Video-based hand pose estimation for Parkinsonian bradykinesia analysis and evaluation

Accurate assessment of Parkinsonian bradykinesia is essential for the clinical diagnosis and monitoring of Parkinson's disease (PD). This paper presents a video-based method designed to automatically quantify the severity of Parkinsonian bradykinesia. Three tasks related to hand bradykinesia are investigated: finger tapping, hand movements, and hand pronation–supination. We propose a Stacked Multi-Scale Hourglass (SMSH) model for hand pose estimation. The SMSH loss function is designed to incorporate the structural characteristics of the hand. Furthermore, hand blur augmentation is employed to enhance keypoint detection in blurred images, while a distribution-aware coordinate decoding method refines keypoint localization. Statistical features extracted from hand motion trajectories are processed by a classifier to assess Parkinsonian bradykinesia. Experiments conducted on real-world datasets demonstrate that the proposed method can effectively estimate hand poses in patients with PD and accurately assess the severity of bradykinesia.