A Real-Time Cardiac Arrhythmia Classifier

Abstract

Cardiovascular diseases (CVD) have increased drastically among Non-Communicable diseases, which have peaked over the past recent years. In 2018, around 17.9 million which is an estimated 31% of the people have died worldwide due to CVDs. A novel machine learning algorithm for continuous monitoring, identification and classification of cardiac arrhythmias from Electrocardiogram (ECG) data is presented here. The proposed solution has two stages where the first stage is a rule based cardiac abnormality identification which has an individual 97.55% ± 0.3% of accuracy (Acc) for a dataset of 705,000 and the second stage is a Neural Network (NN) based classification model which is trained and tested to identify 15 different classes recommended by ANSI/AAMI standard [1], and has 97.1% of individual accuracy for MIT-BIH Arrhythmia dataset [2] of 96265 beat samples. The combined real-time cardiac arrhythmia classifier is parallelized with CUDA in order to utilize the GPU and increase the execution speed by 4.86 times.