Abstract:
The main objective of this research is the automatic detection of life-threatening Cardiac
Arrhythmia and its types accurately based on electrocardiogram (ECG) analysis for proper
treatment in order to save a life. The methodology adopted in the paper includes ECG Signal
Preprocessing, Heartbeat Segmentation, Feature Extraction, and for automatic detection and
decision-making, machine learning algorithms are applied. Prepossessing involved removal of
Noise by applying various methods including Low-Pass filter, Band-Pass filter, and wavelet
transforms. Morphological and dynamic features were used to classify various types of
arrhythmias. RR interval information, extracted from the fiducial points (QRS complex, R-peaks,
etc.) that were computed by Pan Tompkins’s method in heartbeat segmentation, were used as
dynamic features. Discrete wavelets transform (DWT) was then applied on each heartbeat and
each sub-band of DWT is dimensionality reduced using independent component analysis (ICA),
resulting in the selection of twelve coefficients as morphological characteristics. In addition, the
Teager energy operator (TEO) was utilized to capture nonlinear dynamics, which improves the
arrhythmia classification. These hybrid features are then combined and fed to a support vector
machine (SVM) to classify arrhythmia. The method proposed was tested over MIT BIH
Arrhythmia Database to train the system which is consists of 116,137 numbers of QRS complexes.
The different beats included Normal beat (N), Right bundle branch block (RBBB), premature
ventricular complex (PVC), atrial premature beats (APB), and Left bundle branch block (LBBB)
of recorded signals. Using the proposed technique, improved average accuracy of 97.24 % was
obtained and Model Overall Accuracy of 99.74 % was achieved using python and 99.4% using
Cubic SVM implemented in MATLAB Classification Learner App.
