Machine Learning Constructed Based on Patient Plaque and Clinical Features for Predicting Stent Malapposition: A...

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Stent malapposition (SM) remains a clinical challenge after percutaneous coronary intervention (PCI) for myocardial infarction.
• Machine learning (ML) shows promise in predictive modeling for cardiovascular diseases.

Purpose of the Study:

• To develop and evaluate ML models for predicting SM using optical coherence tomography (OCT) imaging, laboratory tests, and clinical characteristics.
• To identify key predictors of SM through feature selection.

Main Methods:

• A study of 337 patients undergoing PCI and coronary OCT was conducted.
• Five ML models (XGBoost, LR, RF, SVM, NB) were developed and optimized using selected clinical and OCT imaging features via Lasso regression.
• Model performance was assessed using ROC curves and calibration accuracy.

Main Results:

• XGBoost and SVM models showed high predictive performance based on calcification features.
• Lasso regression identified five key features: calcification length, age, coronary dissection, lipid angle, and troponin.
• Optimized ML models, particularly XGBoost, demonstrated improved AUC values and high calibration accuracy, with calcification length, age, coronary dissection, lipid angle, and troponin being the most influential factors.

Conclusions:

• ML models integrating plaque imaging and clinical data can accurately predict SM.
• Models incorporating both clinical and OCT imaging features achieve superior performance in predicting stent malappposition.