Surrogate Simulation of Subject-Specific Lateral Pinch via Deep Learning

Area of Science:

• Biomechanics
• Computational modeling
• Deep learning

Background:

• Musculoskeletal modeling and simulation are computationally intensive, hindering personalized model development and application.
• Existing methods require significant time and resources, limiting their use in research and clinical settings.

Purpose of the Study:

• To develop a deep learning-based adaptive surrogate model for rapid musculoskeletal simulations.
• To enable personalized simulations by accepting musculoskeletal parameters and muscle activations as input.
• To accelerate the simulation process for complex biomechanical tasks like lateral pinch.

Main Methods:

• A deep learning-based adaptive surrogate model was created for lateral pinch simulations.
• The model utilizes musculoskeletal parameters and muscle activations for personalization.
• Performance was evaluated against traditional OpenSim forward dynamics simulations.

Main Results:

• The surrogate model achieved an average root-mean-squared error (RMSE) of 2.27 N, comparable to experimental measurement errors.
• The model demonstrated sensitivity to both musculoskeletal parameters and muscle activations.
• Simulations were accelerated by 10-1000x, completing thousands of simulations in seconds.

Conclusions:

• The deep learning surrogate model offers a computationally efficient and accurate alternative for musculoskeletal simulations.
• Its speed and personalization capabilities facilitate broader application in research and potentially clinical settings.
• The model's differentiable nature opens possibilities for integration into optimization frameworks and experimental data modeling.