Automated Identification of Myocardial Infarction Using Harmonic Phase Distribution Pattern of ECG Data

Abstract

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. Most of the published techniques use advanced signal processing tools, a huge number of ECG features, and complex classifiers, which make their hardware implementation difficult. This paper proposes the use of harmonic phase distribution pattern of the ECG data for MI identification. The morphological and temporal changes of the ECG waveform caused by the presence of MI are reflected in the phase distribution pattern of the Fourier harmonics. Two discriminative features, clearly reflecting these variations, are identified for each of the three standard ECG leads (II, III, and V2). Classification of the healthy and MI data is performed using a threshold-based classification rule and logistic regression. The proposed technique has achieved an average detection accuracy of 95.6% with sensitivity and specificity of 96.5% and 92.7%, respectively, for classifying all types of MI data from the Physionet Physikalisch-Technische Bundesanstalt diagnostic ECG database. The robustness of the algorithm is confirmed with real data as well. The algorithm is also implemented and validated on a microcontroller-based Arduino board, which can serve as a prototype ECG analysis device. Apart from providing comparable performance to other reported techniques, the proposed technique provides distinct advantages in terms of computational simplicity of the features, significantly reduced feature dimension, and use of simple linear classifiers which ensure faster and easier MI identification.