Cross subject myocardial infarction detection from vectorcardiogram signals using binary harry hawks feature selection and ensemble classifiers

• Published in: IEEE access Vol. 12; p. 1
• Main Authors: Chaitanya, M Krishna, Sharma, Lakhan Dev
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 01.01.2024; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Binary Harry Hawks feature selection; Classification; Classification algorithms; Electrocardiography; ensemble classifier; Ensemble learning; Entropy; Feature extraction; Feature selection; Heart; Heart attacks; Machine learning; Myocardial infarction; myocardial infarction (MI); Myocardium; Occlusion; Orthogonality; Power lines; Reliability analysis; Spectral analysis; Spectrum analysis; Supervised learning; Support vector machines; Vectorcardiography (VCG)
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2024.3367597

Myocardial infarction (MI), widely referred to as a heart attack, is a leading reason for deaths worldwide. It is frequently caused by coronary artery occlusion, resulting in inadequate oxygen and blood supply, which damages the myocardial structure and function. Therefore, innovative diagnostic methods are required for reliable and timely identification of MI. The typical 12-lead electrocardiogram (ECG) technology causes patient discomfort and makes cardiac monitoring challenging. The frontal, sagittal, and transverse planes (3 orthogonal planes) are where vectorcardiogram (VCG) renders an edge over 12-lead ECG. This study, proposes a method for detecting MI utilising VCG signals of four seconds. Circulant singular spectrum analysis (CSSA) and four stage savitzky-golay (SG) filter were used in the filtering stage for the removal of power-line interference and base-line wander. The signal was time-invariantly decomposed using the CSSA, then features were extracted. The binary harry hawks-based feature selection method is employed on the extracted features to choose the optimal feature subspace which was followed by supervised machine learning based classification. The 10-fold cross validation, an even more practical leave-one-out (LOO) cross validation approach, and inter dataset cross validation (IDCV) were used to evaluate the reliability of the suggested method. Voting-based ensemble classification was used in LOO, IDCV validation, which improves the accuracy of this method. The proposed technique achieved an accuracy of 99.97%, 91.03%, and 99.41% for 10-fold, LOO cross validation, and IDCV, out-performing the state-of-the-art methods in the cross validation scenarios. The proposed technique results in an accurate detection of MI. Successful accomplishment of the LOO cross validation demonstrates the applicability and dependability of the suggested technique in the health care applications.