Eight added SEEG channels increased mean offline movement forecast error at penalty 10 in 10 MetaRDK records

We tested whether eight SEEG channels add useful information to movement forecasts. We used public MetaRDK data, OpenNeuro ds006253 version 1.0.3. We fit models on early trials and tested later trials. At 850 ms after the stimulus, the target was movement within the next 500 ms. The test used 10 records and 563 trials with no movement by 850 ms. At penalty 10, mean Brier score was 0.2751 with neural inputs and 0.2531 with task and history inputs. The paired change was +0.0220. Its descriptive 95% record-bootstrap interval was [+0.0078, +0.0378]. The descriptive interval excludes zero. At penalty 100, the mean change was +0.0004 and its interval was [-0.0048, +0.0057]. The source signal was processed offline. The result does not establish live BCI accuracy. The package includes figures, code, compact data, source hashes, masks, forecasts, and checks.

SEEG · movement forecasting · brain-computer interface · Brier score · high-gamma activity