Video-based quantification of leg agility in Parkinson’s disease and its relationship to upper extremity motor impairments

Bradykinesia is a hallmark sign of Parkinson’s disease (PD). Current clinical assessment is susceptible to inter-rater variability and only provides a single, ordinal severity score. Here, we propose a computer vision-based framework to quantify distinct motor characteristics from video recordings of the leg agility test, including slowness (bradykinesia), reduced amplitude (hypokinesia), progressive decrement (sequence effect) and irregularity (hesitation-halts).

We validate our approach using a large-scale dataset of 3097 video recordings captured from 443 participants in the Personalized Parkinson Project. We demonstrated that features reflecting bradykinesia, hypokinesia and hesitation-halts differed significantly across clinical severity ratings, whereas sequence effect features showed no consistent differences. Furthermore, data-driven analysis using principal component analysis with varimax rotation revealed that two additional feature dimensions (i.e., beyond the four conventional domains) may be necessary to fully characterize motor abnormalities during the leg agility test.

We also explored the relationship between finger-tapping (a distal upper limb task) and leg agility (a proximal lower extremity task). The analyses showed that upper and lower limb tasks provide differential insights into PD motor impairment, and that asymmetry between the least and most affected side was more pronounced in the upper extremities. This work highlights the ability of video-based assessment to provide objective characterization of motor impairment in PD, with the potential to support both in-clinic evaluations and at-home remote monitoring.

Future work will investigate responsiveness to medication and longitudinal disease progression. We would like to express our gratitude to all study participants and study assessors for making this work possible. We are grateful to Yağmur Güçlütürk for her valuable methodological suggestions.

We thank Mona Fayyazi for her assistance with creating the video annotations. This work was financially supported by the Dutch Research Council Long-Term Program (project \#KICH3.LTP.20.006, financed by the Dutch Research Council, Verily, the Dutch Ministry of Economic Affairs and Climate Policy) and the Donders Institute for Brain, Cognition and Behaviour. Michael Tangermann received support from the DBI2 project (024.005.022, Gravitation), which is financed by the Dutch Ministry of Education (OCW) via the Dutch Research Council (NWO), from the Dareplane collaboration project, which is co-funded by PPP Allowance awarded by Health Holland, Top Sector Life Sciences & Health, and by a contribution from the Dutch Brain Foundation.

The Center of Expertise for Parkinson and Movement Disorders was supported by a Center of Excellence grant from the Parkinson's Foundation. Center of Expertise for Parkinson and Movement Disorders, Department of Neurology, Donders Institute for Brain, Cognition and Behaviour, Radboud University Medical Center, Nijmegen, Netherlands Tahereh Zarrat Ehsan, Bastiaan R. Evers Department of Artificial Intelligence, Donders Institute for Brain, Cognition and Behaviour, Radboud University, Nijmegen, Netherlands Verily Life Sciences, Verily Life Sciences LLC, Dallas, USA Prof.