The Clinical & Engineering Problem
Standard clinical hand goniometry suffers from high inter-rater variability, while commercial motion-capture systems require bulky hardware, calibration suits, or transmission of identifiable patient video to cloud servers.
The Technical Approach
Runs on-device computer vision models directly in the browser to extract 21 anatomical landmarks per hand, maintaining transparent separation between raw joint landmarks and clinical kinematic estimates.
Engineering Architecture
- Client-side machine learning pipeline running entirely within the browser sandbox
- Zero telemetry architecture with patient session records encrypted via device-local AES-256-GCM
- Calibrated range-of-motion calculations across flexion, extension, and radial/ulnar deviation
- Structured session exports supporting clinical audit trails and longitudinal tracking
Documented Evidence & Verification
Research kinematic measurement workspace. Not cleared as an automated diagnostic tool; intended to augment calibrated clinical evaluations. Browser camera depth varies across devices.