Towards Video-Based Movement Biomarkers for Neuromuscular Diseases

Neuromuscular diseases are characterized by loss of function due to progressive muscle weakening. Promising new therapies could slow disease progression, but the functional outcomes used in trials (e.g., timed functional tests) lack sensitivity. OpenCap, a tool that quantifies human movement from smartphone videos, could generate movement biomarkers that augment existing outcomes. Here we aimed to assess if machine learning models trained with video-based kinematic measures can outperform those trained with timed functional tests in diagnosis and disease differentiation tasks. We collected a cross-sectional dataset of 90 individuals with neuromuscular diseases and 39 without neuromuscular conditions performing eight activities. The video-based models outperformed the timed functional test models at differentiating between different neuromuscular diseases. Using data from a single activity—walking—video-based features derived from unsupervised dimensionality reduction also out-performed timed functional test features at classifying between individuals with and without a neuromuscular disease. Video-based movement analysis is a promising approach to developing more informative and disease-specific movement biomarkers.