SW-VEI-Net: A Physics-Informed Deep Neural Network for Shear Wave Viscoelasticity Imaging

Area of Science:

• Biomedical Engineering
• Medical Imaging
• Artificial Intelligence

Background:

• Quantitative viscoelasticity imaging using shear wave elastography (SWE) faces challenges due to complex wave physics and reconstruction method limitations.
• Accurate assessment of both elastic and viscous moduli is crucial for diagnosing conditions like liver fibrosis.

Purpose of the Study:

• To develop and validate SW-VEI-Net, a physics-informed neural network (PINN) for simultaneous reconstruction of shear elastic and viscous moduli.
• To enhance the accuracy, interpretability, and clinical applicability of SWE.

Main Methods:

• Developed SW-VEI-Net, a dual-network PINN integrating viscoelastic wave equations and a dual-loss function.
• Validated the model on tissue-mimicking phantoms, a rat liver fibrosis model, and clinical data.
• Compared SW-VEI-Net against state-of-the-art methods including SWENet and dispersion fitting (DF).

Main Results:

• SW-VEI-Net achieved higher accuracy in shear elastic modulus reconstruction compared to SWENet.
• Viscoelastic parameter maps from SW-VEI-Net showed comparable results to DF with enhanced robustness.
• SW-VEI-Net demonstrated superior performance in liver fibrosis staging (AUC 0.85 for ≥F2, 0.91 for =F4) compared to SWENet and DF.
• Strong agreement was observed with a commercial ultrasound system in healthy volunteers.

Conclusions:

• SW-VEI-Net offers a significant advancement in SWE by accurately and simultaneously quantifying elastic and viscous moduli.
• The method shows substantial clinical potential for early detection of hepatic fibrosis and malignant lesions.
• Integrating deep learning with wave physics enhances SWE capabilities for precise viscoelastic biomarker mapping.