Identification of two-dimensional passive motion model of a knee joint

• Published in: Biomedical signal processing and control Vol. 112; p. 108304
• Main Author: Drążkowska, Marta
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.02.2026
• Subjects: Feedforward neural network; Human joint; Kinematic motion model; Knee X-ray; Model identification; search algorithm; Knee X-ray; Kinematic motion model; search algorithm; Human joint; Feedforward neural network; Model identification
• ISSN: 1746-8094; 1746-8108
• DOI: 10.1016/j.bspc.2025.108304

Accurate modeling of knee joint kinematics is essential for both clinical assessment and biomechanical analysis. In this study, we present a method for identifying a two-dimensional motion model of the knee joint using X-ray image data. To ensure consistency across different images and subjects, we propose a robust bone coordinate system that can be unambiguously defined in each image. The model identification process is based on a dataset comprising multiple knee joint configurations representing various movements and pediatric subjects. To enhance interpretability and reduce sensitivity to estimation errors, the proposed motion model is divided into two independent submodels: one describing positional changes and the other describing orientation. Each submodel is identified separately. We investigate two structurally distinct classes of models and perform an extensive search over their parameter spaces using a modified A⋆ algorithm to guide the selection of candidate structures. Multiple candidate models are evaluated through a combined verification process. Final model selection is based on a comprehensive analysis including test set error, penalty metrics, and statistical testing. The chosen model is compared with other well-known models from the literature as well as with the ground truth available for synthetic data. The results demonstrate the ability of the proposed approach to accurately capture knee joint motion with reduced model complexity, providing a promising tool for further biomechanical studies and potential clinical applications. •Proposition of novel coordinate system definition for human joints on 2D images.•Collection of knee joint configurations across varying flexion and subjects.•Decomposition of the kinematic model into two error-resilient submodels.•Model identification employing polynomial and neural network methodologies.•Optimal architecture determined through graph-theoretic exploration.