Application of Dual-Q TQWT for Atrial Fibrillation Detection with Three-Layered Neural Network

T. Pander (Silesian Univ. of Techn., Poland)

Atrial fibrillation is a very serious heart disease that should be detected as early as possible. In the approach presented here, an ECG signal is used, involving the detection of QRS complexes and then partitioning the ECG signal into segments containing 20 QRS complexes. In a subsequent step, this single signal segment is transformed using a Dual-Q Tuneable Q-factor Wavelet Transform. On this basis, the energy distributions in the frequency subbands for the high and low Q-factor resonance components are calculated. This allows the generation of two fixed-length vectors characterising the analysed ECG signal segment, which are fed to the input of the three-layer neural network. The presence of atrial fibrillation in the analysed ECG signal fragment alters the energy distributions in these components. An AF database from physionet.org containing 23 long-term ECG signals was used for the study, but the database only contains 23 signals. A classifier designed on the artificial neural network was then trained and tested. Tests carried out using the 5-fold cross-validation method resulted in Sen=99.02% and Prec=99.11%, among others, which compares very well with the results of reference methods.