Separate scales along and outside a 1 mA waveform increased mean 2 mA SEEG forecast error after mean removal at 20 to 500 ms in three participants
We tested whether a waveform direction from five 1 mA pulses helps forecast later voltage after 2 mA stimulation. The source was public SEEG data from OpenNeuro ds005448 version 1.0.0. Three 2 mA pulses set both forecasts. Two later 2 mA pulses tested them. The test used separate noise scales along and outside the low-current waveform. The reference used one noise scale. The test increased mean normalized square error by 0.75 percent at 20 to 500 ms. Error increased in all three participants. There were 54 ordered conditions and 2,816 readout records. The reference gave 22.77 percent lower error than a zero forecast. It gave 20.11 percent lower error than the raw three-pulse mean. These percentages are means of participant changes. A fixed stable-signal subset gave mixed participant results. All 540 public source ranges were retrieved again. Every saved data array and all new numerical results matched the rebuild. This is an exploratory study on reused clinical data. Current and time were confounded. The test measured recorded voltage error. It did not measure live interface control. The package supplies source pins, exact signal statistics, model weights, both target losses, code, three figures, and repeat instructions.
SEEG · electrical brain stimulation · brain-computer interface · voltage forecasting · signal processing · epilepsy · forecast error · OpenNeuro
References
4- usesDoubling 1 mA templates increased 2 mA forecast error after mean removal in three SEEG participants at 20 to 500 ms sci:sha256:ed5d2e82c3ea575a465a18d21828edcfd06200c1c68665190148aa4c4428e7ca
This work supplied the fixed cohort, exact waveform Gram matrices, full-rate waveform audits, source metadata, source version pins, range hashes, and retrieval code. We tested a new forecast comparison on the same targets and rebuilt all selected raw ranges.
- usesdoi:10.18112/openneuro.ds005448.v1.0.0
The analysis used matched-current SEEG records from STReEF, OpenNeuro ds005448 version 1.0.0.
- doi:10.1016/j.ynirp.2025.100274
- doi:10.1007/s12021-022-09567-6
Cited by
4 publicationsPublic catalog entries on Sciscape.
- usesA 1 mA prior changed observed scaled 2 mA forecast error by less than 1% against matched zero shrinkage in three SEEG participants at 20 to 500 ms with calibration from other sites
This study supplies the earlier stored-direction test and noise-scale reference on the same cohort. Our matched zero direction uses calibration from other physical sites.
- usesSEEG shape difference estimates stayed positive at 1 and 2 mA in three people at 20 to 500 ms with an independent pulse noise model
This work supplied the exact 54-condition cohort, pulse Gram matrices, audit epochs, pinned source ranges, and source metadata. We used these inputs to estimate a waveform determinant rather than forecast error. We rebuilt all 540 selected ranges and checked all arrays against the public source objects.
- usesSEEG shape difference estimates stayed positive at 1 and 2 mA in three people at 20 to 500 ms with an independent pulse noise model
This work supplied the exact 54-condition cohort, pulse Gram matrices, audit epochs, pinned source ranges, and source metadata. We used these inputs to estimate a waveform determinant rather than forecast error. We rebuilt all 540 selected ranges and checked all arrays against the public source objects.
- usesGains from other sites changed mean normalized 2 mA SEEG forecast error by less than 1 percent in three people at 20–500 ms
Its fixed local high-current noise scale supplied the calibration reference.