A Gradient-Based Machine-Learning Inspired Inverse Modeling Approach for Characterization of Nonlinear Tissue...

Area of Science:

• Biomechanics
• Computational mechanics
• Medical imaging analysis

Background:

• Accurate characterization of soft tissue material properties is essential for clinical applications.
• Understanding anisotropic nonlinear behavior and fiber directions is key for modeling tissue mechanics.

Purpose of the Study:

• To develop and validate a gradient-based inverse solver for retrieving anisotropic nonlinear material properties and fiber directions in soft tissues.
• To enable characterization based on known deformations and external loads, addressing a significant clinical goal.

Main Methods:

• A gradient-based inverse solver directly minimizes nodal force residuals, avoiding repeated forward simulations.
• Material properties are parameterized using a multilayer perceptron (MLP) mapping spatial location to behavior.
• Residual smoothing and parallelized Jacobian computation enhance convergence speed; supports general constitutive laws.

Main Results:

• The solver successfully recovers spatially varying elasticity and anisotropic fiber distributions in complex 3D geometries, demonstrated on a bioprosthetic heart valve.
• A residual-only formulation offers faster optimization than Physics-Informed Neural Network (PINN) approaches.
• While the Fung model shows parameter nonuniqueness, fiber direction recovery remains robust.

Conclusions:

• The developed inverse solver effectively recovers nonlinear material properties and fiber directions for complex 3D soft tissue problems, including those with large deformations.
• This method holds promise for advancing the understanding and clinical application of soft tissue mechanics.