ECG compression with retrieved quality guaranteed

Abstract

A filter bank-based algorithm for ECG compression is developed. The proposed method utilises a nearly-perfect reconstruction cosine modulated filter bank to split the incoming signals into several subband signals that are then quantised through thresholding and Huffman encoded. The advantage of the proposed method is that the threshold is chosen so that the quality of the retrieved signal is guaranteed. It is shown that the compression ratio achieved is an improvement over those obtained by previously reported thresholding-based algorithms.Introduction: Several wavelet-based ECG compression methods have recently been developed that report good performance [1][2][3]. Among the reported algorithms, those based on wavelet coefficient thresholding have been shown to be efficient and yield high compression ratios (CRs) [3]. In this Letter, we present an easy to use and efficient ECG compression scheme based on an M-channel maximally decimated filter bank with a parallel structure [4] that guarantees the quality of the retrieved signal. This method is based on that originally reported in [5]. The proposed method incorporates several innovations, including the quantisation of the subband signal samples with less resolution than the original signal samples and entropy-coding by means of a Huffman coder. As a result, the algorithm performs significantly better than other approaches developed using similar techniques.Tests are carried out using the MIT-BIH Arrhythmia Database and the percentage root-mean-square difference (PRD) measurement criteria to evaluate the quality of the retrieved signal. Accordingly, let x[n] and x[n] be the original and the reconstructed signals. The PRD is then defined as:

Optimal Selection of Wavelet Coefficients for Electrocardiograph Compression

“…Common criteria for performance testing are the compression ratio CR and the percentage root-mean-square difference PRD1 (independent from the mean value), defined as in [2]. The values of PRD1 between 0% and 9% guarantee good reconstructed signals.…”

“…The simplicity of the electronic implementation of the proposed HLNTM, optimized for a lower number of samples (512 samples), makes it suitable for portable applications with low memory resources and channel occupancy. Table 3 shows the comparison between the proposed method and the best methods reported in the literature in terms of error figures and compression rates [2], [3] using the same number of samples. Both are very similar in CR versus PRD1.…”

“…However, serious difficulties are encountered in attempting to reduce the channel costs and electronic resources. Several attempts have been made which partly solve the problem using compression algorithms [1], [2].…”

Performance Analysis of ECG Signal Compression using SPIHT

“…8, 10, 14, 18, 22, 26, 28, and 30. The second experiment was intended to compared the performance of proposed method with four other previous methods proposed by Alshamali[18], Benzid [25], Blanco [26], Lu [4]. For the second experiment, we used the second dataset, which includes the same records from MIT-BIH arrhythmia database as used in their experiment.…”

“…proposed by [18], [25], [26], and [4], respectively. The same fact occurred at the range of 14~16 compression rates, the mean of PRD of the proposed method is 2.82, while the other methods generate higher distortion, i.e.…”