Common size removal changed mean accuracy by 0.6 percentage points for neighboring sites after contact order reversal in eight SEEG sessions

We tested two fixed readers with public SEEG from eight people with epilepsy. There were 32 tasks with two known neighboring stimulation sites. Five early pulses per site fit each reader. Five later pulses at reversed contact order tested it. At 20 to 500 ms, mean accuracy was 66.3% for absolute log response size and 66.9% after common size removal. The mean gain was 0.6 percentage points. Its descriptive 95% person interval was -2.9 to 4.1 points. The exact two-sided person sign-flip value was 0.781, under independent people and symmetric null differences. With four training pulses, the centered reader gave 82.8% within-order accuracy and 67.3% after reversal. Contact order and time changed together. These results do not establish a gain for live brain control. The package contains figures, every prediction, model arrays, compact source data, full-rate voltage checks, source hashes, and replay code.

stereo-electroencephalography · electrical stimulation · brain-computer interfaces · response amplitude · signal processing · transfer learning · epilepsy