An Automated Approach for Atrial Fibrillation Detection Using Electrocardiogram

Atrial fibrillation (AF) is a prevalent cardiac rhythm disorder with substantial health consequences that remains concealed in the initial stages due to a lack of symptoms and gradually advances to the chronic phase, emphasizing the urgent need for automated detection models. This paper introduces an automated deep-learning model for AF prediction at earlier stages incorporating a bidirectional long-short-term memory (BiLSTM) that leverages various time series heart rate variability (HRV) features extracted from electrocardiogram signals. The execution of the minimum redundancy maximum relevance algorithm optimizes the feature set, improving the effectiveness of the model. A total of 730 recordings from 54 subjects were used to validate the model. Subjectwise 5-fold cross-validation is performed and the model attains 92.34 ± 3.74% accuracy, 95.23 ± 2.01% specificity, a nd 9 1.00 ± 3.00% F 1 s core. The outstanding performance of our proposed model, employing a limited set of features, highlights its significance and positions it as a strong candidate for AF monitoring.