Automated Classification of Myocardial Infarction Based on Auscultation Position Using Random Forest

Myocardial Infarction is a disease requiring immediate treatment. ECG examination has the disadvantage of diagnosing MI signals. This research used phonocardiogram signals to classify myocardial infarction, specifically STEMI, and NSTEMI based on auscultation position. Signals were acquired from four auscultation positions: APEX, LLSB, LUSB, and RUSB for 30 seconds. The filtered signal is segmented each cycle using Shannon Energy (SE). The feature extraction obtains 12 time-frequency-statistic matrices from the segmented cycles. Normalized z-score used to evaluate all feature values. A random forest was applied to classify normal and abnormal signals. The findings indicated that the proposed approach has the best performance accuracy of 86 %, precision of 84 %, sensitivity of 85%, and F1 score of 84% at the LUSB (Pulmonary) position. This is in accordance with previous research observations that there was a pansystolic murmur in MI patients with optimal audibility at the LUSB. These findings can be used for reference examination of patients with pathological symptoms of MI. Our future research will improve the performance by focusing on features with low redundancy and essential information for each signal feature.