Logo image
Back
Predicting coronary artery stenosis using machine learning to reduce unnecessary angiographies
Journal article   Open access   Peer reviewed

Predicting coronary artery stenosis using machine learning to reduce unnecessary angiographies

Andreas Leiherer, Laura Schnetzer, Sylvia Mink, Axel Muendlein, Bernhard Bermeitinger, Christoph Saely, Peter Fraunberger and Heinz Drexel
Atherosclerosis, v 407, 119535
Aug 2025
DOI: https://doi.org/10.1016/j.atherosclerosis.2025.119535
url
https://doi.org/10.1016/j.atherosclerosis.2025.119535View
Published, Version of Record (VoR)Maybe Open Access (Publisher Bronze),  Open

Abstract

Background and Aims: Coronary angiography is the gold standard for diagnosing coronary artery disease (CAD), yet it is an invasive procedure with associated risks. Predicting angiographic outcomes with machine learning (ML) offers a potential solution to avoid unnecessary procedures. This study aimed to develop and evaluate a machine learning model for predicting angiographic findings, focusing on its ability to minimize false-negative predictions for significant coronary stenosis (≥50% lumen narrowing). Methods: This study analyzed clinical and laboratory data from 2,310 patients undergoing coronary angiography. Angiographic outcomes were classified as no stenosis (X0), non-significant stenosis (X1), or significant stenosis (X2). A total of 114 variables were preprocessed using k-nearest neighbors (kNN) imputation for missing data and min-max scaling for normalization. Data were split into training and test sets (75:25). To address class imbalance, oversampling was applied in the training set. XGBoost was identified as the best-performing algorithm following a comparison with alternative machine learning models. Hyperparameter optimization was conducted via grid search, and 5-fold cross-validation was employed to evaluate model performance. Metrics included sensitivity, precision, F1 score, and (balanced) accuracy. Results: The model achieved an overall accuracy of 62.9% (95% CI: 57.8–67.9). For significant stenosis (X2), it demonstrated a sensitivity of 94.7%, a precision of 62.7%, and an F1 score of 74.5% ensuring high reliability in identifying patients at risk. For no stenosis (X0), sensitivity was 37.3%, precision was 67.6% and the F1 score was 48.1%. Balanced accuracy for X2 and X0 was 60.0% and 66.7%, respectively, indicating robust performance across these critical categories. Conclusions: The XGBoost-based model reliably predicts angiographic findings, particularly excelling in identifying significant stenosis, thus reducing the risk of missed diagnoses. By accurately stratifying patients, this approach has the potential to optimize patient selection for and before coronary angiography and reduce unnecessary invasive procedures in clinical practice.

Metrics

1 Record Views

Details

Logo image