Automated Acute Myocardial Infarction Detection Using Machine Learning From Phonocardiogram

Early detection of myocardial infarction (MI) is critical due to rapid heart muscle deterioration. Current diagnosis relies on electrocardiogram (ECG) and troponin tests, but troponin tests are accurate only after 12-24 hours. Therefore, we propose much earlier detection method combining phonocardiograms (PCG) with machine learning techniques. In our method, the PCG signal is recorded for 30 seconds using an electronic stethoscope, filtered, and included in the pre-double selection feature model by integrating the mutual information method and classification model with a tuning parameter to enhance the optimal results. The study identified 18 crucial features from PCG signals for MI classification. Then, we achieved outstanding high-performance metrics: accuracy of 96.3%, precision of 95.3%, sensitivity of 95.7%, and F1-score of 95.6%, which showed promise as biomarkers for MI classification. Our method demonstrates the potential of using PCG signals combined with machine learning for early and accurate classification of MI types, offering a non-invasive and potentially rapid diagnostic approach compared to current methods.