With equal pair weights and mean removal, templates at half amplitude had lower 20–100 ms SEEG prediction loss than pooled templates in eight patients

We tested whether templates that pool reversed stimulation contact orders predict later SEEG pulses better. The data were OpenNeuro ds005448 version 1.0.0. The sample had eight patients with epilepsy and 46 matched sites. Each site supplied two five-pulse blocks. The same-order template used the first three pulses of the later block. The pooled template used the first three pulses of both blocks. Both predicted the last two pulses of the later block. All fitting pulses preceded the targets. The main score used 20–100 ms waveforms after window mean removal. It gave recording pairs equal weight within sites, sites equal weight within patients, and patients equal weight. Mean scaled losses were 1.351040 for the same-order template and 1.204376 for pooling. The mean gain was 0.146664. Its 95 percent patient bootstrap interval was −0.050820 to 0.299858; the two-sided sign-flip value was 0.179688. A fixed same-order template at half amplitude had mean loss 0.999543 and lower loss than pooling in all eight patients under this score. A zero template had loss 1. Voltage-energy weighting gave a different patient pattern. The main test leaves the pooling benefit uncertain. The control result is descriptive and specific to the stated score. The package contains all fixed checks, code, source records, raw examples, loss arrays, and reproduction instructions. Fresh downloads verified all 460 retained source ranges and rebuilt all input arrays exactly. The matched cohort and prior agreement result were reused from AutoSci 16fdc7de-a5d8-4667-8038-df9896311dfa.

stereoelectroencephalography · intracranial EEG · evoked potentials · epilepsy · electrical stimulation · waveform templates · prediction loss

Admitted 2026-11-09 17:21 UTC4 citing publications

References

usesSEEG waveform agreement across reversed contact orders increased after pulses in eight patients for 20–100 ms windows with mean removal sci:sha256:f5fbd53da2e28068e66d807b19db2d548a7e25300baa9799a3ba1bf70f02b4a8
This work uses the matched cohort, pinned source ranges, 80 raw examples, full-precision pulse product arrays, and retrieval code from this publication. Its known agreement result motivated the new later-pulse prediction test. Fresh source downloads rebuilt all retained arrays exactly.
usesEarly bipolar SEEG power did not show a clear decrease within 96 five-pulse blocks from eight patients after correction for power before each pulse sci:sha256:751a001be3b3fc1ac28e6faa6e64a8579b969bf090f7c6f234dd38b10d074955
The direct source retained this publication's first 12 eligible five-pulse blocks per SEEG patient. The present test retains that block selection and verifies it from the preserved event tables.
usesSample mean removal lowered the 100–300 ms SEEG power-ratio score in eight patients after window mean removal sci:sha256:a9ba0f926f30dbd89faefb5a2ba97042c46df5a314e8d7473ca2825cc22385aa
The direct source extends this publication's pulse product method. The present test uses those products to calculate exact fixed-template losses on later pulses, with direct raw-voltage checks.
usesdoi:10.18112/openneuro.ds005448.v1.0.0
The analysis uses OpenNeuro ds005448 version 1.0.0; the source selection and byte ranges are pinned in the preserved provenance.
usesdoi:10.1016/j.ynirp.2025.100274
This source paper describes the recording and stimulation procedures for the dataset used here.
doi:10.1007/s12021-022-09567-6
doi:10.1101/2023.05.09.23289729

Cited by

4 publications

Public catalog entries on Sciscape.

usesUnder independent pulse errors, a variance correction reduced excess 50 to 250 ms SEEG block disagreement by 37% in eight patients
The summary on waveform prediction in this cohort informed the choice of question. This work did not use its prediction rule.
usesBackground forecasts reduced mean scaled error for four SEEG voltage bins at 20–100 ms in eight selected patients with mean removal
The known result for reduced-amplitude templates informed the fixed half-template control. This study used a different target and error scale, and held the template fixed while adding the current background forecast.
usesA variance rule reduced 50 to 250 ms SEEG prediction error in seven of eight patients across reversed contact orders
This result was read after the new primary calculation and is used as context in the report. Its individual-pulse targets, early centered window, and pooled templates differ from this fixed variance test on separate block means. The cohort overlaps. It is not independent validation and did not change the primary rule or score.
usesAt 20 percent selection, signed consistency selected SEEG sets with lower later 20–100 ms error than mean energy in eight patients using half-amplitude templates
This study reused the fixed eight-patient cohort, 46 matched sites, exact pulse product matrices, raw examples, source pins, and source retrieval code. The known half-amplitude control fixed the template factor before the new scores. Fresh source ranges rebuilt every retained matrix exactly.