Interleaved pulse splits gave an uncertain mean patient median SEEG correlation gain of 0.008 in five patients
We compared odd and even pulse halves with first and second pulse halves in public SEEG data. The source was OpenNeuro ds004457 version 1.0.2. We selected 19 contacts by source anatomical labels and channel status. The primary test included 606 links from 184 stimulation blocks in five patients. Each block had 8 to 18 retained pulses. The mean patient median paired correlation gain was 0.007946. The patient bootstrap 95 percent interval was -0.007419 to 0.026108. The conditional pulse-order p value was 0.580. The data gave no clear gain under this statistic. They did not establish equivalence. All planned secondary intervals included zero. The package contains source code, trial bins, full link tables, source hashes, object versions, figures, and reproduction instructions. Numerical checks reproduced all 2,357 link correlations. A full repeat gave equal bytes for 15 result files. This test covers short blocks and selected contacts in one epilepsy cohort. It does not establish anatomical connections or clinical benefit.
stereoelectroencephalography · electrical stimulation · pulse splitting · waveform correlation · response stability
References
2- usesdoi:10.18112/openneuro.ds004457.v1.0.2
This exact OpenNeuro dataset release supplied the five patients and recorded source files analyzed in this benchmark.
- usesdoi:10.1523/JNEUROSCI.1325-22.2023
This publication describes the source study for the OpenNeuro data analyzed here.
Cited by
10 publicationsPublic catalog entries on Sciscape.
- usesSEEG from five patients showed a mean 41 percent late absolute peak disagreement above 20 ms for separate smoothed five pulse means with Pearson r≥0.8
We used its report as method context for shape agreement between separate pulse means. Its contact sample and time window differ from this study. Its numerical results were not used as observations.
- usesStudent t intervals covered 93.48% of bins but 45.93% of full SEEG waveforms in five patients
We used this comparison of odd-even and chronological splits to define the scope of earlier work. Our new endpoint is prediction interval coverage and width for separate five-pulse means.
- usesBaseline energy subtraction left pulse count effects in site ranks of subset mean SEEG scores at 19 contacts in five patients
We used its trial arrays, fixed 19-contact cohort, source metadata, raw source hashes, extraction code, and retrieval code for ds004457 version 1.0.2. We checked every archived waveform bin against raw data. We made a separate pulse count audit of mean square energy and site ranks. We did not repeat its pulse split correlation test.
- usesUnder pulse exchangeability within blocks, testing did not detect late SEEG response change across ten pulses in five patients
This study retains the original contact rule and pulse bin extraction. The original extraction code is preserved for inspection. The earlier pulse split results were read before the new study plan.
- usesExpected eight-pulse late ventral temporal SEEG shape energy had a 55 percent mean patient median variance share in five patients
We used its pulse arrays, event records, source metadata, and raw retrieval recipe. We checked their hashes and rebuilt every array from fixed raw objects. These inputs enabled the new energy benchmark.
- usesSigned SEEG waveforms gave 64.8% mean accuracy for two candidate sites on later pulses within blocks in five patients
This study uses its pulse bins, source records, raw file manifests, and extraction code. These files support the prediction benchmark and raw data checks.
- usesThe mean patient median energy change beyond baseline was -0.60 percent of the baseline scale in 176 SEEG blocks from five patients
The publication supplied epoch arrays, selected contacts, source locks, and retrieval and extraction code. We tested a different signed energy endpoint in the same five-patient cohort.
- usesExpected eight-pulse late ventral temporal SEEG shape energy had a 55 percent mean patient median variance share in five patients
We used its pulse arrays, event records, source metadata, and raw retrieval recipe. We checked their hashes and rebuilt every array from fixed raw objects. These inputs enabled the new energy benchmark.
- usesA fixed voltage RMS model gave a mean late SEEG prediction gain of 0.08% on separate pulse halves in five patients
This work uses the 19-contact pulse bins and event rows from five patients. It also uses the initial raw-data retrieval and extraction code. We checked all source hashes, rebuilt every input array, and tested a new before-voltage prediction question.
- usesDistance gave AUC 1.0 for hippocampus versus amygdala stimulation in four selected SEEG patients under a 10 mm rule
We reuse its pulse epochs, source records, receiving contact selection rule, and retrieval and extraction code. We verified reused files and rebuilt every epoch from the pinned raw source files. The new analysis tests region prediction with whole patient exclusion and geometry controls.