Predictive stress analysis in simplified spinal disc model using physics-informed neural networks.

Area of Science:

• Biomechanical Engineering
• Computational Modeling
• Artificial Intelligence in Medicine

Background:

• Accurate prediction of stress distribution in spinal discs is crucial for understanding disc degeneration and developing effective treatments.
• Traditional biomechanical models often require extensive computational resources and simplifying assumptions.
• Physics-informed neural networks (PINNs) offer a novel approach by integrating physical laws into deep learning models.

Purpose of the Study:

• To develop and validate a PINN model for predicting stress distribution in a simplified spinal disc.
• To assess the accuracy and potential of PINNs in biomechanical modeling of spinal structures.
• To visualize stress patterns under different loading conditions to inform clinical applications.

Main Methods:

• A 3D physics-informed neural network (PINN) model was developed.
• The model incorporated 3D spatial inputs and enforced mechanical equilibrium using a custom loss function.
• The PINN was trained on synthetic data derived from elasticity principles.

Main Results:

• The PINN model achieved a Mean Absolute Error (MAE) of 0.026 and an R-squared (R²) of 74.6% on synthetic data.
• Visualizations revealed distinct stress patterns under various loading conditions.
• Peak stress was identified at the z = 1 location under top compression loading.

Conclusions:

• PINNs show significant potential for accurate biomechanical modeling of spinal structures.
• This approach can improve predictive accuracy in spinal biomechanics compared to traditional methods.
• The findings suggest that PINN models can inform clinical interventions for spinal conditions.