Automated interpretable detection of myocardial infarction fusing energy entropy and morphological features. (July 2019)

• Record Type: Journal Article
• Authors: Han, Chuang; Shi, Li
• Is Part Of: Computer methods and programs in biomedicine. Volume 175(2019)
• Page Start: 9
• Page End: 23
• Publication Date: 2019-07
• Subjects: Myocardial infarction -- MODWPT -- Energy entropy -- Morphological features -- Automated interpretable detection -- Inter-patient paradigm; Medicine -- Computer programs -- Periodicals; Biology -- Computer programs -- Periodicals; Computers -- Periodicals; Medicine -- Periodicals; Médecine -- Logiciels -- Périodiques; Biologie -- Logiciels -- Périodiques; Biology -- Computer programs; Medicine -- Computer programs; Periodicals; Electronic journals; 610.28
• DOI: 10.1016/j.cmpb.2019.03.012
• Languages: English
• ISSNs: 0169-2607

Abstract

Highlights: The feature extraction method based on MI detection performing a novel combination of global energy entropy features based on MODWPT and local morphological features is proposed. The method of energy entropy based on MODWPT not only enlarges the local characteristics via time-frequency analysis but also captures the small and short changes of 12 leads ECG based on the probability distribution of energy. Automated detection method has the characteristic of interpretability according with diagnostic logic and strategy of clinician and specific change of ECG for MI. The proposed work has more vital clinical significance based upon inter-patient paradigm. The quantization method has achieved superior results with few features compared to other detection methods. Abstract: Background and objective: The 12 leads electrocardiogram (ECG) is an effective tool to diagnose myocardial infarction (MI) on account of its inexpensive, noninvasive and convenient. Many methodologies have been widely adopted to detect it. However, much existing method did not integrate with diagnostic logic of clinician and practical application. The aim of the paper is to provide an automated interpretable detection method of myocardial infarction. Methods: The paper presents a novel method fusing energy entropy and morphological features for MI detection via 12 leads ECG. Specifically, ECG signals are firstly decomposed by maximal overlap discrete wavelet packet transform (MODWPT), then energy entropy is calculated from the decomposed coefficients as global features. Area, kurtosis coefficient, skewness coefficient and standard deviation extracted from QRS wave and ST-T segment of ECG beat are computed as local morphological features. Combining global features based on record and local features based on beat for single lead, all the 12 leads features are fused as the ultimate feature vector. What's more, different methods including principal component analysis (PCA), linear discriminant analysis (LDA) and locality preserving projection (LPP) are employed to reduce the computational complexity and redundant information. Meanwhile, principal component features are ranked by F-value. To evaluate the proposed method, PTB (Physikalisch-Technische Bundesanstalt) database and inter-patient paradigm are employed. Results: Compared with different algorithms, support vector machine (SVM) using radial basis kernel function combined with 10-fold cross validation achieves the best average performance with accuracy of 99.81%, sensitivity of 99.56%, precision of 99.74% and F1 of 99.70% based on 18 features in the intra-patient paradigm. By contrast, the accuracy is 92.69% with only 22 features for the inter-patient paradigm. Conclusions: The experimental results present a superior performance compared to the state-of-the-art method. Meanwhile, above approach has the characteristic of interpretability according with diagnostic logic and strategy of clinician and specific change of ECG for MI.