Current rest subtraction gave no clear log loss gain in eight offline SEEG key readers after task changes with 64 calibration trials

We tested whether earlier rest power improves SEEG prediction of the response key after a task condition change. We used public OpenNeuro ds004473 version 1.0.2 data from eight patients. Each reader used 64 early trials for calibration. Pair selection, scaling, and penalty selection used early trials only. The next condition had 979 test trials. Mean log loss was 0.7212 nats for raw pre-response power and 0.7187 nats for pre-response power minus current rest power. The paired gain was 0.0024 nats. The 95% whole-patient bootstrap interval was -0.0227 to +0.0389 nats. Gain was positive in three patients. Both main readers had higher mean loss than a constant probability of 0.5. A secondary past-rest mean gave a larger gain, but one patient accounted for most of it. The results do not establish a useful gain or equivalent methods. This is an exploratory offline reading test. It uses eventual response time and complete responding trials. It has no stimulation outcome. The package preserves features, raw signal audit ranges, all predictions, model states, figures, source hashes, and retrieval code.

stereoelectroencephalography · brain-computer interface · response-key prediction · log loss · rest subtraction · calibration