Real-time biomechanics using the finite element method and machine learning: Review and perspective

Area of Science:

• Biomechanics
• Computational modeling
• Medical simulation

Background:

• Finite Element Method (FEM) is standard for anatomical simulations but is time-intensive.
• Clinical applications like haptics require real-time simulation capabilities, which FEM alone cannot provide.
• Machine Learning (ML) offers a potential solution to reduce simulation time without compromising accuracy.

Purpose of the Study:

• To review ML applications for accelerating FEM-based biomechanical simulations.
• To assess the impact of ML on simulation speed and performance.
• To identify effective ML strategies for anatomical simulations.

Main Methods:

• Systematic review of 41 publications on ML in FEM biomechanical simulations.
• Analysis of ML algorithms, data collection, anatomical structures, and validation metrics.
• Focus on neural networks and deep learning approaches.

Main Results:

• ML algorithms, particularly neural networks, are predominantly trained using FEM data.
• Tissue deformation is a common simulation target, with other features also addressed.
• Significant time reductions observed, from hours/minutes (FEM) to milliseconds (ML).

Conclusions:

• ML effectively accelerates FEM biomechanical simulations of anatomical structures.
• Achieving real-time simulation speeds with ML is feasible.
• Accelerated simulations can facilitate wider clinical adoption of biomechanical modeling.