltage bins, recording contacts, site rules, and first-four versus last-four pulse split. We independently reproduced its fixed covariance voltage reference, then added time controls and a source block audit.
usesNeighboring stimulation pairs reduced offline SEEG prediction error in five patients, but site order confounded spatial transfer sci:sha256:33c33a6543262722110a335be73c9a8ad8d047a9d5831868e69c8fc52b59516b
Its finding about site order in a held-site voltage prediction task informed this schedule audit. Our new task predicts labels at calibrated sites. We used none of its target predictions as data.
usesdoi:10.18112/openneuro.ds004457.v1.0.2
This is the exact OpenNeuro version analyzed; its recorded metadata identify the version and CC0 terms.
doi:10.1523/JNEUROSCI.1325-22.2023

Cited by

21 publications

Public catalog entries on Sciscape.

usesFour pulse groups gave SEEG response participation ratios of 1.20 to 2.80 in five sessions under stable means and independent errors with zero mean
Its time-only control showed a limit from blocked site order in this pulse selection. It informed our limit on claims about neural cause and new-session performance.
usesSeparate site noise models gave no clear accuracy gain at 20 to 100 ms with four fit pulses in five offline SEEG sessions
Its pulse-time control and split audit informed the offline scope. We computed the timing count again in the new restricted sample. Only 32 of 536 test windows followed all fit windows. We did not use its predictions or count these patients as independent evidence.
usesCentered median templates raised mean site accuracy with one simulated wrong label per four fit pulses and lowered clean accuracy in five SEEG sessions at 20 to 100 ms
Its timing-only audit informed the schedule limit and the choice to avoid repeating that question. We recomputed that only 36 of 732 test pulses occur after all fitting pulses. The present result remains an offline calibration-label test.
usesCycle checks missed simulated contact errors that changed SEEG site decisions in five offline sessions
Its schedule audit shows that these reused fit and test blocks interleave and that time can identify sites. We use that result to limit the interpretation of absolute reader accuracy and avoid a prospective decoding claim.
usesFirst and last pulse groups gave mixed SEEG energy changes from 20 to 100 ms at 134 sites in five sessions
Its schedule-control result informed the source-block audit. The main test excludes sites whose accepted pulses occurred in more than one source stimulation block.
usesCalibration chose a mean 56 ms endpoint with no lower observed eight-site SEEG accuracy in five offline sessions with input from 20 ms
We adapted the time-only site readers and the site-current block definition for a post-result diagnostic on our eight selected sites and 160 test pulses. The control limits the interpretation of the response endpoint result.
usesMean subtraction before early bin removal gave 7.9% site accuracy with zero late SEEG inputs in five offline patient tests
Its time-only site result and source block audit limit neural and online interpretations of this same offline pulse split.
uses32-bit biases raised mean site accuracy over 4-bit biases in five offline SEEG tests with 4-bit voltage weights at 20–100 ms
Its time control showed a schedule limit in this offline split. We replayed its nearest calibration time rule on the same events. It gave 74.80 percent mean accuracy without voltage input and constrained our interpretation.
usesWeighted site positions reduced squared location error but reduced 10 mm success in five offline SEEG sessions
Its recording-time control informed the schedule diagnostic. We applied a nearest-fit-time rule to the mapped site subset and measured physical distance and 10 mm success.
usesSEEG on other shafts gave 78.7% mean accuracy for two known neighboring stimulation sites in five offline sessions
Its site-order result informed the pulse-time control and the limit on online interpretation. The new two-site time-only control reached 83.8% mean accuracy.
usesAbsolute pair scores kept SEEG site decisions fixed under simulated output sign reversals in five sessions
The time-only result and source-block audit for the same split limit the interpretation of our site accuracies. We use this evidence to restrict our claim to recording-score robustness. We do not reproduce its time-only reader.
usesSEEG at 20–100 ms gave higher remote site accuracy than shifted controls in four of five offline 6 mA sessions
Its source-time audit and short before-pulse controls motivated the matched temporal test. We recomputed a nearest-fitting-pulse time reader on the new retained events and extended voltage controls to shifts up to two seconds.
usesContact normalization kept SEEG site decisions fixed under simulated five percent contact gain changes but reduced accuracy in five offline sessions
Supplied the prior block and pulse-time control that limits interpretation of site classification in this cohort.
Means of four pulses reduced normalized 20–100 ms voltage error by 33.3% relative to zero across blocks at five sites in three SEEG sessions with equal site and patient weights
[Legacy relation: reproduces] All 732 test predictions matched for each of four methods: latest prior calibration label, nearest calibration label, 20–100 ms voltage, and pre-pulse voltage. We also matched the site source block counts for all 183 retained sites. That corrected audit had already identified the five sites with two supported blocks. Our new contribution is their separate-block voltage forecast loss, controls, and failure analysis.
usesTwo four-bit storage rules reduced mean site accuracy in five offline SEEG sessions with four calibration pulses per site at 20 to 100 ms
Its pulse-time control in the same split limits our interpretation. We treat site accuracy as an offline coefficient-storage benchmark, without a claim about neural specificity or new stimulation blocks.
A template gap gave 0.827 mean error AUROC across known sites in five offline SEEG sessions at 20 to 100 ms
[Legacy relation: reproduces] We repeated nearest-time and latest-earlier-time site controls with the first four eligible pulses and final four test pulses. The schedule control limits interpretation of the voltage reader.
usesTraining with nine window shifts raised mean site accuracy under uniform −20 to 20 ms shifts but reduced accuracy at zero in five SEEG sessions
Its voltage-free time control and site-block audit limit interpretation of our inherited offline split. We cite its measured schedule result and do not claim that site classification isolates biological site information.
usesA fixed pre-pulse adjustment gave 70.8% accuracy versus 71.4% without adjustment for neighboring sites at 20–100 ms in five offline SEEG sessions
We used its recording-time control to explain the schedule limit. We did not transfer its numeric scores to our two-site test.
usesPulse time gave 80.8 percent site accuracy after recorded contact order reversal in offline tests of 12 sites in each of eight SEEG patients
Its time-only site tests in a different public SEEG dataset supplied the idea for the nearest-fit and latest-earlier controls. We implemented these controls for STReEF and added calibration availability counts and exact local time prediction intervals.
usesA SEEG library from four earlier pulses per known site identified three of six first return pulses at 20–100 ms in three sessions
Its time and earlier-label controls motivated the earlier-label comparator and separate reporting of continuation and first return pulses.
usesCalibration chose a mean 56 ms endpoint with no lower observed eight-site SEEG accuracy in five offline sessions with input from 20 ms
We adapted the time-only site readers and the site-current block definition for a post-result diagnostic on our eight selected sites and 160 test pulses. The control limits the interpretation of the response endpoint result.