From markers to model: Personalized lower limb skeletal reconstruction based on statistical shape models

• Published in: Biomedical signal processing and control Vol. 104; p. 107518
• Main Authors: Zhai, Haoyu, Wang, Junqing, Guo, Lanting, Jin, Xiaoxian, Nie, Yong, Li, Kang, Wang, Hongkai
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.06.2025
• Subjects: Biomechanical analysis; Kinematic chains; Marker-based methods; Musculoskeletal modeling; Statistical shape model; Statistical shape model; Kinematic chains; Musculoskeletal modeling; Biomechanical analysis; Marker-based methods
• ISSN: 1746-8094
• DOI: 10.1016/j.bspc.2025.107518

•Marker-based approach enables lower limb skeletal reconstruction.•Kinematic chains and shape model were combined for dynamic posture and precise geometry.•An iterative method improved posture and shape alignment accuracy.•Model's performance verified through geometric and biomechanical evaluations. Musculoskeletal models are essential in biomedicine, sports science, and engineering for simulating joint and skeletal dynamics in diagnosis, rehabilitation, and performance analysis. While Computed Tomography and Magnetic Resonance Imaging provide accurate models, they are expensive and time-consuming. Marker-based methods offer a quicker alternative, but challenges persist in balancing accuracy, personalization, and cost. We developed a marker-based method for lower limb skeletal modeling, designed to separately handle posture and shape. Posture is adjusted through kinematic chains, reducing position differences and permitting flexible fitting. The statistical shape model governs the shape modeling, accurately reflecting bone morphology. An iterative algorithm refines the model by aligning it with anatomical landmarks, enabling automatic adaptation to individual subjects. We built a statistical shape model using 81 training cases and conducted three-fold cross-validation. The proposed method improved geometric accuracy over linear scaling (Root Mean Square Error: 3.80 mm, Dice Similarity Coefficient: 0.73, Hausdorff Distance: 19.65 mm). Following this, validation on the West China Hospital dataset confirmed the robustness of the method, the method extended its practical utility by accurately mapping bone and surface landmarks, effectively addressing soft tissue artifacts. Finally, biomechanical analysis using Grand Challenge Competition datasets demonstrated that our method provided more accurate estimates of knee contact forces, reducing Root Mean Square Error by 7.5 %, 28.4 %, and 16.0 % for total, medial, and lateral forces, respectively. The proposed method improves the accuracy of skeletal reconstructions, demonstrating its potential for advancing personalized healthcare and musculoskeletal research.