Rest calibration reduced imagined pinch sensitivity in two SEEG participants with fixed linear readers trained on executed pinch

We tested whether rest calibration improves a reader trained on executed pinch. We used public SEEG records from two participants. A fixed linear reader used four power bands and early training blocks. Later imagined blocks supplied 43 paired target and rest periods. Calibration used early imagined rest data and no imagined target training data. With the median reference, balanced accuracy changed from 55.4% to 50.0% in A and 56.7% to 63.3% in B. Target sensitivity fell and rest specificity rose with both tested references. Calibration missed all 28 targets in A with the median reference. The constant score shift left AUC unchanged. These results show an error tradeoff in this offline test. They do not establish benefit or harm in a patient population. The package preserves features, source metadata, hashes, scores, figures, code, and a raw sample retrieval recipe.

stereoelectroencephalography · motor imagery · brain-computer interface · rest calibration · linear ridge reader · signal classification