A generative model of hyperelastic strain energy density functions for multiple tissue brain deformation

Area of Science:

• Computational mechanics
• Biomedical engineering
• Neurosurgery

Background:

• Accurate estimation of brain deformation is critical for neurosurgery.
• Existing biomechanical models face limitations due to experimental conditions, leading to variability in reported tissue properties.
• Understanding tissue mechanics is essential for developing reliable surgical simulations.

Purpose of the Study:

• To demonstrate a generative model capable of estimating tissue elastic properties using observed brain deformation.
• To leverage strain energy density functions within a generative framework.
• To address variability in mechanical characterization of brain tissue.

Main Methods:

• Utilized Gaussian Process regression to learn elastic potentials from 73 scientific manuscripts.
• Evaluated neo-Hookean, Mooney-Rivlin, and 1-term Ogden meta-models for stability.
• Validated the generative model on a synthetic brain model and eight clinical temporal lobe resection cases, comparing pre- and post-operative images.

Main Results:

• Achieved high accuracy on a synthetic model with root-mean-square errors of 0.1 mm and 0.2 mm.
• Reduced root-mean-square error for brain deformation in clinical cases: 1.37 mm to 1.08 mm (ventricle surface) and 5.89 mm to 4.84 mm (resection cavity surface).
• The generative model captured uncertainties in mechanical characterization.

Conclusions:

• The generative model effectively captures uncertainties in tissue mechanical characterization.
• All evaluated meta-models performed similarly on elastography and linear studies; Ogden excelled on hyperelastic studies.
• While the model predicted elastic parameters in a synthetic phantom, it only partially explained deformation observed in clinical neurosurgery cases.