Using skeletal data for human motion analysis has several advantages over the other visual data such as video and images since the features extracted from skeletal data tend to be more compact and more robust to the variation between different subjects as the visual appearance is not included. We propose a graph neural network [Zhang et al. MICCAI2022] for diagnosing Parkinson's diseas (PD) by Parkinson's Tremor (PT) classification as it effectively learns the spatial relationship between body joints from graph-structured data. Inspired by the information gain analysis and the clinician observation that PT usually occurs only on one side of the early stage PD patient’s upper body, we propose a novel attention module with a lightweight pyramidal channel-squeezing-fusion architecture to capture the self, short and long-range joint information specific to PT and filter noise.
In [Rueangsirarak et al. IEEE TNSRE2018], we propose an automatic framework for classifying musculoskeletal and neurological disorders among older people based on 3D motion data. We also propose two new features to capture the relationship between joints across frames, known as 3D Relative Joint Displacement (3DRJDP) and 6D Symmetric Relative Joint Displacement (6DSymRJDP), such that the relative movement between joints can be analyzed. To handle noisy skeletal data captured using depth sensors such as Microsoft Kinect, we propose a framework that accurately classifies the nature of the 3D postures obtained by Kinect using a max-margin classifier in [Ho et al. CVIU2016]. Different from previous work in the area, we integrate the information about the reliability of the tracked joints in order to enhance the accuracy and robustness of our framework.