Neural Network-Driven Finite Element Modeling for Estimating Knee Joint Cartilage Mechanical Responses.

Area of Science:

• Biomechanics
• Biomedical Engineering
• Artificial Intelligence

Background:

• High-fidelity motion capture (Mocap) is standard for human movement analysis.
• Current methods lack low-fidelity approaches for studying knee joint tissue mechanics.
• Finite element (FE) models are crucial for understanding tissue-level responses.

Purpose of the Study:

• To investigate knee cartilage stresses and strains using FE models driven by both high- and low-fidelity motion capture methods.
• To evaluate the comparability of AI-driven low-fidelity approaches with traditional high-fidelity methods for knee biomechanics.
• To assess the potential of AI for out-of-laboratory knee cartilage analysis.

Main Methods:

• Subject-specific FE modeling of knee joints for nine healthy participants.
• Acquisition of high-fidelity kinematic data via Mocap and kinetic data via musculoskeletal modeling.
• Estimation of kinetic data using artificial neural networks (ANNs) from low-fidelity inputs (e.g., subject demographics, static knee angle, walking speed).

Main Results:

• High- and low-fidelity approaches yielded comparable estimates for tibial cartilage stress and strain at the first peak knee contact force.
• Significant differences were noted at the second peak contact force, particularly in the lateral compartment.
• Most differences diminished when comparing average values across cartilage contact areas.

Conclusions:

• Low-fidelity, AI-generated approaches show potential for assessing tibial cartilage mechanics.
• This out-of-laboratory tool could facilitate cartilage failure prediction.
• The method may improve the management of osteoarthritis through accessible biomechanical analysis.