Modified conventional gait model versus cluster tracking: Test-retest reliability, agreement and impact of inverse kinematics with joint constraints on kinematic and kinetic data

Background: Three-dimensional gait analysis is often used to assess kinematics and kinetics to discriminate gait patterns and examine change over time. Test-retest reliability is therefore imperative; however, many variations of gait models currently exist.

Research question: This study examined the test-retest reliability of, and agreement between, two commonly used methods of gait modelling, a modified Conventional Gait Model and cluster-based model, using both six degrees-of-freedom or inverse kinematics computational methods in Visual3D.

Methods: Thirty healthy participants attended two identical sessions. The data for both models were collected concurrently and analysed in Visual3D using either six degrees-of-freedom or inverse kinematics computational methods. Outcomes were taken as the peak measurements for kinematics (joint angles and angular velocity) and kinetics (joint moments and power) for the hip, knee and ankle. Intraclass correlation coefficients were used to examine reliability, with the standard error of measurement and minimal detectable change also calculated. Agreement between models was examined with Pearson correlations and intraclass correlation coefficients.

Results: Test-retest reliability was good to excellent for all models for lower limb kinematics and kinetics. The inverse kinematic models had slightly lower reliability across outcomes compared to the six degrees-of-freedom models. Agreement between the Conventional Gait Model and cluster model was mostly good to excellent. Comparison of the two modified CGMs (with six degrees-of-freedom and inverse kinematics) showed much higher agreement against the comparison of the two cluster-based models (with six degrees-of-freedom and inverse kinematics).

Significance: This study provides a comprehensive assessment of the test-retest reliability and agreement between two gait models with various computational methods. Future research may use these results to guide their decision making for the gait model and outcomes of interest to be used.