Physics-Informed Neural Networks for Tissue Elasticity Reconstruction in Magnetic Resonance Elastography

Area of Science:

• Medical Imaging
• Computational Physics
• Artificial Intelligence

Background:

• Magnetic resonance elastography (MRE) quantifies tissue stiffness for diagnosing liver fibrosis.
• Elasticity mapping involves solving complex inverse problems using partial differential equations (PDEs).
• Existing numerical methods are sensitive to noise and require predefined physical relationships.

Purpose of the Study:

• To apply physics-informed neural networks (PINNs) for solving the inverse problem in tissue elasticity reconstruction.
• To develop a method that avoids numerical differentiation and respects physical constraints.
• To improve the accuracy and robustness of elasticity mapping in MRE.

Main Methods:

• Utilized PINNs to solve the inverse problem for elasticity reconstruction.
• Developed a method that does not rely on numerical differentiation.
• Integrated anatomical information to enhance learning of physical correlations.
• Validated the approach on simulated and in vivo patient data (non-alcoholic fatty liver disease).

Main Results:

• The PINN-based method demonstrated increased robustness to noise compared to traditional numerical techniques.
• Achieved higher accuracy in elasticity reconstruction on realistic simulated and in vivo data.
• Incorporating anatomical information further improved the method's performance.

Conclusions:

• PINNs provide a powerful, noise-resilient tool for non-invasive tissue elasticity reconstruction via MRE.
• The proposed method enhances diagnostic capabilities for liver fibrosis and other conditions.
• This AI-driven approach offers a more accurate and adaptable alternative to conventional numerical methods.