Super-fast and accurate nonlinear foot deformation Prediction using graph neural networks

Area of Science:

• Biomechanics
• Computational modeling
• Artificial intelligence

Background:

• Increasing prevalence of foot diseases necessitates efficient non-surgical treatments.
• Customized insoles offer a promising solution but traditional design is slow and costly.
• Finite Element Analysis (FEA) is computationally intensive for predicting foot deformation.

Purpose of the Study:

• To explore the use of Graph Neural Networks (GNNs) for predicting 3-D foot shape under load.
• To evaluate GNN performance based on dataset size.
• To assess the speed and accuracy of GNNs compared to FEA for insole design.

Main Methods:

• Utilized MeshGraphNet framework for GNN development.
• Trained GNN with 186 3-D foot geometries and FEA-predicted deformations.
• Optimized GNN weights for accurate prediction of foot displacement.

Main Results:

• GNN model achieved over 95% accuracy (R² values) in predicting foot displacement.
• GNN was approximately 97.52 times faster than traditional FEA simulations.
• Performance was tested across varying dataset sizes.

Conclusions:

• GNNs significantly enhance the efficiency and reduce the cost of custom insole manufacturing.
• This AI-driven approach represents a major advancement in non-surgical foot condition treatments.
• GNNs show strong potential for revolutionizing custom orthotic design.