Please use this identifier to cite or link to this item:
Title: ECG Data Compression for Telecardiology
Authors: Singh, Mandeep
Supervisor: Saxena, S. C.
Kumar, Vinod
Keywords: ECG;Data Compression;Telecardiology
Issue Date: 7-Aug-2008
Abstract: The electrical signal generated by heart and acquired from the body surface is known as Electrocardiogram (ECG). It is used to know the status of heart to diagnose its malfunctioning at an early stage, so that corrective action can be taken to prevent any major non-reversible failure. For critical cardiac patients, persons under cardiac surveillance, ambulatory patients and for creation of ECG database, continuous recording of ECG is required. The recorded data becomes so voluminous that it becomes practically impossible to handle it without compression. The importance of data compression further increases by the fact that the rate at which cardiac patients are increasing all over the world, we do not have a matching number of cardiologists to provide the required healthcare, especially in remote and rural areas. One way to overcome this problem is to transmit ECG, along with other vital statistics of a patient, over internet to a cardiologist for expert advice. For one day’s continuous recording, the amount of multichannel ECG data exceeds several gigabytes. Moreover if this data is to be transmitted over a telephone line or a slower digital communication network, the time of transmission goes beyond the human patience. Compressing the data is the only solution to this problem. The main goal of any compression technique is to achieve maximum data volume reduction while preserving the significant signal morphology on reconstruction. Our work starts with literature survey of the techniques used for compression of ECG signal, and identifies a wavelet compression method of Set Partitioning In Hierarchical Trees (SPIHT) as superior to any other technique reported so far. We have proposed two additional steps in the SPIHT algorithm, which are “Blank-fire removal” and “Polishing”. These additional steps increase the compression ratio and reduce the percentage root-mean-square difference, while retaining all features of the existing SPIHT algorithm. The performance of existing SPIHT has been compared with that of the modified SPIHT algorithm on the same database set i.e. ECG signals from MIT-BIH arrhythmia test base (mitdb) (sampling rate 360 per second), with the same wavelet filters and using the same distortion measure. Out of a total of 84 signals tested, 59 signals (70.2%) have shown an improvement in the compression ratio in the range of 3.24% - 20.22%, averaging to 7.95% improvement, while the remaining 25 signals (29.8%) have maintained the same compression ratio as given by the existing SPIHT algorithm. Not even a single case has shown deterioration in compression ratio. Improvement in distortion has been found in all 84 signals (100%) ranging from 3.26% to 19.23%, averaging to 9.99% reduction in PRD. To handle lengthy ECG records, a set of executable files has been developed in C++ environment. Data downloaded from the Long–Term ST Database (ltstdb) has been used to test the executable files. Some peculiar programming problems were encountered while encoding the compressed data bit streams to ASCII characters. Since ASCII character for decimal value 13 in MS-DOS, Windows, and various network standards, is used as part of the end-of-line mark, while ASCII character for decimal value 26 and 255 are used for marking end-of-text and end-of-file (EOF) respectively, they disrupt the normal file reading process. These characters are identified in the bit stream itself and are subsequently modified to circumvent this problem. The concepts of “fragmentation” and “looping” are used to handle long records. The number of refinement passes determines the extent to which a signal can be safely compressed using SPIHT algorithm. With every refinement pass, the distortion in the reconstructed signal decreases, but this also results in decrease in the compression ratio. A criterion for the number of refinement passes in SPIHT algorithm is therefore required to achieve optimal compression ratio, while retaining all clinically significant morphological features. A study carried out to find this number has revealed that for the same number of refinement passes, different signals give vastly different PRD. Thus recommending a fixed number of refinement passes is not possible. However, if we take into account the presence of “blank-fire” in the compression, a criterion can be evolved. We recommend five refinement passes in absence of “blank-fire”, and six in its presence. This has resulted in largely improved consistency in the distortion. The proposed refinement criterion has been validated by visual inspection of the signals by a physician, as well as by a statistical analysis over a larger database. Implementing the improved SPIHT algorithm, with the proposed refinement criterion over 42 sets of two-lead ECG signals, the original and the reconstructed signals were randomly presented to a physician, in sets of 12-15 pairs of signal. The interpretations of these ECGs were same for all 42 reconstructed signals (100%) as for the original ones. We have also proposed a new form of root-mean-square distortion measuring parameter called Dynamically Derived Percentage Root-mean-square Difference (DDPRD), which gives its numerical value proportional to the distortion measured by other existing parameters. If the original signal is scaled by any factor, or the baseline is shifted in any direction or a baseline wander with zero dc value is introduced in it, subsequent to reconstruction, the numerical value of distortion measured by DDPRD remains unaffected. While all other existing parameters fail to give the same numerical value when the ECG test signal is subjected to scaling/baseline shift/baseline wander, DDPRD shows immunity to the mentioned changes in original signal. Taking valuable suggestions from a panel of reputed physicians, a user-friendly, interactive and dedicated telecardiology based website has been designed. The website has been designed using HTML as the front end user GUI, MySQL for maintaining the related database, and PHP as the interface programming language. Password protected login to the website can be done as a manager or a client or an expert. The site has built-in security features against unauthorized intrusions in the site. The jobs are prioritized as per the requirement of the client and the transfer of credits from client’s account to expert’s account is done as per the priority, i.e. more for high priority jobs and vice or versa. Ensuring online availability of experts and making fiscal transactions from clients and to experts is responsibility of the manager. The manager also does new registrations as a client or as an expert after verifying their credentials. The website has been designed keeping in mind the practical requirements and convenience of both client and expert, with an objective of bridging the gap between the client and the expert. The concept of 24 hour availability of authentic experts in the required area of specialization like cardiac medicine and cardio thoracic surgery, facility to compress and decompress long records, prioritizing the jobs instead of having a general queue, enhanced security features, safe fiscal transaction through a manager and a platform for settling disputes, if any, are some of the features of the developed website, which are not available on general chat sites. It may be concluded that the present work has contributed significantly in the area of ECG data compression for telecardiology, offering an opportunity to deliver better healthcare services for entire population of the world.
Description: Doctor of Philosophy
Appears in Collections:Doctoral Theses@EIED

Files in This Item:
File Description SizeFormat 
T562.pdf50.15 MBAdobe PDFThumbnail

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.