Estimation of left ventricular parameters based on deep learning method.

Area of Science:

• Biomedical Engineering
• Computational Mechanics
• Cardiovascular Research

Background:

• Personalized human left ventricular (LV) modeling requires accurate estimation of myocardial material properties.
• Inverse problems in biomechanics are computationally intensive, limiting clinical applications.

Purpose of the Study:

• To develop a deep learning (DL) framework for inversely evaluating passive myocardial mechanical properties.
• To establish a nonlinear mapping between geometric parameters, pressure-volume (PV) curves, and constitutive parameters using DL.

Main Methods:

• Standardized LV geometric models were created and optimized using data from 27 healthy volunteers.
• Statistical methods and Latin hypercube sampling (LHS) generated geometric parameter datasets.
• A structure-based orthotropic Holzapfel-Ogden constitutive law described the LV myocardium.
• Multiple neural networks were trained to predict PV curve parameters and constitutive parameters.

Main Results:

• The DL method significantly improved computational efficiency compared to traditional optimization methods.
• A nonlinear mapping was successfully established between geometric, PV curve, and constitutive parameters.
• The approach demonstrated potential for rapid feedback integration with clinical data.

Conclusions:

• Deep learning offers a computationally efficient alternative for estimating myocardial material properties in LV modeling.
• This method enhances the feasibility of applying biomechanical simulations for personalized medicine and clinical insights.