Predicting and understanding arterial elasticity from key microstructural features by bidirectional deep learning.

Area of Science:

• Biomedical Engineering
• Computational Biology
• Materials Science

Background:

• Soft biological tissues exhibit complex relationships between microstructure and mechanical properties.
• Inter-individual variations and changes due to aging or disease complicate understanding these relationships.
• Existing methods struggle to predict mechanical properties from microstructure or quantify feature contributions systematically.

Purpose of the Study:

• To develop a bidirectional deep learning architecture for predicting macroscopic mechanical properties from tissue microstructure.
• To enable automated and unbiased quantification of microstructural feature contributions to mechanical properties.
• To address key questions regarding predictive modeling and feature relevance in soft biological tissues.

Main Methods:

• Utilized a bidirectional deep learning architecture.
• Integrated data from histological analyses, two-photon microscopy, and biaxial biomechanical testing.
• Applied the model to murine aorta tissue during maturation and aging.

Main Results:

• Achieved high accuracy (R²=0.92) in predicting evolving mechanical properties of the murine aorta.
• Demonstrated the model's capability to predict mechanical properties from microstructural data.
• Identified extracellular matrix composition and organization as key determinants of mechanical properties.

Conclusions:

• The proposed deep learning architecture accurately predicts soft tissue mechanical properties from microstructural imaging.
• This approach allows for automated and unbiased quantification of microstructural feature relevance.
• Opens new avenues for predictive mechanical modeling in soft tissues and tissue engineering.