Hybrid physics-based and data-driven modeling of vascular bifurcation pressure differences.

Area of Science:

• Biomedical Engineering
• Computational Fluid Dynamics
• Machine Learning

Background:

• Reduced-order models (ROMs) offer efficient blood flow simulation in patient-specific vasculatures but often sacrifice accuracy due to simplifying assumptions.
• A key assumption in many ROMs is pressure continuity at vascular bifurcations, which can lead to significant errors in pressure prediction.

Purpose of the Study:

• To develop and validate a novel model that accurately predicts pressure differences across vascular bifurcations.
• To enhance the accuracy of cardiovascular reduced-order models for improved clinical utility.

Main Methods:

• A machine learning approach was integrated into a common ROM structure to predict pressure differences at vascular bifurcations.
• The model was tested on steady and transient blood flow data across three distinct bifurcation geometries.
• Different machine learning techniques were evaluated, with neural networks showing the most robust performance.

Main Results:

• The proposed model significantly improved prediction accuracy for bifurcation pressure losses compared to existing methods.
• A neural network-based approach demonstrated superior performance in predicting pressure differences.
• The model showed good generalization capabilities when tested on combined datasets of different bifurcation types.

Conclusions:

• The developed model effectively addresses the limitations of pressure continuity assumptions in cardiovascular ROMs.
• This work represents a significant advancement in improving the accuracy and reliability of reduced-order models for cardiovascular flow analysis.
• The findings suggest a promising direction for more precise and efficient patient-specific hemodynamic simulations.