DATA COMPRESSION OF ECG RHYTHMS

Abstract

Heart diseases, such as coronary heart disease, congestive heart failure, and congenital heart disease, are the leading cause of deaths for men and women all over the world. The continuous monitoring for twenty-four-hour is necessary to detect heart abnormalities for the critical cardiac patient. Therefore, recent developments in Electrocardiogram (ECG) signal processing, information technology and communication has brought a new dimension to the medical world. ECG is a quasi-periodic and non-stationary signal that is generated by the action of depolarization and repolarization of the cardiac cells. The analysis of ECG signal is very significant for feature extraction and interpretation of heart diseases. Timely diagnosis is an important factor in the treatment of cardiac abnormalities; however, it is not possible for every cardiac patient. Therefore, incorporation of wireless communication technology in the field of telemedicine, especially tele-cardiological systems has played an incredible role in the timely monitoring of a heart patient. However, these technologies are resource-constrained applications with limited bandwidth and power. Transmission /transfer of large amount of physiological data to healthcare /e-healthcare centers is very expensive. Therefore, these systems require efficient data reduction before transmission. In many cases, cardiologists need observation of the continuous heart activity of the patient after release from the hospital. In these cases, ambulatory monitoring or Holter monitoring system plays an important role. In these systems, data recording and storing is done for twenty-four-hour. For effective data storing, data size must be as less as possible. Therefore, ECG data compression is an essential part of these types of devices. As a result, ECG data compression is an important area of research for the last few decades. Data compression is the process of converting the bits structure of data in such a way that it consumes less space on disk. The major aim of every data compression method is to accomplish maximum data volume reduction while preserving the important features of the signal morphology during reconstruction. Data compression techniques can be divided into four major types, direct data compression (DDC), transform based data compression, feature extraction based data compression, and two dimensional (2D) data compression. In DDC techniques, data is compressed in time domain, therefore, these methods do not provide efficient results in terms of compression ratio (CR) and signal reconstruction quality. In feature extraction ii based data compression techniques, useful features are extracted and used for reconstructing the signal. These methods provide a good CR; however, the signal reconstruction quality is very poor. In 2D data compression techniques first, 1D ECG signal is converted into 2D, then 2D data compression techniques such as joint photographic experts group (JPEG) and set partitioning in the hierarchical trees (SPHIT) are employed for data reduction. These techniques provide good results in terms of CR and signal reconstruction quality both. However, conversion of signal from 1D to 2D is a time taking and complex process for the non-stationary signals such as ECG, and hence these techniques cannot be used for real-time processing. Therefore, for the last two decades, several studies have been carried out to compress the ECG rhythms, in which researchers have used transform based techniques. are generally used to compress the data compression of ECG rhythms. In these techniques, data is transformed from time domain to frequency domain. Among all transform based techniques, wavelet and filter bank based techniques are preferred, because of multi-resolution property of them. Therefore, the aim of this thesis is to develop an ECG signal compression algorithm that has a high compression ratio while guaranteeing signal quality. A wavelet filter bank is a tree-structured filter bank that decomposes the signal into sub-bands, and the power of time and frequency based parallel signal processing is exploited. Modulated filter banks (MFBs) is a cost-effective way to design and implement filter banks. Amongst all modulated filter banks, cosine modulated filter banks (CMFBs) provide the best results in terms of computational efficiency with small number of optimization parameters. Computational efficiency can be improved further by using computationally efficient prototype filter. An interpolated finite impulse response (IFIR) filter is a highly efficient filter in terms of computational complexity. Hence, in the present research work, the mammoth task of signal compression has been accomplished using computationally efficient CMFB, which also provides low implementation cost, higher compression ratio, and good signal reconstruction quality. Noise elimination is the first step of this work, because feature extraction and comparison is utilized here to examine the data decompression performance, and presence of noise may lead to false feature extraction. Therefore, in this work, elimination of noise is carried out using IFIR and frequency response masking (FRM) techniques. This filtering technique provides linear phase, inherently stable output with low computational complexity, which are the important factors for any signal processing. iii The design of nearly perfect reconstructed 4, 8, 16 and 32 channel uniform and 3, 4 and 5 channel non-uniform cosine modulated filter banks are done. A nearly perfect filter bank suffers from three types of distortions; a) amplitude distortion, b) aliasing distortion and c) phase distortion. Phase distortion can be eliminated completely using linear phase filters (FIR or IFIR), aliasing distortion and amplitude distortion can be minimized by suitable optimization technique. Here, for designing the filter banks, three approaches are used such as: a) a simple and proficient linear iterative technique, b) Schittkowski algorithm, and c) passband edge iteration to minimize the cost function. In technique (a), the cut-off frequency of the model filter is optimized to satisfy the perfect reconstruction condition in CMFB. Different fixed (Blackman, rectangular, Bartlett, Hanning and hamming) and adjustable (Kaiser, Dolph-Chebyshev, Saramaki, ultraspherical, symmetric exponential, symmetric hyperbolic cosine, symmetric Nuttall, extended Norton-beer, modified Kaiser, Gaussian and Taylor) window functions are used for designing the linear phase IFIR prototype filter for the CMFB. 8-channel uniform and 5 channel non-uniform filter bank are used for feature extraction and data compression, respectively. The non-uniform CMFB /QMF /WT is used for ECG data compression by decomposing the ECG signal into various frequency bands. Subsequently, thresholding is applied for truncating the insignificant coefficients. The estimation of threshold value is performed by examining the significant energy of each band. Encodings (RLE /Huffman /LZW) are utilized for the compression. In this work, performance is measured in terms of CR and signal reconstruction quality (peak mean square difference (PRD), signal to noise ratio (SNR), mean square error (MSE), mean error (ME), peak mean square difference normalized (PRDN) and quality score (QS). And also signal reconstruction quality is measured by extracting and comparing the features of both signals (original signal before the compression and reconstructed signal after decompression). R-peak extraction through wavelet transforms using bi-orthogonal mother wavelet and 8-channel uniform CMFB is done. The extraction of other features, viz., Q waves, S waves, P waves, T waves, P wave onset & offset points, T wave onset & offset points, QRS onset and offset points are identified using rule-based algorithms developed for the study. In this work, weighted diagnostic distortion (WDD) a measure of signal reconstruction quality and accuracy, has been used. In addition to eighteen, two new features (number of P waves and ventricular late potential) are used to compute the value of WDD. The tabular results show that the filter iv banks provide better results as compared to other related work (in terms of computational complexity) without affecting the other fidelity parameters. The data compression performance of the method based on CMFB provides efficient results in terms of CR and preserving diagnostic information (WDD). Finally, it can be stated that the work contributes significantly to the area of ECG data compression techniques. The developed methods are very much useful for telemedicine, especially in telecardiology. The overall work done in this thesis may be considered a positive and significant contribution for effective healthcare services to remotely located patients. In the current open society and with the growth of human rights, people are more concerned about the privacy of their information and other important data. Water marking and compression of ECG data can protect the individual identification and information. An ECG signal cannot be only used to analyse disease, but also to provide biometric information for identification and authentication. Integrating ECG water marking and compression approach can be taken as an area for future study, as ECG water marking can ensure the confidentiality and reliability of users’ data while reducing the amount of data.