Detection and Analysis of Cardiac Arrhythmias from Heartbeat Classification

The Electrocardiogram (ECG), one of the biological signals, can be utilized to identify heart arrhythmias. Detecting a single irregular heartbeat that can occur alone or in repetition helps in discovering an arrhythmia. Early detection of arrhythmias and taking necessary precautions can help cure or prevent life-threatening arrhythmias. Depending on the shape and features of ECG, they are categorized into multiple arrhythmias and grouped as classes based on their threat level, such as Unknown Beats (Q), Supraventricular Ectopic Beat (SVEB), Fusion Beat (F), Ventricular Ectopic Beat (VEB) and Non-ectopic Beat (N). The openly accessible Massachusetts Institute of Technology-Beth Israel Hospital(MIT-BIH) database is considered in this paper. Three stages are suggested for detection. The first stage is pre-processing, which is done by the 1-Dimensional Wavelet Discrete Transform (1D-DWT) method. The second stage is feature extraction, carried out by the Empirical Mode Decomposition (EMD) method. Features now extracted are then fed for the classifiers. Deep Neural Network (DNN) is capable of automatically extracting features and analyzing data patterns, eliminating the need for complex signal processing. For the classification stage, the dataset considered has 20% test data and 80% trained data. The Deep Learning (DL) originated Convolutional Neural Network (CNN) is compared with K-Nearest Neighbor (KNN) algorithm, which is originated from Machine Learning (ML) for secondary confirmation. These classifiers achieved a Maximum Accuracy (MAAC) of 90.87%, Maximum Sensitivity (MASE) of 90.56%, and Maximum Specificity (MASP) of 91.18% with KNN, and a MAAC of 93.8%, MASE of 92.52%, and MASP of 95.08% with the CNN classifier.