Eight-pair SEEG readers gave 48.6 percent balanced accuracy for digit versus spoken format on later trials across separate number values in 11 patients

This study tested whether fixed eight-pair SEEG readers could distinguish seen digits from spoken number words on later trials with separate number values. Public OpenNeuro ds005415 version 1.0.0 supplied 616 trials from 11 patients. The readers trained on early values 1 to 4 and tested later values 6 to 9, and then tested the reverse direction. Mean balanced accuracy was 48.6 percent, with a descriptive 95 percent patient interval of 42.6 to 54.6 percent. The conditional label test gave p = 0.5763. Response accuracy was 10.3 percentage points higher than the before-onset control, which scored 38.3 percent. Under independent symmetric patient differences, the exact two-sided sign-flip test gave p = 0.0449. A gain over this control did not establish useful absolute decoding. The package includes figures, code, compact data, source records, model fits, predictions, random label draws, and reproduction instructions. Fresh raw-data checks and an exact offline repeat passed. The result tests one reader design in a reused epilepsy cohort. It does not establish intended-output decoding or brain stimulation.

stereoelectroencephalography · brain-computer interfaces · neural decoding · digits · spoken numbers · balanced accuracy