Can Gait Deviations Be Identified through Video-Based Gait Analysis? Validation of the BlazePose Pose Estimation Algorithm.

Accurate identification of gait deviations is critical for effective intervention in children with cerebral palsy (CP). However, traditional marker-based gait analysis systems, while highly precise, are costly and impractical for broad clinical applications. Video-based pose estimation algorithms (PEAs) such as BlazePose offer a potential alternative, leveraging standard video inputs to provide scalable and accessible gait assessment. This study assessed the validity of BlazePose in diagnosing gait deviations and analyzing joint kinematics in children with CP, using marker-based motion capture as the gold standard and comparing the diagnostic results with observational gait analysis (OGA). Confusion matrices, root mean square error (RMSE), and correlation analyses were used to assess diagnostic accuracy and joint angle validity. BlazePose achieved higher accuracy and precision but lower recall and F1 scores than the traditional observational gait analysis in diagnosing gait deviations. Additionally, the joint angle errors exceed 5°. While limitations remain, BlazePose has potential to be used for gait screening and remote monitoring, particularly in resource-limited situations.Clinical Relevance-This study demonstrates the clinical relevance of BlazePose as a cost-effective, accessible tool for gait analysis in children with cerebral palsy, particularly in resource-limited settings. By offering scalable and objective gait deviation diagnosis, it supports early detection and intervention, complementing traditional marker-based systems for broader clinical application.