An improved deep learning approach based on exponential moving average algorithm for atrial fibrillation signals identification

Abstract

Atrial fibrillation (AF) is one of the leading causes of heart diseases worldwide. An accurate AF detection method for timely treatment is attracting great attention by widespread scientific and clinical research in recent years. However, routine AF detection using the visual examination of electrocardiogram data is strenuous and relatively subjective. In this study, an improved Elman neural network (IENN) model is specially designed for automated AF signals identification. Inspired by the exponential moving average (EMA) algorithm, a weight strategy is installed into the existing ENN network that provides more effective and comprehensive discrimination ability to the model for ECG signals recognition. Besides, our IENN network is embedded into a convolutional network architecture and the existing multi-layer perceptron and ENN networks are considered as control subjects to evaluate the validity on two public databases. The results exhibit that the proposed model yields the accuracy, specificity and sensitivity of 97.9%, 97.8% and 98.0% on the MIT-BIH AF database and 97.1%, 97.1% and 97.2% on the MIT-BIH arrhythmia database, respectively. These superior performances enable the proposed IENN model to have considerable potential as an effective tool to aid cardiologists in detecting AF signals accurately.