Deep neural networks enable quantitative movement analysis using single-camera videos.

Area of Science:

• Biomedical Engineering
• Clinical Biomechanics
• Machine Learning in Healthcare

Background:

• Neurological and musculoskeletal diseases significantly impair patient mobility, affecting daily function and social engagement.
• Accurate quantitative motion assessment is vital for medical decisions but relies on costly, specialized equipment and expert analysis.
• Existing methods limit accessibility to quantitative motion analysis in clinical and home settings.

Purpose of the Study:

• To develop and validate a novel method for predicting key clinical motion parameters from standard patient videos.
• To enable accessible, quantitative gait analysis using commodity cameras.
• To facilitate large-scale research on movement disorders.

Main Methods:

• Development of machine learning models trained to predict specific motion parameters from video data.
• Utilizing ordinary videos of patients as input for the motion analysis system.
• Validation of model predictions against established motion capture techniques.

Main Results:

• Machine learning models accurately predicted walking speed (r=0.73), cadence (r=0.79), and knee flexion angle (r=0.83).
• The Gait Deviation Index (GDI), a comprehensive gait impairment metric, was predicted with high accuracy (r=0.75).
• Achieved correlations approach theoretical limits, indicating high reliability of the proposed method.

Conclusions:

• This video-based machine learning approach provides a cost-effective and accessible solution for quantitative motion analysis.
• The method enhances the availability of gait pathology quantification in diverse settings, including clinics and homes.
• Enables broader research participation and data collection for neurological and musculoskeletal disorder studies.