A 5% rest calibration rule gave 3.9% mean later rest activation and 26.1% task window detection in eight SEEG recordings

We tested a rest calibration rule with public SEEG data from eight participants in OpenNeuro ds004473 version 1.0.2. We used separate time periods to fit a model, set a threshold, and test later trials. Each model used eight bipolar pairs and three power bands. The 5% rank rule gave 3.9% mean rest activation and 26.1% mean pre-response detection on later trials. A fixed score threshold of 0.5 gave 27.1% and 66.1%, respectively. Each participant had equal weight. Two recordings had rest activation above 5%. A 1% rule gave an infinite threshold in all eight recordings because each calibration set had fewer than 99 rest windows. These results show the detection cost of a low rest activation target for this model and task. They do not establish a continuous false alarm rate or movement intent. The package preserves features, source records, code, models, scores, figures, and checks.

stereoelectroencephalography · brain-computer interface · rest calibration · rank threshold · task detection · false activation