A five-pulse shape check selected later repeatable SEEG readouts with 73 percent mean precision in eight patient sessions

This study tested a five-pulse shape check on public SEEG data from eight epilepsy patient sessions. It used 886 eligible readouts from OpenNeuro ds004696, version 1.0.1. The later test used two pulse groups that were separate from calibration. A size rule selected the same readout count at each stimulation site. Both rules selected 196 readouts. Mean patient precision was 73.2 percent for shape and 63.5 percent for size. The paired difference was 9.7 percentage points. The descriptive bootstrap 95 percent interval was 1.8 to 18.2 points. The primary exact sign-flip test gave p = 0.125. Four patients favored shape and four tied. Thus the primary test does not establish that shape selection is better. The package supplies a separate pulse benchmark, three figures, exact trial products, all selections, source hashes, and code. One public source derivative was rebuilt and matched the archive exactly. All stored data passed hash and numerical checks. The result concerns waveform agreement within a session. It does not establish better decoding or closed-loop control.

SEEG · electrical stimulation · waveform agreement · brain-computer interfaces · trial selection · signal processing