DATA COMPRESSION OF ECG SIGNAL USING NEURAL NETWORKS

Abstract

The continuing proliferation of computerized ECG processing systems along with the increased feature performance requirements and demand for lower cost medical care have mandated reliable, accurate, and more efficient ECG data compression techniques. The practical importance of ECG data compression has become evident in many aspects of computer, electro-cardioyrphy including : (a) increased storage capacity of ECG's as data bases for subsequent comparison or evalution, (b) economical rapid transmission of off-line ECG's over public phone lines to a remote interpretation center, and (c) improved functionality of ambulatory ECG monitors and recorders. In this work Artificial Neural Networks (ANN) have been used for compressing the ECG signal. But, before working with ECG signal, ANN was trained and tested for various sinusoidal signals, to have an idea regarding the performance of ANN. Multilayer Perceptron has been used in this work and the net is trained using back-propagation algorithm. Using neural nets for compressing, does not entail the reduction in the number of sample points for data compression and then to generate an ECG wave, but an inherent capability of artificial neural networks has been utilized. Firstly neural network has been trained for ECG signals. Then 'ECG signals to be compressed are presented to the trained net, on a cycle-to-cycle basis. The resulting activation levels of the hidden units and the interconnecting weights express the respective features of the waveform for each consecutive heart beat and are eventually used to reproduce the original waveforms. Therefore, in this method, instead of storing the entire ECG data sequence only the interconnecting weights of the network and the activation levels of the hidden units, for each consecutive heart beat are stored and hence data compression is accomplished. But this method is suitable only for Holter ECG's, as in the case of Holter ECG's only certain short-period segments of the 24 hour recording show abnormality relative to the large number of normal sinus ECG's. ECG signal used for this work is signal recorded at sampling frequency of 100 hz, at "Military Hospital", Roorkee. Software is written in "C" language and entire work was carried out on PC-AT-286.