Signed SEEG waveforms gave 64.8% mean accuracy for two candidate sites on later pulses within blocks in five patients

This study tested fixed classifiers on public SEEG data from five patients. Each test used two candidate stimulation sites. Five earlier pulses trained each site template. Five later pulses in the same site block tested it. Signed normalized waveforms from 100 to 500 ms gave 64.80% mean patient accuracy. Log RMS gave 61.69%. Before-pulse waveforms gave 49.88%. The mean gain over log RMS was 3.11 percentage points. Its descriptive 95% patient bootstrap range was -2.28 to 8.32 points; the exact patient test gave p=0.3125. An absolute shape score gave 58.22%. The package preserves all bins, predictions, code, source hashes, and checks. Raw reconstruction matched every stored bin. These within-block results do not show a general advantage over size or a new neural pathway.

SEEG · stimulation · waveform classification · CCEP · patient-level inference · OpenNeuro