Prediction of coronary artery disease and evaluation of relative risk factors using artificial neural networks

AIM:To establish the hazard model of coronary artery disease (CAD) using artificial neural networks (ANN) and to evaluate the relative risk factors. METHODS: A retrospective case-control study was conducted in 265 patients diagnosed with CAD by coronary angiography (at least one coronary artery stenosis >50% in major epicardial arteries) and 102 subjects with normal coronary arteries were used as control. ANN models trained with different algorithms were performed in 367 records, divided into training (n=300) and testing (n=67) data sets randomly. The performance of prediction was evaluated by accuracy, sensitivity and specificity values based on standard definitions. RESULTS: The results demonstrated the ANN models trained with the 12 smallest mean-square-error algorithms were promising. Accuracy, sensitivity and specificity values varied, respectively, between 98.51 and 100%, 98.04 and 100% and 87.5 and 100% for testing. The best ANN model showed the value of 100% for accuracy, sensitivity and specificity. Using mean impact value of the ANN, total cholesterol, LDL cholesterol, HDL cholesterol and systolic blood pressure were found to be the most important risk factors for CAD. CONCLUSION: The proposed ANN models trained with the algorithms can be used as a promising approach for predicting CAD without the need for invasive diagnostic methods and for making prognostic clinical decisions.