An efficient AR modelling-based electrocardiogram signal analysis for health informatics. (12th November 2021)

• Record Type: Journal Article
• Authors: Gupta, Varun; Mittal, Monika; Mittal, Vikas; Gupta, Anshu
• Is Part Of: International journal of medical engineering and informatics. Volume 14:Number 1(2022)
• Page Start: 74
• Page End: 89
• Publication Date: 2021-11-12
• Subjects: electrocardiogram -- ECG -- AR coefficients -- atrial tachycardia -- premature atrial contractions -- PAC -- KNN classifier -- PCA classifier -- Burg method -- Yule-Walker; 610.2805
• Languages: English
• ISSNs: 1755-0653

Abstract

Today, health informatics not only requires correct but also timely diagnosis much before the occurrence of critical stage of the underlying disease. Electrocardiogram (ECG) is one such non-invasive diagnostic tool to establish an efficient computer-aided diagnosis (CAD) system. In this paper, autoregressive (AR) modelling is proposed that is an efficient technique to process ECG signals by estimating its coefficients. In this paper, two parameters viz. atrial tachycardia (AT) and premature atrial contractions (PAC) are considered for evaluating the performance of the proposed methodology for a total of 17 recordings (6 real time and 11 from MIT-BIH arrhythmia database). As compared to K-nearest neighbour (KNN) and principal component analysis (PCA) with AR modelling [also known as Yule-Walker (YW) and Burg method], KNN classifier coupled with Burg method (i.e., Burg + KNN) yielded good results at model order 9. A sensitivity (Se ) of 99.95%, specificity (Sp or PPV) of 99.97%, detection error rate (DER) of 0.071%, accuracy (Acc) of 99.93% and mean time discrepancy (MTD) of 0.557 msec are obtained. Consistent higher values of all the performance parameters can lead to the development of an autonomous CAD tool for timely detection of heart diseases as required in health informatics.