Full five-pulse subtraction increased mean 70–110 Hz threshold crossings at 100–500 ms after stimulation in eight SEEG patients with fixed raw thresholds

We tested fixed waveform subtraction in public SEEG data from OpenNeuro ds004696 version 1.0.1. The sample had eight patients, 86 sites, and 886 recording-pair instances. Five early pulses fitted each template; five later pulses supplied test data. Four short blocks covered 100 to 500 ms after each pulse in the nominal 70 to 110 Hz band. At a fixed threshold from 20 early raw before-pulse powers, full subtraction changed after-pulse crossings by 5.60 percentage points (95% patient interval 4.81 to 6.43). The matched before-pulse change was 4.89 points (3.81 to 5.74). The primary after-minus-before contrast was 0.71 points (-0.57 to 2.24); it did not show a clear mean reduction. The exact patient sign-flip probability was 0.429688. Signals had no intended-command labels; crossings are not proven false commands. Activity in the first 100 ms and task decoding were not tested. The package retains compact inputs, source hashes, results, figures, and reproduction instructions. GPL analysis code is omitted; two proofs explain small-sample thresholds and template-noise power under stated independence assumptions.

stereoelectroencephalography · brain-computer interface · stimulation artifact · high-gamma activity · waveform subtraction · threshold crossings · calibration · signal processing