In eight SEEG participants, a 0.1 s power window reduced offline key-press detection AUC by 0.063 versus 0.5 s

This study tested how power-integration length affects offline cued key-press versus rest detection from stereo-EEG (SEEG). Using the first 100 trials from eight participants, the same zero-phase filtering and event-centered windows were used while power-averaging length changed from 0.05 to 0.5 s. With 70–150 Hz power and shrinkage LDA, mean AUC was 0.737 at 0.1 s and 0.800 at 0.5 s (mean difference −0.063; lower in 7 of 8 participants; exact two-sided sign-flip p = 0.023). Mean AUC increased monotonically across tested lengths. At 0.3 s the mean loss was 0.014 (p = 0.17), below the planned 0.02 practical margin. Adding 13–30 Hz power showed the same pattern (mean AUC 0.782 at 0.1 s and 0.832 at 0.5 s). None of eight secondary contrasts was significant after Holm correction (minimum adjusted p = 0.11). The result applies to an offline cued task with windows centered on the event. The zero-phase filter uses data outside each short integration window, so an online causal system would add filter delay not measured here. The analysis included eight participants and did not estimate information-transfer rate.

stereo-EEG · brain-computer interface · key-press detection · power window · AUC · high gamma · shrinkage LDA · offline decoding