Telecardiology under Resource Constraint: Low:complexity Compact Representation of ECG

Abstract

Compact, yet faithful, representation of ECG signals to meet bandwidth and power constraints remains central to successful telecardiology, primarily in infrastructuredeficient areas. Towards practical realization, we seek desired compactness in the class of low-complexity transform representations. A typical ECG signal consists of a strong rhythmic (low-pass) component, with compact Fourier transform representation, and temporally localized (high-pass) features, efficiently represented by wavelet transform. Accordingly, we propose a compact representation consisting of suitable Fourier and wavelet coefficients. As computation of such coefficients is at most O(n log n), the proposed representation inherits the desired low complexity. Our method achieves targeted representation accuracy using fewer transform coefficients compared to the well-known fixed transform representations. IntroductionElectrocardiogram (ECG) is an indispensable tool in monitoring and management of cardiac health. ECG records are increasingly being maintained in electronic form, and machines directly producing digitized ECG signals are now commonplace [1]. Very often, such signals are stored locally, or transmitted to a remote location, respectively, for decision making at a later point, or at a distance [2]. Accordingly, with a view to minimizing storage requirement and/or communication bandwidth, one desires to represent ECG signals as compactly as possible without adversely affecting eventual clinical interpretation.Interestingly, the problem of practical ECG compression has been studied since several decades. Efficacy of a compression algorithm depends on signal sparsity. In this context, various researchers have reported ECG signals to be sparse in wavelet bases, and in particular "Daubechies 4" (db4) wavelet basis [3,4]. In the process, various researchers observed signal sparsity in wavelet and related domains, and demonstrated the respective efficacy of discrete cosine transform (DCT) SPIHT (set partitioning in hierarchical trees) algorithm [7]. Signal sparsity is also central to compressed sensing of ECG signals [3]. Moreover, in denoising applications, a sparser representation allows more coefficients to fall below a threshold thus allowing more noise to be removed [8].[In an earlier work, we proposed a Hybrid Fourier/wavelet technique for ECG signal approximation [9]. The method was demonstrated on signals derived from ANSI/AAMI EC13 dataset [10]. In this paper, we present a comparative study of ECG signal representation using Fourier transform (FT), wavelet transform (WT), discrete cosine transform (DCT) and the proposed hybrid method. The signals for the experimental dataset were picked from MIT-BIH Arrhythmia database [10]. Compared to the DCT, FT and WT representations, the proposed method saves of 46.15%, 25% and 4.54% coefficients on average, for certain target representation accuracy, respectively.The rest of the paper is organized as follows. We begin with describing theoretical background and the key idea in Sec. ...