Can Generative Video Address the Clinical Video Data Gap? Evaluation of Synthetic Parkinsonian Hand Motions
- 1University of Pennsylvania
Abstract
Computer vision approaches to disease motor assessment are limited by clinical video dataset scarcity and distributional imbalance, motivating interest in generative video models as a potential source of training data. We introduce a three-component framework for evaluating the clinical fidelity of synthetic medical motion data, assessing visual fidelity, biomechanical stability, and pathophysiological accuracy, and apply it to videos generated by the text-to-video model Sora 2 depicting Parkinsonian hand-fisting motions across MDS-UPDRS severity levels 0-2 (normal to mild-moderate motor impairment). We further use real-versus-real bootstrapping to compare synthetic-real differences against the natural variability of real clinical videos. Although generated videos appear visually realistic, quantitative analysis shows substantial divergence from real clinical recordings, including weaker severity-dependent patterns of bradykinesia and hypokinesia, less realistic tremor rhythms, and substantial temporal instability. These findings reveal a motion-domain gap between Sora-2-generated and real clinical motor behavior in this Parkinsonian hand-motion setting and demonstrate that visual realism alone is insufficient to establish the clinical utility of synthetic video data for training or evaluating medical computer vision systems.
Related
- Parkinsonian syndromes in institutionalized older adults in the French Caribbean: Clinical characteristics and one-year mortality
- Harnessing conversational generative artificial intelligence and natural language processing to address the colorectal cancer screening adherence gap: the Azores endeavour
- Economic evaluation protocol of the Ntshembo (Hope) Trial (RCT) in rural and urban South Africa: an intervention to optimise adolescence body mass index (BMI) pre-conception to address the triple burden of malnutrition
- Advancing T Cell Engagers in Multiple Myeloma: Structural Design and Clinical Optimization to Address Current Limitations.
- Multidimensional evaluation of generative AI chatbots for public consultation on sudden cardiac death: safety, accuracy, empathy, reliability, quality, and readability across six models
BibTeX
@misc{feng2026generative,
title = {{Can Generative Video Address the Clinical Video Data Gap? Evaluation of Synthetic Parkinsonian Hand Motions}},
author = {Y. Feng and S. Mopidevi and K. Kinfu and J. Park and R. Vidal and K. B. Johnson},
howpublished = {medRxiv},
year = {2026},
month = oct,
doi = {10.64898/2026.10.05.26363719},
url = {https://www.medrxiv.org/content/10.64898/2026.10.05.26363719v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - Can Generative Video Address the Clinical Video Data Gap? Evaluation of Synthetic Parkinsonian Hand Motions
AU - Y. Feng
AU - S. Mopidevi
AU - K. Kinfu
AU - J. Park
AU - R. Vidal
AU - K. B. Johnson
PY - 2026
DA - 2026/10/05
PB - medRxiv
DO - 10.64898/2026.10.05.26363719
UR - https://www.medrxiv.org/content/10.64898/2026.10.05.26363719v1
AB - Computer vision approaches to disease motor assessment are limited by clinical video dataset scarcity and distributional imbalance, motivating interest in generative video models as a potential source of training data. We introduce a three-component framework for evaluating the clinical fidelity of synthetic medical motion data, assessing visual fidelity, biomechanical stability, and pathophysiological accuracy, and apply it to videos generated by the text-to-video model Sora 2 depicting Parkinsonian hand-fisting motions across MDS-UPDRS severity levels 0-2 (normal to mild-moderate motor impairment). We further use real-versus-real bootstrapping to compare synthetic-real differences against the natural variability of real clinical videos. Although generated videos appear visually realistic, quantitative analysis shows substantial divergence from real clinical recordings, including weaker severity-dependent patterns of bradykinesia and hypokinesia, less realistic tremor rhythms, and substantial temporal instability. These findings reveal a motion-domain gap between Sora-2-generated and real clinical motor behavior in this Parkinsonian hand-motion setting and demonstrate that visual realism alone is insufficient to establish the clinical utility of synthetic video data for training or evaluating medical computer vision systems.
ER -