Score centering over the past 20 seconds did not show higher detection before key presses at a 5/min early budget in eight SEEG recordings

We tested whether removal of a past score mean helps a continuous SEEG command rule. We used eight patients in OpenNeuro ds004473, version 1.0.2. Each decoder used the first 40% of valid trials for fitting. The next 20% set thresholds under a limit of five background commands per eligible minute. We tested fixed scores and scores minus the mean of the past 20 seconds on later data. Both rules skipped the same first minute and used the same command wait of one second. The test included 753 responses and 104.24 minutes of continuous scans. Mean detection before key presses was 19.71% with fixed scores and 19.84% with centered scores. The paired change was +0.13 percentage points. Its 95% patient bootstrap interval was -1.29 to +1.79 points. The exact two-sided sign-flip p value was 0.9375. Mean background rates were 4.42 and 4.00 commands per eligible minute. Later rates were not matched. The detection result did not show a clear gain. Background commands give a task proxy for unwanted output. They do not show known intent errors. This is an exploratory test of existing data from one session per patient. The package has source code, input features, figures, provenance, and a raw-data retrieval recipe. This study tests signal reading. It does not test brain stimulation.

SEEG · brain-computer interface · score centering · causal decoder · command detection · continuous monitoring