Deep learning approach for MI detection using SVM-based classifier

Incorporation of automated electrocardiogram (ECG) analysis techniques in home monitoring applications can ensure early detection of myocardial infarction (MI), thus reducing the risk of mortality. The computerized identification and tracking of irregularities from multi-lead electrocardiograms (MECGs) is a common pattern recognition issue that has been studied in recent years. Ubiquity and feature minimization are the key challenges. The main objective of the proposed study is to demonstrate increased classification precision for myocardial infarction (MI) identification from MECG by fusing a support vector machine (SVM) with a deep autoencoder (DAE). The validation of the suggested method was conducted with 100 numbers of ECG signal/data from Physionet, including three major classes (anterior, inferior and posterior) of myocardial infarction (MI). Both the normal and MI ECG beats were fed to feature extraction tool as input and a minimized set of 60 output attributes (/features) was fed to the binary classifier. This proposed classification method used ten-fold cross validation. The proposed classification technique achieved 97.3% and 97.12% of average sensitivity and positive predictivity respectively. The result of the proposed method is compared with other existing classification technique in terms of features and the sensitivity and data length.