Machine learning accuracy for assessment of functional movement in Low back pain based on clinically applicable performance Metrics: A systematic review.

PURPOSE: To assess whether machine learning (ML) can accurately evaluate functional kinematics in people with low back pain (LBP) when judged by psychometric properties, including validity, reliability, and measurement error.

METHODS

A systematic search of PubMed, Scopus, Web of Science, and IEEE Xplore identified studies applying ML with kinematic inputs for LBP assessment. Risk of bias was assessed using the Newcastle-Ottawa Scale and selected COSMIN domains.

RESULTS

Twenty studies met inclusion. Most reported criterion validity via accuracy, while few examined reliability or measurement error. Inertial sensors and support vector machines were the most common methods.

CONCLUSIONS

ML shows strong validity for LBP movement assessment, but limited psychometric reporting constrains clinical use.