AUTOMATED DETECTION OF ATRIAL FLUTTER AND FIBRILLATION USING ECG SIGNALS IN WAVELET FRAMEWORK

In this paper, an electrocardiogram (ECG)-based pattern analysis methodology is presented for the detection of artrial utter and atrial fibrillation using fractal dimension (FD) of continuous wavelet transform (CWT) coefficients of raw ECG signals, sample entropy of heart beat interval time series, and mean heart beat interval features. Accurate diagnosis of atrial tachyarrhythmias is important, as they have different therapeutic options and clinical decisions. In view of this, we have made an attempt to develop a discrimination mechanism between artrial utter and atrial fibrillation. The methodology consists of mean heart beat interval detection using Pan Tompkins algorithm, calculation of sample entropy of heart beat interval time series, computation of box counting FD from CWT coefficients of raw ECG, statistical significance test, and subsequent pattern classification using different classifiers. Different wavelet basis functions like Daubechies-4, Daubechies-6, Symlet-2, Symlet-4, Symlet-6, Symlet-8, Coiflet-2, Coiflet-5, Biorthogonal-1.3, Biorthogonal-3.1, and Mayer wavelet have been used to compute CWT coefficients. Features are evaluated using statistical analysis and subsequently two-class pattern classification is done using unsupervised (k-means, fuzzy c-means, and Gaussian mixture model) and supervised (error back propagation neural network and support vector machine) techniques. In order to reduce the bias in choosing the training and testing set, k-fold cross validation is used. The obtained results are compared and discussed. It is found that the supervised classifiers provide higher accuracy in comparison to the set of unsupervised classifiers.