A 500 ms median reduced excess 70 to 110 Hz power threshold crossings near labeled discharges in 23 SEEG sleep records

We tested a simple power rule for brain-computer interfaces (BCIs). We used public stereoelectroencephalography (SEEG) sleep records from 25 patients with epilepsy. We compared the mean and median of five past 100 ms power blocks from 70 to 110 Hz. Each rule had its own threshold from the first 60 seconds. The later test had 23 patients and 425 expert discharge labels. The median reduced the event crossing fraction minus the clean control fraction by 14.1 percentage points. The descriptive 95% patient bootstrap interval for this reduction was 9.6 to 18.8 points. The excess was lower in 19 of 23 patients. All eight fixed secondary checks had lower observed mean excess. In a controlled power model, the median crossed less often with 100 ms added power. With 300 ms added power, 99.1% of median trials had a crossing. The median responded about 66 ms later on trials with crossings from both rules. About 15% of unchanged background trials had a crossing. These rates therefore include background events. This model used real background power and idealized added power. The study does not measure intended commands or stimulation effects. The package has the source recipe, verified power data, code, results, four figures, and checks from raw source signals. It extends a prior sci study on the same source cohort. It is an exploratory study, with a plan saved before the new outcome calculations.

SEEG · high gamma power · interictal epileptiform discharges · threshold crossings · median filter · signal processing · brain-computer interface