Relationship Between the Results of Arm Swing Data From the OpenPose-Based Gait Analysis System and MDS-UPDRS Scores

Abstract

Motor functions of individuals with Parkinson’s disease (PD) have been studied by many methods such as sensors, 3D measurement devices, smartphone applications, and deep learning tools. OpenPose is a deep learning tool for human pose estimation. However, few studies have focused only on gait arm swing and analyzed it separately from that of healthy subjects. Furthermore, none of the previous studies compared arm swing data measured from video images using OpenPose or other methods with the relevant items in the MDS-UPDRS. In this study, we calculated the thresholds to distinguish between normal and abnormal gaits from the data of healthy subjects. We compared the peak-to-peak (P-P) data of the left and right arm swing and arm swing asymmetry (ASA) using an OpenPose-based gait analysis system developed in our previous study with the MDS-UPDRS scores. It showed 72.73–82.35% accuracy; thus, the threshold between normal and abnormal gaits can be improved. From these results, we can conclude that there is a significant relationship between the MDS-UPDRS score and the magnitude of the gait arm swing angle measured using the OpenPose-based gait analysis system. The method proposed in this study is simpler and easier to use in clinical practice than the methods proposed in previous studies. Moreover, videos of timed up-and-go (TUG) tests can be used. We hope that this study will enable the estimation and early detection of PD symptoms using a simpler index and method.

Keywords

• Arm swing
• deep learning
• MDS-UPDRS
• OpenPose
• Parkinson’s disease
• timed up-and-go test