JSM-2 Based ECG Compression with Statistical Support Prediction Sucheng Yu, Bin Liu, Wei Qiao and Chi Zhang School of Information Science and Technology University of Science and Technology of China Hefei, Anhui, 230027, P. R. China Email: ysc@mail.ustc.edu.cn, flowice@ustc.edu.cn Chang Wen Chen University at Buffalo, the State University of New York Buffalo, NY, 14260-2000 USA Email: chencw@buffalo.edu Jian Cai Institute of Microelectronics of the Chinese Academy of Sciences Beijing, 100029, P. R. China Email: caijian@ime.ac.cn Abstract—This paper addresses the problem of developing an efficient compression scheme with high quality and low computational complexity for ECG signal compression. Taking into account the joint sparsity existing in ECG data and the temporal dependencies in ECG signal sequence, a novel scheme for JSM-2 based ECG compression is developed to exploit these characteristics. We first predict support information in sparse domain from the previous ECG data for the current recovery process. Then a modified Simultaneous Orthogonal Matching Pursuit Algorithm (SOMP) algorithm is proposed to incorporate the idea of support information establishment for JSM-2 based ECG compression. Simulation results show that the proposed JSM-2 based ECG compression scheme with statistical support prediction outperforms existing schemes with enhanced performance and low computational complexity. I. I NTRODUCTION The electrocardiogram (ECG) data is an important class of bio-signals for diseases diagnostics. For contemporary health-care applications in wireless body area networks (WBANs), long term ECG recordings need to be stored and transmitted to reflect the patient’s health states. However, due to the limitation of storage space and transmission bandwidth for mobile wearable sensors, it is necessary to compress ECG data. In recent years, compressed sensing (CS) has been developed for simultaneous sampling and compression. It can reduce a large amount of data for transmission and reconstruct original signals from much fewer samples than suggested by conventional sampling theory. The work by Candes et al. [1] [2] and Donoho [3] has built the foundation of this novel sampling scheme for a variety of applications. Much works have been carried out to show the potential of exploiting CS method in ECG compression. The work in [4] presented a complete system-level comparison between a CS-based ECG algorithm and the state-of-the-art discrete- wavelet-transform(DWT)-based embedded ECG compression algorithms. It is found that the CS-based compression achieved the best energy efficiency due to its low complexity in the encoder which makes it more suitable for WBAN platforms. In [5] [6], the authors presented an ECG compression method based on specific dictionaries and compressed sensing. By establishing and using overcomplete training dictionaries of ECG beats associated with normal and several cardiac diseases, the ECG data can be accurately represented and properly reconstructed. However, they model the input signals as a sequence of separate heartbeats, and do not consider the correlations between heartbeats. Recently, distributed compressed sensing (DCS) [7] [8] has been adopted in ECG compression to utilize the correlations between heartbeats. DCS approach is based on the assumption that there exists joint sparsity in signal ensemble, which means that there exists common support in sparse domain between signals, and it can be used to improve the reconstruction performance. Here the support is the set of indices that indicates the positions of the important coefficients in sparse domain. Once the support information is obtained, it could orientate the reconstruction algorithm towards a proper result. In [7], Duarte et al proposed three models for different scenarios to deal with the joint sparsity. Specifically, JSM-2 model assumes that all signals can be present from the same support with different coefficients. This model can be properly adopted for ECG compressed sensing to exploit both intra- and inter-signal correlation structures. In [9], Carrillo et al showed that an accurate support can greatly improve the performance of the reconstruction. Polania et al proposed a JSM-2 based compression method [10] for single-lead ECG data, which focuses on the joint sparsity of ECG signal and exploits the common support between adjacent heartbeats. They assume that the positions of the first low-pass wavelet subband are the partial known support (PKS), i.e., they assume all low-pass wavelet elements are important coefficients, and this assumption of PKS is utilized in joint reconstruction. However, such support, which is established based on empirical evidence, is not accurate. In [11], the authors presented a more accurate way to establish PKS and exploited these PKS into joint reconstruction to improve the performance under low compression ratio (CR). However, the performance degrades fast when CR is high since the established support is no more accurate when the sampling number is low. Also, the computational cost of the algorithm to obtain the accurate PKS is high. In this work, we propose a novel ECG signal compression scheme, which considers not only the joint sparsity existing in ECG data, but also the temporal dependency in ECG signal sequence. Here the temporal dependency is drawn out from the fact that the support of heartbeats signal in sparse domain changes slowly. With this property, a statistical support prediction (SSP) method is used in the proposed scheme to provide more accurate support information. The statistical support information (SSI), which is defined as the probability of each index occuring in support, can be obtained from the previous reconstructed heartbeats and exploited into the reconstruction of the next heartbeats. We also propose a new 2013 IEEE 15th International Conference on e-Health Networking, Applications and Services (Healthcom 2013) 978-1-4673-5801-9/13/$26.00 ©2013 IEEE 218