Gains from three pulses reduced centered SEEG forecast error at 20–500 ms in two recordings

We tested whether three calibration pulses can set the size of a later SEEG response forecast. We used OpenNeuro ds003848 version 1.0.3 and data products from sci publication cd3aeba9-afaa-4013-9e3d-3dca48204436. The scope was two recordings, 59 conditions, 590 distinct pulses, and 3,276 bipolar readout instances. The primary gain uses shared calibration-pulse energy divided by mean-template energy, limited to zero to one. It is motivated by a common-signal and independent pulse-noise model. The gain uses no later target values. On fixed later pulses, it reduced centered 20–500 ms forecast error by 16.0% and 32.5% against the new-three mean. The equal-person normalized error fell from 2.0212 to 1.4606. It also beat fixed half-new and pooled means in both recordings. It did not beat zero in both recordings. It also reduced pre-pulse error. The uncorrected leave-one-pulse-out gain had slightly lower mean normalized error. These controls support noise reduction in this test. They do not establish useful neural response information in both people. This is a dependent analysis of the same source records, with no population claim. The package includes the gain derivation, code, data products, raw examples, source versions and hashes, tables, four figures, and reproduction instructions.

SEEG · intracranial EEG · signal processing · forecasting · brain-computer interfaces · calibration · noise reduction