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

Area of Science:

• Biomechanical analysis
• Pediatric rehabilitation
• Medical technology innovation

Background:

• Accurate gait deviation identification is crucial for children with cerebral palsy (CP).
• Marker-based gait analysis is precise but expensive and impractical for widespread clinical use.
• Video-based pose estimation algorithms (PEAs) present a scalable and accessible alternative.

Purpose of the Study:

• To validate the use of BlazePose, a PEA, for diagnosing gait deviations in children with CP.
• To compare BlazePose's diagnostic accuracy and joint kinematics analysis against marker-based motion capture (gold standard) and observational gait analysis (OGA).

Main Methods:

• BlazePose was used to analyze gait in children with CP.
• Diagnostic results were compared with OGA.
• Marker-based motion capture served as the gold standard for validation.
• Statistical analyses included confusion matrices, RMSE, and correlation analyses.

Main Results:

• BlazePose demonstrated higher accuracy and precision compared to OGA for diagnosing gait deviations.
• However, BlazePose had lower recall and F1 scores than OGA.
• Joint angle errors using BlazePose exceeded 5 degrees.

Conclusions:

• BlazePose shows potential as a cost-effective tool for gait screening and remote monitoring in children with CP, especially in resource-limited settings.
• It can complement traditional systems by offering objective gait deviation diagnosis.
• Further refinement is needed to address limitations in recall and joint angle accuracy.