An effective wavelet-based lossy compression of noisy ECG signals

Abstract

Noise degrades the rate-distortion performance of any electrocardiogram (ECG) compression algorithm. In distortion driven lossy coding approach, the bit stream is truncated at the bit rate that corresponds to a guaranteed user defined distortion level measured using the percentage root mean square difference (PRD). In many lossy compression methods, noise filtering may be implicitly done when performing the thresholding or/and quantization. In such a case, noise decreases the compression ratio for the specified distortion level. In this paper, we propose an effective wavelet-based lossy compression of noisy ECG signals based on the set partitioning in hierarchical trees (SPIHT) coding algorithm and novel wavelet energy based weighted PRDs (WEWPRDs) criterion. The PRDs measures normalized root mean squared difference between wavelet subband coefficients of the original and compressed signals. The dynamic weights based on wavelet energy feature represent the actual contribution of the subbands that are used to discriminate different frequency subbands, particularly subbands corresponding to noise. The WEWPRDs criterion appears to be a correct representation of the amount of signal distortion at all subbands and robust to insignificant errors in some bands. Thus, this criterion leads to a better measure of the rate-distortion (R-D) performance of the coder. Experiments on several noisy records from the widely used MIT-BIH arrhythmia (mita) database show that the proposed scheme outperforms PRD and wavelet based weighted PRD (WWPRD) measurement criteria based schemes.