Sit-to-stand-to-sit analysis as a digital biomarker for Parkinson’s disease

Abstract Background Parkinson’s disease requires continuous assessment and long-term rehabilitation; however, access to in-person care is often limited by mobility constraints and healthcare availability. Remote monitoring approaches that rely on simple, widely accessible devices may help expand access to objective motor assessment in daily life. We aimed to investigate whether smartphone-recorded sit-to-stand and stand-to-sit movements can be used to quantify motor characteristics in individuals with Parkinson’s disease using pose estimation, center of gravity trajectories, and machine-learning-based classification. Methods Smartphone videos of repeated sit-to-stand-to-sit movements were collected for 32 patients with Parkinson’s disease. Pose estimation was used to derive normalized center of gravity trajectories, which were clustered using time-series analysis with dynamic time warping, and a simplified recurrent neural network was trained to classify the resulting movement patterns. Results Center of gravity trajectory clustering revealed three distinct sit-to-stand-to-sit movement patterns associated with differences in age, functional independence, balance, and Parkinson’s disease motor severity. The recurrent neural network classified these movement patterns with an accuracy of 81.2%. Conclusions Sit-to-stand-to-sit movements recorded with a smartphone contain interpretable motor signatures that provide an approximate indication of Parkinson’s disease motor severity and thereby offer a basis for smartphone-based remote assessment of motor function.