Are we ready to translate markerless motion capture in clinical gait analysis? A reliability assessment of different AI-driven approaches in healthy and Parkinson's disease subjects.

Neurodegenerative disorders represent the leading cause of physiotherapy demand worldwide, with incidence expected to increase. Parkinson's disease (PD) is a progressive neurodegenerative condition affecting the nervous system, leading to gait and balance alterations and requiring regular monitoring. Clinical scale assessments may suffer from operator subjectivity, while instrumental gait analysis is time-consuming and cumbersome. Recently, markerless (ML) motion capture technology has emerged as an alternative, enabling quantitative gait and posture evaluation in low-resource ecological settings and facilitating more frequent screening of disease progression. However, its clinical validity remains an open challenge. This study aims to assess the reliability of different markerless approaches applied to PD individuals adopting different commercial camera setups. A convenient sample of ten PD and ten healthy individuals (HS) were acquired synchronously with an optoelectronic system and two commercial cameras. Joint kinematics and spatiotemporal parameters, obtained through both marker-based and markerless approaches, were calculated. Additionally, the effects of the chosen biomechanical model and camera configuration were examined. In terms of validity assessment, results revealed good to excellent correlation (>0.7) in the comparison of joint rotation and gait speed. The lowest RMSE was observed for hip abb-adduction (5.10°) in the HS group, whereas the highest RMS was found for hip flexion extension (19.20°) still in the HS. In considering clinical applicability, results revealed higher values of Minimal Detectable Change (MDC) across the tested ML solutions (i.e., max = 28.57°, min = 4.87°) with respect to a stereophotogrammetric reference. Errors increased with fewer degrees of freedom, reduced camera number, and in pathological subjects.