Movement-detection channels did not improve left/right decoding in 7 SEEG patients

We studied whether channels selected to detect movement onset also carry left/right choice information in a two-stage brain-computer interface. Using archived 70–150 Hz high-gamma window features from 7 SEEG patients performing a random-dot motion task with mouse responses, we selected 8 bipolar channels per patient by movement response on training trials. These channels detected movement well (mean test AUC 0.75; range 0.62–0.91). In the first 0.5 s of movement, they did not decode choice better than 8 random channels (mean AUC 0.504 vs. 0.508; difference −0.004; 4/7 patients positive; exact sign-flip p = 0.89). Choice-selected channels did not outperform detection-selected channels (mean difference +0.005, p = 0.83), and detection rank did not predict choice rank (mean Spearman rho = −0.004, p = 0.92). All three pre-registered tests were null after Holm correction. Choice decoding was near chance for all channels in the movement window (mean AUC 0.52), so the floor effect limits conclusions about small differences. In a secondary, uncorrected analysis, all channels decoded choice above chance in the 0.5 s before movement onset in 7/7 patients (mean AUC 0.533, p = 0.016). This offline result does not show that detection channels are poor for choice; it shows that, for these 0.5 s high-gamma means, a two-stage BCI cannot assume that channels selected for movement detection also provide movement direction. The study used one time-ordered split per patient and clinical electrode coverage with little motor cortex. No closed-loop BCI was tested.

brain-computer interface · SEEG · movement detection · choice decoding · high gamma · channel selection · left-right choice