ActiTect ML Pipeline Detects REM Sleep Disorder with 0.95 AUROC

ActiTect addresses a critical gap in prodromal Parkinson's detection: identifying isolated REM sleep behavior disorder (iRBD), a strong predictor of α-synucleinopathies (Parkinson's, Lewy body dementia, multiple system atrophy). While wrist-worn actimeters can capture abnormal nocturnal movements, no reliable, generalizable analysis pipeline existed. The team—spanning 20 authors from institutions including University of Cologne, Aarhus University, and University of Oxford—built a fully automated, open-source ML pipeline that preprocesses multi-device actigraphy data, performs automated sleep-wake detection, and extracts physiologically interpretable motion features.

Model training used a cohort of 78 individuals, achieving AUROC 0.95 under nested cross-validation. Generalization held across a blinded local test set (n=31, AUROC 0.86) and two independent external cohorts (n=113, AUROC 0.84; n=57, AUROC 0.94). Leave-one-dataset-out cross-validation confirmed consistent performance (AUROC range 0.84–0.89). By being open-source and easy to use, ActiTect promotes widespread adoption, independent validation, and collaborative improvements—paving the way for a unified, wearable-based RBD screening model.