Later trials in eight SEEG recordings have mean 2.4% rest alarms and 9.5% response detection after five-window calibration at a nominal 5% level

We tested alarm calibration on public OpenNeuro ds004473 version 1.0.2. The test used 2014 quality-selected trials from eight SEEG recordings. Each person had separate fit, calibration, and later test trials. An alarm was any decision in five overlapping rest windows that covered 1.5 seconds before fixation. At a nominal 5% level, one-window calibration gave mean 14.4% scan alarms and 27.5% response detection. Calibration of the five-window maximum gave 2.4% alarms and 9.5% detection. 2 of eight persons had observed scan alarms above 5%. The higher threshold lowers both alarms and detection by construction. The later trials measure the size of this tradeoff. All persons had fewer than 99 calibration trials. Thus, the rank rule at 1% gave infinite thresholds and no detections. A rank false-alarm bound needs exchangeable calibration and test scores. Time order does not establish this assumption. The package preserves new scan features, exact raw retrieval records, individual scores, code, and figures. This offline test does not measure a continuous alarm rate or stimulation benefit.

stereoelectroencephalography · SEEG · brain-computer interface · false alarms · rank calibration · response detection · chronological test · signal processing