Efficient detection of myocardial infarction from single lead ECG signal. (July 2021)

• Record Type: Journal Article
• Authors: Fatimah, Binish; Singh, Pushpendra; Singhal, Amit; Pramanick, Dipro; S., Pranav; Pachori, Ram Bilas
• Is Part Of: Biomedical signal processing and control. Volume 68(2021)
• Publication Date: 2021-07
• Subjects: Electrocardiogram -- Fourier decomposition method -- Fourier intrinsic band functions -- Machine learning -- Myocardial infraction; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2021.102678
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: According to WHO, heart diseases are number one cause for deaths globally. FDM provides an efficient decomposition of ECG signal and removes noise. Superior classification accuracy of 99.96% in detecting myocardial infarction. Performance comparison provided for four different classifiers. Proposed method holds a strong potential for real-time applications. Abstract: Myocardial infarction (MI) is a heart condition arising due to partial or complete blockage of blood flow to heart muscles. This can lead to permanent damage to the heart and can be fatal, if not detected early. In this work, we use single channel electrocardiogram (ECG) signal to develop two automated MI detection algorithms, namely, primary and modified. The primary algorithm is an ECG beat-based detection algorithm, while the modified algorithm considers frames of 4096 samples for detecting MI. Fourier decomposition method (FDM) is used to remove baseline wander and powerline interference, and then decompose ECG beats/frames into Fourier intrinsic band functions (FIBFs). Features including entropy, kurtosis, and energy are computed from each FIBF and relevant features are selected using the Kruskal–Wallis test. Various machine learning classifiers such as k-nearest neighbor (kNN), support vector machine (SVM), ensemble bagged trees and ensemble of subspace of kNN, are used to build the detection models. The best results are obtained for the primary algorithm with kNN classifier, where the accuracy obtained is 99.96%, sensitivity is 99.96% and selectivity is 99.95%. The modified algorithm, on the other hand, is computationally more efficient as it bypasses the beat extraction step and uses FDM only once for both noise removal and extraction of FIBFs. It achieves an accuracy of 99.65%, with 99.61% sensitivity and 99.73% selectivity. The proposed method performs better than the existing state-of-the-art techniques, and it has the potential for efficient real-time implementation in MI detection systems.