Ventricular arrhythmias classification based on deep learning algorithm

Abstract

Cardiac arrhythmia is a group of conditions in which the heartbeat is irregular, where it can be too fast or too slow. This happens when electrical impulses that coordinate the heartbeat fail to work in a correct manner. Some of these diseases show no symptoms, but ECG (electrocardiogram) can help to diagnose as it extracts the rhythmic information of the heart and heartbeat. This information is important, and it can distinguish the cardiac condition of the patient. Some of these diseases can cause serious condition to the patient if not treated immediately, for instance, Ventricular Fibrillation can result in loss of consciousness and even death in the matter of minutes. Due to limitation in the availability of doctors or cardiologist, machine can help to perform ECG interpretation task. Deep Learning (DL), which is a subset of Machine Learning (ML), does not require human intervention as the nested layers in the neural networks put data through hierarchies of different concepts, which eventually learn through their own errors, is suitable to perform such task. This project aimed to have ECG data extraction to classify the cardiac arrhythmias using deep learning approach for Premature Atrial Contraction, Atrial Tachycardia, Atrial Flutter, Atrial Fibrillation, Premature Ventricular Contraction, Ventricular Tachycardia, Ventricular Fibrillation and Normal Sinus Rhythm. In this project, the final classifying model has achieved an average accuracy of 94.52 across 6 cardiac arrhythmias. All ECG information will be selected from a few databases such as MIT-BIH arrhythmia database, Creighton University Ventricular Tachyarrhythmia Database, Intracardiac Atrial Fibrillation Database, Long-Term AF Database, MIT-BIH Atrial Fibrillation Database, MIT-BIH Normal Sinus Rhythm Database and MIT-BIH Supraventricular Arrhythmia Database. All these databases have annotated ECG files by cardiologist annotators. These ECG data will have to go through pre-processing to remove noises such as the baseline wanders and powerline interference. After that, the ECG data will be broken down into segments of PQRST where it will serve as the input data for the deep learning model. Step segmentation process and CNN deep learning are both done in Python with TensorFlow package for deep learning model and SciPy and NumPy packages for signal processing.