An Interpretable Machine Learning Model to Classify Coronary Bifurcation Lesions

Area of Science:

• Cardiovascular Medicine
• Biomedical Engineering
• Computational Fluid Dynamics

Background:

• Coronary bifurcation lesions are a significant cause of Coronary Artery Disease (CAD).
• Understanding the hemodynamic impact of lesion anatomy is crucial for effective treatment.
• Current treatment strategies are limited by incomplete knowledge of lesion-hemodynamic interactions.

Purpose of the Study:

• To model the impact of coronary lesion geometric features on local hemodynamic quantities.
• To develop an interpretable machine learning model for predicting hemodynamic changes.
• To advance personalized treatment planning for CAD patients.

Main Methods:

• Utilized the Classification and Regression Tree (CART) machine learning algorithm.
• Generated a synthetic arterial database using computational fluid dynamic (CFD) simulations.
• Applied CART to predict time-averaged wall shear stress (TAWSS) based on lesion anatomy.

Main Results:

• CART successfully created a simple, interpretable, and predictive nonlinear model of TAWSS.
• The model accurately estimates TAWSS based on geometric features of coronary bifurcation lesions.
• Demonstrated the capability of machine learning to link anatomical features with hemodynamic alterations.

Conclusions:

• Interpretable machine learning models can effectively predict hemodynamic disturbances caused by coronary bifurcation lesions.
• Fitted tree models offer potential for refining hemodynamic flow predictions based on patient-specific anatomy.
• This approach contributes to personalized treatment strategies for Coronary Artery Disease.