A noninvasive method for determining elastic parameters of valve tissue using physics-informed neural networks

Area of Science:

• Biomedical Engineering
• Computational Biology
• Medical Imaging

Background:

• Accurate in vivo mechanical parameters of heart valves are crucial for predictive computer simulations of virtual interventions.
• Current noninvasive methods for determining these parameters are limited, hindering clinical applications.
• Physics-informed neural networks offer a promising avenue for estimating complex biomechanical properties.

Purpose of the Study:

• To develop and validate a novel noninvasive method for determining patient-specific elastic parameters of heart valve tissue.
• To apply this method to estimate the mechanical properties of a child's tricuspid valve.
• To enhance the accuracy of computer simulations for predicting valve repair outcomes.

Main Methods:

• Utilized 3D echocardiogram time sequences to track tricuspid valve displacements via image registration.
• Employed physics-informed neural networks to estimate nonlinear mechanical properties from first principles and reference displacements.
• Validated the method by comparing simulation results with reference image segmentation.

Main Results:

• The physics-informed neural network method successfully estimated patient-specific elastic parameters for the tricuspid valve.
• Simulations using patient-specific parameters achieved a mean symmetric distance of less than 1 mm compared to reference segmentation.
• The accuracy of the simulated model was doubled compared to models using generic literature parameters.

Conclusions:

• The proposed noninvasive method effectively determines patient-specific valve mechanical properties.
• This advancement holds significant potential for improving the accuracy and clinical applicability of virtual interventions for valve repair.
• Accurate patient-specific modeling can lead to more personalized and effective surgical planning.