Evaluation of range of motion using vision artificial intelligence (AI) in musculoskeletal medicine

Background: Measurements of mobility in musculoskeletal medicine are typically acquired using manual instruments that often have significant measurement variability and may lead to discrepancies in data. This study aims to evaluate the accuracy of vision artificial intelligence (AI) compared to human measurements when measuring participants’ range of motion (ROM). Methods: Thirty-one healthy adult participants and patients at a tertiary academic outpatient clinic were recruited. The active ROM of shoulder, hip, knee, and ankle are measured with the ViFive mobile application using vision AI. ROM was measured simultaneously using ViFive and a goniometer as the conventional standard. Mixed-effects linear regression models were used to determine the strength of the relationship between ViFive and goniometer measurements. Results: There were strong associations between measurements with the ViFive app and those measured by researchers using a goniometer for all joints examined [shoulder flexion (β=0.93, P<0.001), shoulder abduction (β=0.96, P<0.001), shoulder external rotation (β=0.78, P<0.001), shoulder internal rotation (β=0.81, P<0.001), hip flexion (β=0.62, P<0.001), knee flexion (β=0.78, P<0.001), and ankle dorsiflexion (β=0.36, P<0.001)]. Conclusions: Our study demonstrates that the ROM of various joints can be accurately measured by vision AI on the ViFive app as compared to the gold standard. There are new possibilities of using mobile device technologies to assess and rehab musculoskeletal conditions through virtual visits in lieu of the traditional physical exam and physical therapy.