Mouse movement start and pre-click SEEG patterns were separable in 15 patients, but pre-click activity also triggered a movement detector

In archived 70–150 Hz SEEG features from 15 patients performing a random-dot motion task, a shrinkage linear discriminant decoder separated the first 0.5 s after mouse-movement onset (MOVE) from the 0.5 s before the response click (CLICK) on held-out trials (mean AUC 0.670, 95% bootstrap CI 0.624–0.722; all 15 patients above 0.5; 12/15 label-swap p < 0.05; exact one-sided sign-flip p = 3.1 × 10⁻⁵). A MOVE-vs-REST detector also classified CLICK windows above chance (AUC 0.785 vs 0.818 for MOVE; difference −0.034, 95% CI −0.060 to −0.009; p = 0.022); at a training-rest threshold with 10% false alarms, it fired in 65% of CLICK windows. Separation remained for trials with movement-to-click gaps of at least 1.5 s (AUC 0.653, 13/14 patients above chance; one-sided p = 1.8 × 10⁻⁴, Holm-adjusted p = 3.7 × 10⁻⁴). Mean power alone did not explain the separation: 50% of channels had higher mean power in CLICK than MOVE on average. This offline, event-aligned result supports the possibility of a second, weak decoder (AUC about 0.67) for a point-and-click brain-computer interface. CLICK is the final 0.5 s of a reach, not the click itself; MOVE always preceded CLICK, so slow within-trial changes may contribute. The result is not a causal real-time click detector.

brain-computer interface · stereo-electroencephalography · high-gamma activity · mouse movement · click selection · linear discriminant analysis · OpenNeuro ds006253