Predicting Mid-Term Risk of Coronary Plaque Progression in Subclinical Non-Obstructive Coronary Atherosclerosis Using Explainable Machine Learning.
Authors
Affiliations (5)
Affiliations (5)
- Zhejiang Key Laboratory of Imaging and Interventional Medicine, Zhejiang Engineering Research Center of Interventional Medicine Engineering and Biotechnology, Key Laboratory of Precision Medicine of Lishui City, The Fifth Affiliated Hospital of Wenzhou Medical University, Lishui 323000, China; Department of radiology, Lishui Hospital of Zhejiang University, School of Medicine, Lishui, 323000, China.
- Department of Pharmacy, The Fifth Affiliated Hospital of Wenzhou Medical University, Lishui 323000, China.
- Department of Cardiology, The Fifth Affiliated Hospital of Wenzhou Medical University, Lishui 323000, China.
- Zhejiang Key Laboratory of Imaging and Interventional Medicine, Zhejiang Engineering Research Center of Interventional Medicine Engineering and Biotechnology, Key Laboratory of Precision Medicine of Lishui City, The Fifth Affiliated Hospital of Wenzhou Medical University, Lishui 323000, China; Department of radiology, Lishui Hospital of Zhejiang University, School of Medicine, Lishui, 323000, China. Electronic address: [email protected].
- Zhejiang Key Laboratory of Imaging and Interventional Medicine, Zhejiang Engineering Research Center of Interventional Medicine Engineering and Biotechnology, Key Laboratory of Precision Medicine of Lishui City, The Fifth Affiliated Hospital of Wenzhou Medical University, Lishui 323000, China; Department of radiology, Lishui Hospital of Zhejiang University, School of Medicine, Lishui, 323000, China. Electronic address: [email protected].
Abstract
Coronary atherosclerosis (CAS) is a leading cause of cardiovascular morbidity and mortality. Subclinical non-obstructive CAS, often underrecognized in practice, carries substantial risk. We developed a clinically interpretable machine learning model integrating traditional risk factors with quantitative plaque features derived from coronary computed tomography angiography (CCTA) to predict mid-term plaque progression. We retrospectively studied 664 adults with subclinical non-obstructive CAS from a community screening program who underwent serial CCTA (median follow-up: 56 months). Candidate variables were filtered by Spearman correlation and a bootstrap enhanced least absolute shrinkage and selection operator. Seven algorithms were compared; the optimal model was explained with SHapley Additive exPlanations (SHAP). Participants had a mean age of 62 years; 53.6% were men. The random forest (RF) model performed best. The final eight-feature RF achieved an AUC of 0.875 (95% CI, 0.827-0.920), with precision 0.863, accuracy 0.779, recall 0.726, F1 score 0.788, specificity 0.849, and negative predictive value 0.702. SHAP analysis identified fibrofatty plaque volume and coronary artery calcium category as the principal drivers of plaque progression. The proposed machine learning SHAP framework integrates imaging and clinical features to enable individualized risk stratification for subclinical non-obstructive CAS. Incorporating quantitative imaging biomarkers into longitudinal follow-up may facilitate the early identification of high-risk individuals and support the timely implementation of intensive lipid-lowering, anti-inflammatory therapy, and lifestyle modification, thereby maximizing plaque stabilization or regression and ultimately reducing the risk of future cardiovascular events.