A CNN Based Similarity Learning For Cardiac Arrhythmia Prediction

Abstract

Cardiovascular arrhythmia can be described as quickened, hindered, or irregular pulse brought by abnormalities caused by the muscular tissue of the heart. The Proposed Heartbeat classification technique uses deep two-dimensional Convolutional Neural Network (CNN) which gives extraordinary results in the field of pattern recognition and it is named as Cardiac Arrhythmia Prediction using ECG signals (CAPE). Each heartbeat was converted into a binary image. The converted binary image is used as the input data for the CNN classifier. Finally, the CNN classifier is enhanced to classify seven distinct kinds of ECG beats as pursues: Normal Beat (NOR), Paced Beat (PAB), Premature Ventricular Contraction Beat (PVC), Left Bundle Branch Block Beat (LBB), Right Bundle Branch Block Beat (RBB), Atrial Premature Contraction Beat (APC) and Ventricular Escape Beat (VEB). In order to achieve better accuracy, the developed CNN model incorporates optimization techniques such as data augmentation, batch normalization, dropout and Xavier initialization. The proposed model was compared with the popular CNN models: GoogleNet and VGGNet. In comparison, VGGNet takes lesser training time but gives higher accuracy. MITBIH arrhythmia database was used for the assessment of the classifier. The modified CNN classifier (VGGNet) with the converted ECG images accomplished a classification accuracy of 93.57% without any ECG pre-processing techniques such as noise filtering, feature extraction, and feature reduction.