Predicting capillary vessel network hemodynamics in silico by machine learning.

Area of Science:

• Biomedical Engineering
• Computational Fluid Dynamics
• Physiology

Background:

• Microcirculation blood flow is complex, deviating from simple models.
• Current imaging lacks 3D velocity and RBC concentration data crucial for physiological analysis.
• Existing computational models are too resource-intensive for large-scale microvascular networks.

Purpose of the Study:

• To develop machine learning models for accurate 3D blood flow and RBC concentration profiling.
• To predict key hemodynamic variables like wall shear stress (WSS) and cell-free layer (CFL).
• To enable large-scale, organ-level hemodynamic analysis in microvascular networks.

Main Methods:

• Utilized artificial neural networks and U-net architectures.
• Trained models on high-fidelity computational fluid dynamics data.
• Validated ML model predictions against established computational results.

Main Results:

• Achieved highly accurate 3D blood velocity and RBC concentration profiles.
• Successfully predicted WSS and CFL with excellent agreement to true data.
• Reduced computational time by several orders of magnitude compared to traditional methods.

Conclusions:

• Machine learning offers a computationally efficient alternative for detailed microvascular analysis.
• This approach enables organ-scale hemodynamic predictions previously unattainable.
• The developed ML models are vital for advancing research in blood flow regulation and disease.