ECG DATA COMPRESSION USING ARTIFICIAL NEURAL NETWORKS

Abstract

Multi-channel electrocardiograms (ECGs) of high precision are the base and modem techniques for the non-invasive monitoring of cardiac conditions. The quality and accuracy of medical diagnosis is increasing due to improved probing techniques and instrumentation. At the same time, it also implies that vast amounts of data are generated. Usually the electrocardiogram needs to be stored as part of the clinical record or for further diagnosis or to be transmitted for real-time tele-diagnosis and monitoring. A recommended remedy of solving the costly economy of storing and transmitting these signals is to compress them. Even though Many ECG compression techniques have been developed before, most of them do not meet the requirements for clinical acceptability. This work intends to develop a compression system by adopting the application of Artificial Neural Network in dimension reduction. This system first pre-processes the ECG data using digital filters to improve the quality of the ECG signals. Then it adopts amplitude and first derivative algorithm to detect the R-points accurately. Finally the partitioned beats are trained using artificial neural network which adopts a robust learning algorithm called Resilient Backpropagation algorithm. Weights and hidden layer activations are stored to represent the original ECG data. Impermanent and highly aberrant ECG beats are stored uncompressed. The outcome of this result shows, using backpropagation artificial neural network based compression system is an efficient and effective way of compression ECG signals with high compression ratio, improved compression precision and less compression time. This technique also provides high reconstruction fidelity.