DATA COMPRESSION AND FEATURE EXTRACTION OF ECG SIGNALS

Abstract

In the present work a method has been developed both for data compression and feature extraction of ECG signals. Data compression is performed with the help of Error-Back-Propagation (EBP) Artificial neural Network (ANN). A comprehensive study is done, regarding the optimum configuration of the ANN, to achieve better data compression and signal retrieval, The ANN is trained separately, once for single, and then for two hidden layers. Also, the number of nodes in the hidden layers is varied during the process of training. Finally the best network structure (335-4-4-335) is chosen on the basis of results obtained. This network is now used for compression of 12 ECG lead data taken from CSE data base. Then feature extraction is performed on the data retrieved after compression, using the criteria of slope, amplitude and duration. In the end, a comparison is made of the features extracted for both the original and retrieved ECG signals..