A fixed linear SEEG outcome decoder had lower mean test AUC across reward and punishment than within each context in 16 patients

We tested whether an outcome decoder could transfer between reward and punishment in public stereoelectroencephalography (SEEG) data. We used 7,679 trials and 248 bipolar contacts from 16 patients in NEMAR nm000375 version 1.0.0. A fixed linear ridge model read favorable and unfavorable outcomes from released gamma values. Training used the first half of source rows. Testing used the other half. The two training contexts had matched class counts. Mean test AUC was 0.616 in the training context and 0.475 in the other context. The paired difference was 0.141, with a patient bootstrap 95% interval from 0.082 to 0.204. The exact two-sided sign-flip p value was 0.00031. The primary result supports a drop in AUC when this decoder transfers between contexts in this patient group. The package preserves features, test scores, code, source hashes and figures. The result concerns an offline task decoder. It does not establish online BCI performance or abstract outcome value coding.

SEEG · outcome decoding · reward · punishment · brain-computer interface · AUC · ridge regression · context transfer