AfibPred

Abstract

According to the World Health Organization, cardiovascular disease (CVD) is the leading cause of death worldwide. One of the most common heart diseases that lead to stroke and some other heart problems is an Atrial Fibrillation (AFib). To cope with this dangerous risk, an early diagnosis of CVD is a global emergency helping patients to get timely treatment, and decreasing the prevalence of mortality. The early diagnosis of CVD can be conducted by an efficient and accurate analysis of a heart electrocardiogram (ECG) signal. In this paper, we propose a novel atrial fibrillation prediction approach called (AfibPred). AfibPred is based on the recent deep transfer learning technology to detect AFib disease through reusing knowledge gained from classifying some various CVD diseases. This proposal is consisted of three main processes, the first one is to discover pretrained models trained on the large multi-leads dataset, whereas the second process exploits the pretrained models for classifying Atrial Fibrillation (AFib), Noisy (P), Normal (N) and Other rhythms (O) by analyzing short single-lead ECG signal. Finally, we are stacking different models gained using Extreme Gradient Boosting (XG-boost) method. After a set of validations, the best trained AfibPred achieved an overall F1-score of 97%, and AFib F1-score of 99% proving the accuracy of AFib prediction.