WNC-ECGlet: Weighted non-convex minimization based reconstruction of compressively transmitted ECG using ECGlet. (March 2019)

• Record Type: Journal Article
• Authors: Ansari, Naushad; Gupta, Anubha
• Is Part Of: Biomedical signal processing and control. Volume 49(2019)
• Page Start: 1
• Page End: 13
• Publication Date: 2019-03
• Subjects: Compressed sensing -- Wavelet transform learning -- Sparse recovery -- Non-convex minimization; 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.2018.10.005
• Languages: English
• ISSNs: 1746-8094

Abstract

Abstract: Data compression is very important in wireless body area network (WBAN), where a huge amount of ECG data is continuously transmitted to the receiver. Recent research has shown that compressive sensing (CS) is a better alternative for ECG data compression over traditional compression methods. In this work, we provide a method to improve CS-based ECG reconstruction that can be directly implemented in WBAN technology and hence, may enhance the use of WBAN technology in healthcare informatics and telemedicine. The methodology is named as WNC-ECGlet and consists of the following key contributions: (1) non-convex minimization is proposed for use in CS based ECG reconstruction, (2) weighted sparsity on wavelet coefficients of ECG signals is proposed to minimize for reconstruction, and (3) wavelet transform learning is proposed for ECG signals, named hereby, as ECGlet. An algorithm is provided to solve weighted non-convex minimization (WNC) problem. ECGlet is learned from an ensemble of ECG signals in the lifting framework rendering learning to be convex having closed-form solution with compactly supported filters that can be easily implemented on hardware. The learned wavelet transform is used as the sparsifying transform in the CS-based reconstruction of ECG signals transmitted compressively to the receiver. Extensive experiments are performed with comparing results of WNC-ECGlet with standard wavelets, l 1 norm minimization and, Gaussian, Bernoulli and sparse binary sensing matrices on MIT-BIH Arrhythmia ECG dataset. The proposed WNC-ECGlet method is observed to perform better than conventional methods in CS-based ECG signal reconstruction.