Shape features preserved fixed-threshold SEEG key-press accuracy under 3 dB gain drift but reduced AUC without drift in eight subjects

A fixed-threshold SEEG key-press detector may lose accuracy when channel gain changes after calibration. In a pre-registered offline analysis of eight subjects, shrinkage LDA used trials 1–100 for calibration and later trials for testing. We compared full high-gamma plus beta log-power features (HG+beta), their gain-invariant difference (HG−beta), and HG alone while adding simulated, independent per-channel gain offsets to test data (0–3 dB SD; 200 draws per subject and condition). At 3 dB, HG−beta had higher balanced accuracy than HG+beta (0.713 vs 0.636; mean difference +0.078; all 8 subjects; Holm-adjusted p=0.016). Without drift, HG−beta had lower AUC (0.784 vs 0.828; mean difference −0.044), and non-inferiority at a 0.02 margin was not shown (Holm p=0.875). The balanced-accuracy crossover was between 1 and 2 dB SD. These results support shape features for this simulated motor SEEG setting when expected channel gain drift exceeds about 1–2 dB SD. The drift was simulated, frequency-flat, independent across channels, and constant during each test block. The study included eight subjects and one session each and did not compare label-free re-centering.

stereo-EEG · SEEG · motor decoding · key-press detection · gain drift · spectral shape · high-gamma power · beta power · shrinkage LDA · balanced accuracy