Third-order tensor based analysis of multilead ECG for classification of myocardial infarction. (January 2017)

• Record Type: Journal Article
• Authors: Padhy, Sibasankar; Dandapat, S.
• Is Part Of: Biomedical signal processing and control. Volume 31(2017)
• Page Start: 71
• Page End: 78
• Publication Date: 2017-01
• Subjects: Multilead ECG -- Myocardial infarction -- Higher-order SVD -- Mode-n singular values -- Support vector machine; 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.2016.07.007
• Languages: English
• ISSNs: 1746-8094

Abstract

Abstract : Highlights: We propose a novel method for detection and localization of myocardial infarction. MECG data is represented as a third-order. The feature vector is very small compared to the state-of-the-art methods. Algorithm outperforms during both detection and localization. Experiments are conducted without over-fitting the training and testing datasets. Abstract: Electrocardiogram (ECG) feature extraction and classification are challenging tasks for correct diagnosis of cardiac diseases. Conventional feature extraction methods were based on evaluating instances from a 2-D multilead ECG (MECG) data matrix. In this article, we proposed a novel method for detection and localization of myocardial infarction (MI) from the reduced MECG tensor. A third-order tensor structure was employed to represent the MECG data in three dimensions (leads × beats × samples). The higher-order singular value decomposition exploits intra-beat, inter-beat and inter-lead correlations of wavelet transformed MECG tensor. The mode- n singular values (MSVs) and the normalized multiscale wavelet energy (NMWE) of each subband tensor were considered as the mode features for detection and localization of MI. The support vector machine was used as the classifying technique. Datasets from the PTB database that comprise of healthy and different MI cases were considered for evaluation purpose. Experimental results showed that the proposed method assured a detection accuracy of 95.30% with sensitivity and specificity of 94.6% and 96.0%, respectively. The MSV and the NMWE features from the third-order MECG tensor were tested to be accurate in detecting and localizing MI.