Réseaux de neurones appliqués à la modélisation et à la compression du signal ECG.

Abstract

The electrocardiogram (ECG) is the signal that represents time variations of the electric activity of the heart. It constitutes an effective tool for diagnostic of heart anomalies, for this reason we are in need for long recordings of ECG that reflect the state of the heart but this will cause problems of storing or transmission of the ECG to a distant interpretation center, this has motivated research works towards compression Recently, Neural Networks have occupied a great place in signal processing, specifically Recurrent Neural Networks because they are capable to adapt to time variations of non linear and non stationary signals such as the ECG signal. In this regard, a compression algorithm via parameters extraction using Recurrent Neural Networks has been developed and tested on electrocardiography signals of the “MIT-BIH Arrythmia Data Base” and results obtained are presented, discussed and compared with those of some most recent algorithms of ECG compression.