Contact selection refitting reduced background and pulse calls in eight SEEG patients at 4–30 Hz with five controls

We tested contact selection and rank cutoffs in public SEEG from eight patients with epilepsy. The test used 86 stimulation sites, five early calibration pulses, and five later test pulses. The main feature was 4–30 Hz Hann energy from 100 to 500 ms after a pulse, with matched before-pulse windows. One rule reused five backgrounds to select one bipolar pair and set its cutoff. The other rule refitted pair selection for each of six rank roles. With equal person weights, refitting reduced background calls from 28.19% to 19.69% and pulse calls from 70.75% to 60.62%. Net calls fell by 1.64 percentage points; the 95% paired person bootstrap interval was −4.67 to 1.61. Thus the test showed no clear net gain. In independent Gamma noise, selection reuse raised background calls as candidate count rose. A symmetric rank bound requires conditional exchangeability, which these patient data do not establish. The archive preserves inputs, code, all decisions, four figures, source checks, and reproduction instructions; the GPL code is omitted from the SCI payload. The cohort was reused. The reader requires query data from all candidate pairs.

stereoelectroencephalography · SEEG · conformal ranking · contact selection · calibration · brain interfaces