Combining Deep Data-Driven and Physics-Inspired Learning for Shear Wave Speed Estimation in Ultrasound Elastography.

Area of Science:

• Medical physics
• Biomedical imaging
• Ultrasound technology

Background:

• Shear wave elastography (SWE) quantifies tissue stiffness via shear wave speed (SWS) measurement.
• Current deep learning (DL) methods for SWS estimation often rely on simulated data, leading to real-world artifacts.
• Robust SWS mapping is crucial for accurate tissue characterization in clinical settings.

Purpose of the Study:

• To develop a physics-inspired deep learning approach for accurate and robust SWS estimation.
• To address the limitations of supervised DL methods trained on simulated data.
• To improve SWS map reliability in clinical applications.

Main Methods:

• A physics-inspired learning approach utilizing real-world data without known SWS values.
• Implementation of an adaptive unsupervised loss function for training.
• Validation using experimental phantom and in vivo human liver data.

Main Results:

• Demonstrated enhanced accuracy and reliability in SWS estimation compared to conventional and supervised methods.
• Successfully minimized artifacts in SWS maps using real, noisy data.
• Validated the robustness of the proposed method on phantom and in vivo datasets.

Conclusions:

• The proposed hybrid approach effectively combines data-driven and physics-inspired learning for improved SWS mapping.
• This method offers a promising solution for more accurate and reliable quantitative ultrasound elastography.
• The technique has the potential to enhance clinical diagnostic capabilities through better tissue characterization.