Physics-Informed Deep-Learning For Elasticity: Forward, Inverse, and Mixed Problems.

Area of Science:

• Medical Imaging
• Biophysics
• Machine Learning

Background:

• Elastography measures tissue elasticity using ultrasound, but current methods have limitations.
• Poor lateral resolution and reliance on incompressibility assumptions hinder clinical accuracy.
• Existing techniques often require both axial and lateral displacement data.

Purpose of the Study:

• To develop a novel physics-informed deep learning method for improved elastography.
• To reconstruct tissue elasticity using only axial displacement data.
• To overcome limitations of current elastography, including the incompressibility assumption.

Main Methods:

• A deep learning framework integrating displacement and elasticity networks was employed.
• The method reconstructs Young's modulus from axial displacement, removing the incompressibility constraint.
• Simultaneous reconstruction of Young's modulus and Poisson's ratio is enabled.

Main Results:

• The proposed method accurately reconstructs the Young's modulus field of heterogeneous objects.
• It successfully reconstructs both Young's modulus and Poisson's ratio without assuming incompressibility.
• Utilizing multiple measurements mitigates errors from the 'eggshell' effect.

Conclusions:

• This physics-informed deep learning approach significantly advances elastography.
• It offers more accurate tissue elasticity measurements for medical imaging and material science.
• The method provides a robust tool for diverse applications, enhancing diagnostic capabilities.