Artificial Intelligence in Lower Limb Kinematics: Reliability and Reproducibility in the Analysis of Runners with Chronic Ankle Instability

Kinematic motion analysis plays a fundamental role in understanding motor patterns, identifying injury risk factors, and optimizing sports performance. In the context of chronic ankle instability, kinematic assessment has traditionally relied on marker-based optoelectronic systems, which, despite their high accuracy, have limitations related to high costs, the need for controlled environments, and susceptibility to errors caused by soft tissue artifacts. In recent years, machine learning-based approaches, such as markerless posture detection, have been developed to overcome these constraints. However, the accuracy and reliability of these tools still require systematic validation. This study aims to evaluate the accuracy, reliability, and reproducibility of different publicly available deep learning models for markerless posture detection by comparing them with a reference optoelectronic system. Forty young adult runners will be recruited to perform the Single-Leg Step Down functional test, with simultaneous data acquisition from digital cameras and an optoelectronic system. Statistical analysis will include the Bland-Altman test to assess agreement between methods, as well as the calculation of the coefficient of variation to measure inter-trial variability. The study¿s findings will provide evidence on the feasibility of using publicly available deep learning models for kinematic motion analysis, contributing to the validation of accessible and easily implementable tools in clinical and sports settings.