A Fast and Robust Non-Sparse Signal Recovery Algorithm for Wearable ECG Telemonitoring Using ADMM-Based Block Sparse Bayesian Learning

Abstract

Wearable telemonitoring of electrocardiogram (ECG) based on wireless body Area networks (WBAN) is a promising approach in next-generation patient-centric telecardiology solutions. In order to guarantee long-term effective operation of monitoring systems, the power consumption of the sensors must be strictly limited. Compressed sensing (CS) is an effective method to alleviate this problem. However, ECG signals in WBAN are usually non-sparse, and most traditional compressed sensing recovery algorithms have difficulty recovering non-sparse signals. In this paper, we proposed a fast and robust non-sparse signal recovery algorithm for wearable ECG telemonitoring. In the proposed algorithm, the alternating direction method of multipliers (ADMM) is used to accelerate the speed of block sparse Bayesian learning (BSBL) framework. We used the famous MIT-BIH Arrhythmia Database, MIT-BIH Long-Term ECG Database and ECG datasets collected in our practical wearable ECG telemonitoring system to verify the performance of the proposed algorithm. The experimental results show that the proposed algorithm can directly recover ECG signals with a satisfactory accuracy in a time domain without a dictionary matrix. Due to acceleration by ADMM, the proposed algorithm has a fast speed, and also it is robust for different ECG datasets. These results suggest that the proposed algorithm is very promising for wearable ECG telemonitoring.

Keywords: alternating direction method of multipliers (ADMM); block sparse Bayesian learning (BSBL); compressed sensing (CS); electrocardiogram (ECG); wireless body area networks (WBAN).

Figure 1

Typical compressed sensing (CS)-based ECG wireless telemonitoring system.

Figure 2

Block diagram of a wearable ECG telemonitoring system.

Figure 3

Devices of a wearable ECG collecting system.

Figure 4

Recovery results of all 48 of the recordings in MIT-BIH Arrhythmia Database and seven recordings in MIT-BIH Long-Term ECG Database at different CRs (compression ratios). (a–d) show that the PRD (percentage root-mean squared distortion) and Pearson correlation of BSBL-ADMM are very close to BSBL-BO, the other three have much worse recovery accuracy; (e,f) show that the BSBL-ADMM is much faster compared with BSBL-BO. These results on MIT-BIH Databases demonstrate that the proposed BSBL-ADMM algorithm has both satisfactory speed and accuracy compared with the other algorithms.

Figure 5

Histogram of the difference between the PRD of BSBL-ADMM and the PRD of the other four algorithms at each frame over all the recordings in MIT-BIH Arrhythmia Database and MIT-BIH Long-Term ECG Database at the CR of 60%. It can be seen that the PRD of BSBL-ADMM is significantly smaller than BSBL-L1, BSBL-FM and AIHT. In addition, the PRD of BSBL-ADMM is larger than BSBL-BO, but the difference is relatively small and the recovery accuracy of the proposed BSBL-ADMM and BSBL-BO is very close.

Figure 6

Recovery results of eight randomly selected recordings in MIT-BIH Arrhythmia Database and all seven of the recordings in MIT-BIH Long-Term ECG Database. (a,b) show that the BSBL-ADMM has relatively stable recovery accuracy for different recordings; (c,d) show that the the BSBL-ADMM has relatively stable CPU times for different recordings. All these results on MIT-BIH Databases demonstrate the robustness of the proposed BSBL-ADMM algorithm.

Figure 7

Recovery results of all the recordings in practical wearable ECG datasets at different CRs (compression ratios). (a,b) show that the PRD (percentage root-mean squared distortion) and Pearson correlation of BSBL-ADMM is very close to BSBL-BO, the other two have much worse recovery accuracy; (c) shows that the BSBL-ADMM has much faster speed compared with BSBL-BO. These results on practical wearable ECG datasets demonstrate that the proposed BSBL-ADMM algorithm has both satisfactory speed and accuracy compared with the other algorithms.

Figure 8

Recovery results of two ECG signals selected from recording 3 and recording 7, respectively. It can be seen intuitively that the recovered ECG signals by BSBL-ADMM and BSBL-BO are very similar and have no obvious distortion compared with the original signal, but the recovered ECG signals by BSBL-L1, BSBL-FM and AIHT are obviously distorted.