At threshold 0.5, a fixed response-versus-rest SEEG decoder had higher mean activation before go cues than at rest in later trials from eight patients

We tested decoder activation before the go cue in public SEEG data from eight patients with epilepsy. We used 1765 trials from OpenNeuro ds004473 version 1.0.2. Early trials selected four bipolar pairs and fitted each model. The last 711 trials formed the test. The response-versus-rest model had 76.4 percent mean balanced accuracy. It activated in 37.7 percent of pre-go windows and 21.6 percent of rest windows. The equal-patient difference was 16.1 percentage points, with a descriptive 95 percent interval of 8.4 to 23.0. The exact two-sided patient test gave p = 0.0234. A second model used pre-go windows as negative training examples. Its pre-go activation was 32.7 percent and its response sensitivity was 71.1 percent, compared with 74.3 percent for the first model. The test used event-defined windows within the same recordings. It does not measure continuous event detection or brain stimulation. The package preserves code, features, source records, models, predictions, three figures, and reproduction steps.

stereoelectroencephalography · brain-computer interface · premature activation · go cue · binary classification · task-period control