Related Experiment Video
Updated: May 24, 2026
Automated Gait Analysis to Assess Functional Recovery in Rodents with Peripheral Nerve or Spinal Cord Contusion Injury
Published on: October 6, 2020
Automated gait classification: Comparison of automated algorithms to expert classification
Karen M Kruger1, Joseph J Krzak2, Ross S Chafetz3
1Shriners Children's Chicago, 2211 N. Oak Park Ave., Chicago, IL 60707, USA; Orthopedic and Rehabilitation Engineering Center, Marquette University & Medical College of Wisconsin, 1250 W. Wisconsin Ave., Milwaukee, WI 53233, USA.
Automated gait classification algorithms can accurately reproduce expert clinical assessments for children with cerebral palsy (CP), especially for systems with clear biomechanical boundaries, enhancing gait analysis consistency.
Area of Science:
- Biomechanical analysis
- Pediatric orthopedics
- Clinical gait analysis
Background:
- Accurate gait pattern classification in children with cerebral palsy (CP) is crucial for effective treatment guidance.
- Expert interpretation of gait data is time-consuming and can lack consistency.
- Existing automated clinical tools for gait analysis are limited.
Purpose of the Study:
- To evaluate if automated algorithms can replicate expert clinical gait classifications in children with CP.
- To compare the performance of automated classification against established gait analysis systems.
Main Methods:
- Developed an automated MATLAB algorithm for gait classification using Rodda & Graham and Rozumalski & Schwartz systems.
- Collected gait data from children with CP diagnosed with crouch gait.
- Quantified agreement between automated classifications and three independent expert analysts using statistical measures (Cohen's κ, weighted κ, percent agreement, macro F1).
Main Results:
- Substantial inter-rater reliability (κ = 0.753) was observed for the Rodda & Graham system.
- Moderate inter-rater reliability (κ = 0.456) was found for the Rozumalski & Schwartz system, with lower and more variable rater agreement.
- Automated classification showed higher agreement with experts for systems with clear biomechanical boundaries.
Conclusions:
- Automated gait classification aligns with expert analysis for systems with well-defined biomechanical criteria.
- Complex, cluster-based systems show lower agreement due to inherent classification ambiguity.
- Automated tools offer a reliable and objective ground-truth for large-scale gait analysis and training assessment algorithms.
