Abstract
Idiopathic REM sleep behavior disorder (iRBD) is a strong predictor of neurodegenerative diseases like Parkinson's disease (PD). Early detection of motor impairments such as bradykinesia is critical for identifying at-risk populations. This study analyzed Finger Tapping Task videos from 66 participants, including healthy controls (HC) and individuals with iRBD and PD. Only videos that received a clinician score of zero on the MDS-UPDRS Part-III finger tapping item were analyzed. Movement amplitude, speed, and their decrements during the task were directly estimated from the videos using machine learning algorithms. Bradykinesia and hypokinesia were detectable in PD but not in iRBD, while decrement in movement amplitude and speed were observed in PD and iRBD. Classification models achieved 81.5% accuracy distinguishing PD from HC, 79.8% distinguishing iRBD from HC, and 81.7% differentiating iRBD from PD. Video-based assessments offer a low-cost, scalable solution for supporting the identification individuals at risk of developing neurodegenerative diseases.