Models from motor sites separated arithmetic from rest in EEG and fNIRS on first trials of blocks in 24 adults

We tested task labels with EEG and fNIRS from one public study. Five people supplied development. A fixed test used 24 other adults. Each personal model fit two runs and scored each block's first trial in the excluded run. AUC measures correct ordering of condition pairs. Mean AUC for arithmetic and rest was 0.739 in EEG and 0.735 in fNIRS. Their 95% person intervals were 0.690 to 0.788 and 0.679 to 0.791. The models exceeded a model of eye power by 0.156 and 0.152 AUC. Both shared tests passed Holm correction for six claims. Shared tests of hand labels and both tests of fusion gain failed. The arithmetic fusion gain had Holm p=0.0544. A source audit found one incomplete NEMAR EEG file. It held 25.315% of its declared bytes and matched the corresponding author-file prefix. We used the complete author file. The package gives code, source hashes, all outcomes, figures, and replay steps. The test concerns one healthy volunteer cohort and fixed models. Visual, eye, skin, and body blood signals can remain. The results do not identify a common neural cause. No independent study supplied confirmation.

EEG · fNIRS · motor imagery · mental arithmetic · cross-modal prediction · data provenance · brain-computer interface