Research use only. NeuroPD studies EEG-biomarker robustness across datasets. It does not diagnose Parkinson's disease and is not a medical device.
Robustness

Do the findings survive different choices?

Pre-declared checks, not a search for the best result. Values are balanced accuracy / ROC-AUC.

Channel harmonization — region vs. shared channels

featuresinternalexternal
region (210)0.563 / 0.7180.667 / 0.671
channel (1302)0.559 / 0.7150.500 / 0.529

Age/sex adjustment — does EEG add signal beyond demographics?

Age and sex are linearly partialled out of the EEG features inside each training fold (no leakage). If residualized EEG stays above chance, EEG carries signal independent of demographics.

modelinternal balanced acc / AUC
demographics only0.645 / 0.688
eeg0.563 / 0.718
eeg residualized on agesex0.577 / 0.668
eeg plus demographics0.618 / 0.717

Feature families

familyninternalexternal
relative power400.665 / 0.6870.833 / 0.867
absolute power400.692 / 0.7220.600 / 0.591
log power400.758 / 0.7730.567 / 0.640
slowing (paf/sef/ratios)400.664 / 0.7610.433 / 0.476
complexity500.674 / 0.7040.567 / 0.493

Full interpretation: ADR 0010.