Predicting High-Risk Patients: A Biomechanical-Based Machine Learning Approach for Coronary Vulnerable Plaques Detection

Detecting vulnerable coronary plaques from coronary computed tomography angiography (CCTA) is crucial for preventing cardiac events, yet current methods are suboptimal. Biomechanical indicators as plaque structural stress and strain, though promising, are underutilized in vulnerability machine learning prediction models. Our study proposes a machine learning pipeline to predict vulnerable patients using biomechanical markers from finite element analysis (FEA), while also considering the variability in mechanical properties. The study involved 40 patients (33 non-vulnerable and 7 vulnerable) who underwent both CCTA and coronary optical coherence tomography. The 3D coronary artery models were segmented from CCTA, and FEA was performed considering average or variable arterial tissues mechanical properties. Plaque structural stresses and strains, along with patient’s pressure data, were used to train decision tree classifiers. Incorporating variability of the mechanical properties enabled improved classification performances (from AUC of 0.95 vs. 0.99, leave-one-out) demonstrating the effectiveness of considering the material parameter uncertainty in model training. This investigation underscores the promise of biomechanical plaque phenotyping for patient stratification.