Atrial fibrillation recognizing using wavelet transform and artificial neural network confirmed by PCA and ICA

Cardiac arrhythmias detection is important because they determines the emergency conditions (risk of life). Special kind of arrhythmia: atrial fibrillation (AF) has its special pattern on the shape of ECG, it perturbs the electrocardiogram and in the same time complicates automatic detection of other kinds of arrhythmia. The problem has been described as a challenge by both Computers in Cardiology and PhysioNet. Authors have used a continues wavelet transform (WT) for data pre-processing to extract a classifier features allowing to differentiate AF from the sinus rhythm (SR). Next step was to determine structure and train an artificial neural network (ANN). The database of fibrillation, taken form PhysioNet, has been used to develop this study. Coupling of two methods: signal preprocessing by WT and analysis by ANN gives a structure of wavelet neural networks (WNN). As result of above described method the fully controlled correlation operation can be obtained. Additionally, the PCA and ICA methods have been used to confirm above described method.