Application of higher-order spectra for the characterization of Coronary artery disease using electrocardiogram signals. (January 2017)

• Record Type: Journal Article
• Authors: Acharya, U. Rajendra; Sudarshan, Vidya K.; Koh, Joel E.W.; Martis, Roshan Joy; Tan, Jen Hong; Oh, Shu Lih; Muhammad, Adam; Hagiwara, Yuki; Mookiah, Muthu Rama Krishanan; Chua, Kok Poo; Chua, Chua K; Tan, Ru San
• Is Part Of: Biomedical signal processing and control. Volume 31(2017)
• Page Start: 31
• Page End: 43
• Publication Date: 2017-01
• Subjects: Coronary artery disease -- Electrocardiogram -- Higher-order-spectra -- Cumulant -- Bispectrum; 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.2016.07.003
• Languages: English
• ISSNs: 1746-8094

Abstract

Graphical abstract: Highlights: Characterization of CAD and normal is proposed using ECG signals. 137587 normal and 44426 CAD ECG beats are used. HOS bispectrum and cumulant are extracted from ECG beats. PCA and different ranking methods are used. Obtained classification accuracy of more than 98%. Abstract: Coronary Artery Disease (CAD) is the root cause for chain of catastrophic heart diseases such as Ischemic Heart Disease (IHD), Myocardial Infarction (MI) or Heart Attack (HA) and Heart Failure (HF). Early detection and treatment of this CAD condition is essential and may help in preventing it from progressing further. However, faster and accurate identification of CAD from Electrocardiogram (ECG) signals using manual interpretations is not an easy task to achieve. Thus, computer-aided techniques are necessary for the automated characterization of CAD condition. Therefore, this work proposes application of Higher-Order Statistics and Spectra (HOS) for an automated classification of normal and CAD conditions using ECG signals. In this paper, 182, 013 beats (137, 587 normal beats and 44, 426 beats with CAD) ECG beats are used. HOS bispectrum and cumulant features are extracted from each ECG beat. The features extracted are applied to Principal Component Analysis (PCA) dimension reduction technique. Then PCA coefficients are ranked using Bhattacharyya method, entropy, fuzzy Max-Relevancy and Min-Redundancy (mRMR), Receiver Operating Characteristics (ROC), t -test, Wilcoxon ranking methods. All ranked features are subjected to k-Nearest Neighbors (KNN) and Decision Tree (DT) classifiers to obtain the highest classification performance. The proposed methodology has achieved 98.17% accuracy, 94.57% sensitivity, and 99.34% specificity, using KNN classifier using 13 bispectrum features. Similarly, we have obtained 98.99% average accuracy, 97.75% sensitivity, and 99.39% specificity using DT classifier with 31 cumulant features. In addition, we have formulated and developed an integrated index called Coronary Artery Disease Index (CADI) for automated characterization of normal and ECG signals with CAD condition using a single number. This proposed CADI works efficiently to discriminate normal and CAD ECG classes for the any dataset with priory knowledge of the database.