Comparison of Performance of Artificial Intelligence Algorithms for Real-Time Atrial Fibrillation Detection using Instantaneous Heart Rate

Abstract

Atrial Fibrillation (AFib) is an abnormal heart rhythm (arrhythmia) condition that may cause a fatal cardioembolic stroke. The episode of AFib can be paroxysmal which increases challenges for its clinical manual diagnosis and affects the quality of life. Real-time cardiac monitoring with wearable health trackers can improve the chances of detecting this unpredictable event. In this paper, various Artificial Intelligence (AI) algorithms have been developed to classify beat-to-beat variation of AFib episodes in real-time using Instantaneous Heart Rates (IHR). Publicly-available clinical datasets from Physionet.org have been used for training and testing the AI algorithms. The accuracy, sensitivity, specificity, precision, F1 score, recall, and area under the receiver operating characteristic curve of these algorithms are evaluated and compared. It was found that, in comparison to other AI algorithms, the deep Recurrent Neural Network (RNN) with Bi-directional Long Short-Term Memory (LSTM) demonstrates better performance for classifying the AFib episodes. The models developed can be integrated into wireless health tracker-based mHealth applications to detect AFib using IHR in real-time.