Improved accuracy of PCG signal classification for myocardial infarction biomarker using automatic feature selection and boosting process

Abstract Myocardial Infarction (MI) is a leading cause of death worldwide. Traditional MI diagnosis relies on clinical evaluation, Electrocardiogram (ECG) findings, cardiac biomarkers, and imaging studies. This study introduces an innovative approach by applying machine learning to phonocardiogram (PCG) data, aiming to differentiate between normal, segment elevation myocardial infarction (STEMI) and non-ST segment elevation myocardial infarction (NSTEMI). Building upon previous research where a bagging method achieved 86% accuracy, this study enhances predictive performance using advanced boosting algorithms such as AdaBoost and Gradient Boosting. Our model, tested on 104 subjects of Indonesian and Japanese origin, utilized 18 extracted features with a mutual information value of 0.821 bits, demonstrating their high relevance. AdaBoost and Gradient Boosting achieved 88.3% and 93% accuracy without parameter tuning, respectively. With parameter tuning, these methods further improved, with AdaBoost reaching 94% accuracy and sensitivity, while Gradient Boosting achieved an accuracy of 98.3% and a sensitivity of 95%. These results highlight the model’s potential to revolutionize MI diagnosis using PCGs, showcasing the efficacy of machine learning in enhancing clinical decision-making and improving patient outcomes.