MACHINE LEARNING IN CORONARY ARTERY DISEASE DETECTION: A TECHNICAL-SCIENTIFIC REVIEW

Cardiovascular diseases remain the leading cause of death worldwide, making advances in prevention essential. This review summarizes recent work on machine learning (ML) applied to structured clinical data for detecting or predicting coronary artery disease (CAD). This narrative review was conducted under the PRISMA 2020 framework. Searches in PubMed, IEEE Xplore, and SciELO (Jan 2020 to Apr 2025) yielded 3,780 records. After screening and full-text appraisal, 10 papers were included: seven primary studies and three reviews. Sample sizes ranged from 303 to 70,000 individuals. Tree based algorithms and ensembles posted the best scores, with accuracy between 0.82 and 0.99 and AUROC from 0.86 to 0.96. Explainability with SHAP was applied in four studies, and one paired SHAP with LIME. One paper added a cardiologist’s input to the decision loop, raising accuracy from 0.7829 to 0.8302. Only one article evaluated their models on external datasets and noted performance drops. Calibration was rarely addressed. Just one investigation reported a Brier score of 0.14 and a slope of 0.93. ML models trained solely on routinely collected demographic, laboratory, and clinical variables show strong classification ability for CAD, supporting use as a non-invasive screening aid and decision support. Prospective trials, external validation, and detailed calibration reports are required before clinical adoption.