Soft Tissue Simulation and Force Estimation From Heterogeneous Structures Using Equivariant Graph Neural Networks.

Area of Science:

• Computational mechanics
• Medical simulation
• Machine learning in healthcare

Background:

• Accurate soft tissue deformation simulation is vital for surgical training, pre-operative planning, and haptic feedback.
• Traditional physics-based models like finite element method (FEM) are computationally intensive and require extensive preprocessing.
• There is a need for efficient and robust simulation methods for real-time surgical applications.

Purpose of the Study:

• To develop a graph neural network (GNN) architecture for predicting soft tissue deformation and applied force from sparse point clouds.
• To incorporate internal anatomical information and leverage E(n)-equivariant message passing for improved model robustness.
• To provide a computationally efficient, data-driven alternative to traditional simulation methods.

Main Methods:

• Proposed a graph neural network (GNN) architecture utilizing sparse point clouds and incorporating internal anatomical data.
• Employed E(n)-equivariant message passing for enhanced robustness against variations in orientation and resolution.
• Collected experimental data from silicone and bone-like phantoms and supplemented with FEM simulations.

Main Results:

• The GNN model achieved comparable performance to baseline GNNs on standard tests.
• Demonstrated significantly improved performance in rotated and cross-resolution scenarios, indicating strong generalization.
• Achieved substantial speed improvements, making it suitable for real-time applications.
• Maintained sub-millimeter deformation accuracy on experimental data, even with limited samples and noise.

Conclusions:

• The proposed GNN approach offers an efficient, data-driven method for simulating soft tissue deformation.
• The model exhibits strong generalization capabilities across different anatomical configurations and data resolutions.
• This method provides a viable solution for real-time surgical simulation, training, and interactive environments.