Multiscale Energy Based Detection of Multifocal Atrial Tachycardia in ECG Signal

Abstract

Electrocardiogram (ECG) signals give lots of information about the health condition of a person especially the condition of the heart. In this thesis, we have proposed a technique for the detection of Multifocal Atrial Tachycardia (MAT) by using multiscale energy approach. MAT is often misdiagnosed with other diseases, thus an efficient detection technique is essential for the accurate detection of the disease. In this technique, pre-preprocessing of the signal, redundant feature extractions and classification of the disease are performed. The ECG signals are often corrupted with various artifacts such as power line interference, muscle noise, baseline wandering etc. Various filtering methods such as band pass filter, notch filter, wavelet filters are used for removing these artifacts. They are used according to the type of noises. Extraction of features is performed by using Daubechies 9 7 biorthogonal wavelet transform. The features selected are energy due to the wavelet coefficients and covariance of the wavelet coefficients. SVM classifier is used for the classification and its performance analysis is also done. Datasets having normal sinus rhythm and that of MAT from MGH Waveform ECG database are used for evaluation. We have implemented the proposed technique using MATLAB R2016a.