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

A brain-wave biomarker that works — until you change the dataset.

NeuroPD asks a plain question: do interpretable resting-state EEG features that separate Parkinson's patients from controls in one cohort still work in an independent one? Across two public datasets, the answer is sobering — and it is the point.

Same EEG features · what can a model actually predict? (ROC-AUC, 0.50 = chance)
Predict which dataset a recording is from 0.95
Predict Parkinson's vs. healthy (within a dataset) 0.72

A classifier tells which dataset a recording came from almost perfectly (0.95), but tells disease from the same features only modestly (0.72). The feature space is dominated by site and equipment, not by the disease.

6
models evaluated internally & across datasets →
0.95
ROC-AUC predicting dataset identity →
3
sensitivity analyses of the findings →