Adding a pre-click window did not improve offline SEEG movement detection in 15 patients

In an offline analysis of archived 70–150 Hz high-gamma activity (HGA) window features from 15 patients in OpenNeuro ds006253, we tested whether adding the 0.5 s before each response click (PRECLICK) as a second positive training example improved shrinkage-LDA detection of movement (MOVE) versus pre-stimulus rest (REST). For the pre-registered primary analysis using the first 20 calibration trials and the last 40% of trials for testing, augmentation changed mean AUC by −0.0036 (95% patient-bootstrap CI −0.018 to +0.010; 6/15 patients improved; exact two-sided sign-flip p = 0.626). At 10, 40, and all calibration trials (60%), changes were also not significant after Holm correction (p = 1.0); all confidence intervals were within −0.02 to +0.025 AUC. With all calibration trials, the MOVE-vs-REST detector already scored PRECLICK above REST (mean AUC 0.812). The base detector's mean AUC was 0.833, matching the reported value for the archived feature contribution. The analysis is offline, uses one session per patient and event-aligned windows, and PRECLICK overlaps MOVE in trials where the click follows movement onset by less than 1 s (8%–66% per patient). With 15 patients, effects smaller than approximately 0.02 AUC cannot be excluded. These results apply to the tested extra-window rule; other movement windows were not tested.

SEEG · brain-computer interface · movement detection · high-gamma activity · shrinkage LDA · calibration trials · AUC · PRECLICK · OpenNeuro