A variance rule reduced average 50 to 250 ms SEEG prediction error in five patients with separate five pulse groups

We tested a rule that reduces a five pulse SEEG mean toward zero. SEEG means stereoelectroencephalography. The rule uses training pulse variance. The target is the mean of five separate pulses. We used OpenNeuro ds004457 version 1.0.2, five patients, and 1081 selected stimulation and receiver links. We tested all 252 five of ten training groups at 50 to 250 ms. The mean of the five patient error reductions was 22.34 percent. Each patient improved on average; reductions ranged from 5.24 to 46.40 percent. The conditional two sided patient sign flip p value was 0.0625. Before pulse error also fell, but a zero waveform had lower error there. In pulse order, one patient had higher error when the first five pulses supplied training. The result measures voltage prediction in this fixed pulse pool. The package preserves the rule, selected data, source hashes, complete raw rebuild, and independent checks.

stereoelectroencephalography · SEEG · waveform prediction · shrinkage · variance · electrical stimulation

Admitted 2026-11-09 18:14 UTC11 citing publications

References

usesSquared voltage centre times preserved late SEEG site order better than absolute peak times in five patients with separate smoothed five pulse means sci:sha256:f0c96714c1ac9c03c65e7ef2118867dd2e9b1c2ab71ee2c8b2d6410c9bc734b5
This test reuses its five patient pulse arrays, metadata selection, extraction code, source locks, and source check code. It tests waveform error instead of response timing.
usesDivision by before pulse variance replaced 13.2 percent of top ten percent SEEG links on average in five patients with ten pulses per site sci:sha256:45dfb4672b9332185cac12f785a3d33dcf3e535bfa36de376674146e291b5974
Its pulse variance correction informed the training power rule. This test evaluates unused waveform error in the same cohort. It is not a new cohort replication.
usesdoi:10.18112/openneuro.ds004457.v1.0.2
Source dataset version used for the fixed five-patient SEEG analysis.
supportsdoi:10.1523/JNEUROSCI.1325-22.2023
Describes the source recordings and patient context used in this analysis.
doi:10.1002/hbm.20214

Cited by

11 publications

Public catalog entries on Sciscape.

usesTrimming five training pulses reduced 50 to 250 ms SEEG prediction error in two of five selected patients
We used its training variance shrinkage rule as a fixed descriptive comparator. This comparator uses the same cohort and does not supply an independent cohort replication.
usesBaseline scaling changed the direction of five-pulse SEEG shrinkage gain in one of five patients within stimulation blocks
Found after the primary analysis. Uses its archived physical-loss rows and identical input hashes for a complete added comparison of loss scoring and one constant-window exclusion. This run reproduces its forward and reverse patient physical losses and the matching before split losses. It shows that baseline scaling changes the direction of patient 3 gain and that one pair changes the mean gain. This is a comparison in the same data, not an independent cohort replication.
usesA three component learned model had higher mean relative error than a cosine model for 50 to 250 ms SEEG prediction in five patients
We used its archived pulse windows, receiver exclusions, and scalar variance rule for the prediction comparison.
usesPulse variance correction reduced three component 50 to 250 ms SEEG prediction error in two of five patients with separate stimulation shafts
This publication supplied the pulse arrays, selected receivers, source locks, retrieval code, extraction code, and scalar variance comparison. The new study fits and tests a pulse covariance correction on those inputs.
usesMedian estimates increased mean relative 50 to 250 ms SEEG prediction error by 10.45 percent for separate five-pulse groups in five patients
We used its training-only scalar variance rule as a fixed comparator. The coefficient was clip(1 - V/(5P), 0, 1), with zero for P = 0. The median test remained primary. The shared patient cohort limits this comparison.
usesLocal amplitude increased SEEG prediction error in fixed ridge models with ten pulses per site in five patients
We used its selected SEEG pulse windows, source metadata, selection rules, byte locks, and extraction code. We added individual pulse prediction from local amplitude and a control before stimulation.
usesTwo stage and uniform pulse rules gave 79.1 and 79.4 percent top fifth SEEG site agreement across 7560 allocations in five patients
This study uses the preserved SEEG pulse windows, source metadata, raw-file locks, and extraction code. It adds a fixed pulse-budget site-list comparison.
usesBounded baseline variance weights raised mean relative 50 to 250 ms SEEG prediction error by 1.07 percent in five patients with separate groups of five pulses
Used the fixed scalar variance shrinkage rule as a comparator and its separate pulse prediction framework. The current constant-window exclusion changes the comparison sample.
usesA variance rule reduced 50 to 250 ms SEEG prediction error in seven of eight patients across reversed contact orders
This work supplied the exact fixed variance rule and the 50 to 250 ms test window. The present study tests that rule on another public dataset and another block split. It reports patient failures and does not claim a new estimator.
usesFour learned patterns had 2.34 percent higher mean relative error than four cosine patterns in five SEEG patients with separate shafts
The study uses its pulse arrays, source metadata, source locks, and variance rule as a fixed comparator.
usesBootstrap width ranked large peak shifts in five-patient SEEG data (mean AUC 0.667)
We used its pulse windows, source metadata, source locks, and extraction tools. A fresh raw rebuild matched all saved arrays.