Featured Expert:
Co-Authors:
Jan Stenum, PhD
Alexander Pantelyat, MD
Ryan Roemmich, PhD
Objectives: Repetitive motion is an important part of many neurologic exams like the Unified Parkinson's Disease Rating Scale (UPDRS) for persons with Parkinson's disease (PD). Bradykinesia is a prevalent symptom in persons with PD, and the UPDRS tests this via observational assessment of repetitive movements. Bradykinesia is typically evaluated subjectively in the clinic via visual observation by a movement disorders neurologist, but computer vision offers an opportunity for remote and objective assessments.
Design: We evaluated a pose estimation approach for measuring movement frequency in persons with PD from smartphone videos. We used 48 videos of the UPDRS done on 22 individuals with PD. The videos were segmented into portions containing two repetitive motion tasks: finger tapping (FT) and hand opening and closing (HOC). Tasks were scored by a neurologist and given UPDRS subscores from 0-4 (no symptoms to severe symptoms). We used a workflow that included Google MediaPipe (a freely available pose estimation algorithm) to estimate movement frequency and compared that to ground-truth manual measurements.
Results: The means of the manually calculated and Mediapipe frequency for the FT tasks were 4.37±0.89 and 4.03±0.98 Hz, respectively. The means of the manually calculated and Mediapipe frequency for the HOC task were 2.98±0.88 and 2.66±0.75 Hz, respectively. The correlation coefficient between the manual and Mediapipe calculations for both movements showed strong positive correlations (0.74, 0.61). These results suggest that pose estimation is a promising tool to provide accurate assessments of upper limb bradykinesia in PD.
Conclusions: As this is a preliminary analysis, we restricted our analysis to the FT and HOC tasks, but we plan to include the remaining bradykinesia items in future work.