SWENet: A Physics-Informed Deep Neural Network (PINN) for Shear Wave Elastography

Area of Science:

• Biomedical Engineering
• Materials Science
• Artificial Intelligence

Background:

• Shear wave elastography (SWE) non-invasively measures soft material elastic properties.
• Conventional SWE methods struggle with wave diffraction in heterogeneous materials.
• Accurate material property inference is crucial for various scientific and medical applications.

Purpose of the Study:

• To develop an advanced SWE method overcoming limitations of current techniques.
• To introduce a physics-informed neural network (PINN)-based approach for enhanced elastography.
• To accurately infer heterogeneous mechanical properties in soft materials.

Main Methods:

• Proposed a physics-informed neural network (PINN)-based shear wave elastography (SWENet) method.
• Encoded spatial variations of elastic properties into governing equations as loss functions within SWENet.
• Trained neural networks using snapshots of wave motions for simultaneous elastic property inference.

Main Results:

• SWENet accurately identified shear moduli in soft composites with millimeter-sized inclusions.
• The method demonstrated high accuracy for both regular and irregular inclusion geometries.
• Validation performed using finite element simulations, phantom experiments, and ex vivo studies.

Conclusions:

• SWENet offers superior performance over conventional SWE by utilizing more wave motion features.
• The method allows seamless integration of multi-source data for inverse analysis.
• SWENet shows potential for broad applications in inferring heterogeneous mechanical properties, including medical imaging.