Computer Vision-Based Platform for Quantitative Bradykinesia Assessment in Parkinson's Disease

Abstract

Parkinson's disease is a rapidly growing neurodegenerative disorder, with bradykinesia being a core motor symptom for clinical diagnosis. However, conventional assessment based on observational scales is limited by subjectivity and inter-rater variability. This study presents a low-cost, contactless Internet of Things platform for objective assessment of hand bradykinesia using computer vision. Unlike most existing approaches that focus solely on finger tapping, the proposed system integrates three clinically relevant motor tasks: finger tapping, hand opening and closing, and forearm pronation and supination. MediaPipe Hands was used to track 21 hand landmarks and extract kinematic variables related to amplitude, frequency, and rhythm regularity. Machine learning classification was evaluated using a publicly available finger-tapping dataset comprising 100 participants, with logistic regression achieving the best performance, reaching an accuracy of 0.85 and an area under the receiver operating characteristic curve of 0.908. The results demonstrate the system's feasibility for extracting reproducible motor metrics and support the proposed classification approach. Therefore, the proposed platform provides a scalable solution for remote longitudinal monitoring and supports quantitative and more standardized bradykinesia assessment across diverse environments, with potential applicability in clinical follow-up scenarios.