Application of a spatiotemporal distribution-based multidimensional gait feature algorithm in KOA gait assessment.

To address the lack of quantification in assessing KOA gait dysfunction due to subjective analyses, a multidimensional gait feature algorithm based on spatiotemporal distribution is proposed. Using monocular RGB video input, the algorithm reconstructs 3D human mesh via TokenHMR and dynamically fits an anatomically constrained SKEL model through Depth-aware Progressive SpatioTemporal Modeling (DPSTM). It constructs 3D hip-knee-ankle cyclograms for quantifying inter-joint coordination and extracts morphological features. Sample entropy, multiscale entropy, permutation entropy, Lyapunov exponents, and generalized fluctuation coefficients assess gait dynamic characteristics. Analysis of 45 subjects (22 controls, 23 KOA patients) demonstrated that the algorithm significantly improved 3D kinematic reconstruction accuracy for hip, knee and ankle joints by approximately 40% compared to mainstream 2D pose estimation methods. The KOA group showed 32.3% reduced cyclogram volume and 60.7% smaller knee-ankle cyclogram area compared to controls, revealing compressed three-joint synergy space and significantly weakened inter-joint coupling. The KOA group exhibited systematic abnormalities in multiple complexity features, with the knee joint’s multiscale entropy and Lyapunov exponent significantly increasing by threefold compared to the control group, reflecting aggravated gait complexity and dynamic instability. This algorithm offers important methodological support for precise clinical assessment through 3D reconstruction, tri-joint synergy quantification, and multidimensional complexity analysis.