Application of neural networks to separate interferences and ECG signals

The analysis of the electrocardiogram (EGG) is not an easy job when muscle signals and other noisy signals corrupt the ECG hiding important information. These interferences normally overlap the heart signal spectrum; hence analog or digital filtering implies changes in ECG morphology that can affect the diagnostic result. In this work, an artificial neural network (ANN) which separates the interferences from the ECG is presented. This separation is based on the independence of these signals. The ANN combines the ECGs acquired simultaneously at different positions of the body surface, and is trained in order to minimize the averaged mutual information between its outputs. The ANN output signals are the ECG independent components.