An empirical mode decomposition based detection theoretic strategy for T-wave alternans analysis. (September 2019)

• Record Type: Journal Article
• Authors: Bakhshi, Asim D.; Latif, Muhammad; Bashir, Sajid
• Is Part Of: Biomedical signal processing and control. Volume 54(2019)
• Publication Date: 2019-09
• Subjects: Empirical mode decomposition -- Matched-filter -- T-wave alternans -- ECG analysis; 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.2019.101609
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: All the suggested revisions have been incorporated. Paper has been revised thoroughly for minor errors of language and clarity. Editor's suggestions has also been complied with. Abstract: T-wave alternans (TWA) is a well-recognized marker for malignant cardiac arrhythmia which may lead to sudden cardiac arrest. Micro-volt TWA detection and estimation becomes a challenge under the presence of process noises, transient outliers and physiological artifacts. This paper proposes an empirical mode decomposition (EMD) based template matched-filter approach (EMF) for TWA analysis, targeting detection of distributed energy in beat-to-beat alternant waveform rather than peak amplitude. The template is dynamically generated through decomposition of alternant waveform into its intrinsic modes and subsequent reconstruction. Consequently, the study proposes six different variants of matched-filter based upon estimated reconstructions of intrinsic mode functions through spectral purity (EMFSPI ), detrended fluctuation analysis (EMFDFA ), and four different correlation based thresholding schemes (EMFCORR1/2/3/4 ). Performance evaluation of the proposed strategy and subsequently formulated variants has been done using realistically synthesized noisy data-sets against previously proposed median and mean based template matched-filtering detectors (TMFD). The classical spectral method (SM) has been used as a bench-marking algorithm. All six variants of EMF detectors outperform the previously proposed TMFD variants. The alternant signal and IMF cross-correlation based correlation strategy shows >2 dB improvement as compared to the benchmark. Among correlation based methods, the hard thresholding approach outperforms SM by >1 dB . Subsequent to detection under improved signal conditions, the estimator trends of all the proposed variants are generally comparable with heuristic approaches outperforming the correlation based matched filter estimators.