Compression of Electrocardiogram Using Neural Networks and Wavelets

The real-time transmission of the electrocardiogram (ECG) in urgent situations can improve the chances of the patient. However, one of the greatest problems involving this kind of telemedicine application is the leakage of network bandwidth. ECG exams may generate too much data, which makes difficult to apply telecardiology systems in real life. This problem motivated many authors to look for efficient techniques of ECG compression, such as: transform approaches, 2-D approaches, similarity approaches and generic approaches. The present work proposes a new hypothesis: neural networks may be applied together with Wavelet transforms to compress the ECG. In this approach, the Wavelet transform acts as a pre-processor element for a multilayer perceptron neural network, trained with the backpropagation algorithm. The original signal was divided in two parts: the “plain” blocks and the “complex” ones. The “plain” blocks were compressed with a 40:1 ratio while the “complex” blocks were compressed with a 5:1 ratio. The use of both compressors guaranteed a compression rate of 28:1, approximately. The process obtained good grades in the quality aspect: percent root mean squared difference (1.846%), maximum error (0.1789) and standard derivation of errors (0.1044).