Fast Straightforward RR Interval Extraction Based Atrial Fibrillation Detection Using Shannon Entropy and Machine Learning Classifiers for Wearables

Atrial fibrillation (AF), a complex arrhythmia with substantial morbidity and mortality implications, demands timely detection to preempt chronic cardiac complications. The need for continuous AF monitoring rises the demand for an automatic, fast, and reliable detection approach that ensures low computational complexity in terms of model size and processing time. This study presents an AF detection method using a fast straightforward RR interval extraction method and Shannon entropy (ShE). The method utilizes symbolic dynamics from electrocardiogram (ECG) segments’ heart rate sequences to calculate ShE. When tested on two datasets (2-lead and 12-lead) of 10 s and 30 s durations, the method achieves an accuracy of 99.958% and 100%, respectively, utilizing five machine learning classifiers. Furthermore, it showcases an exceptionally fast detection time of 0.286 µs with multilayer perception neural network. The best performance is achieved with 10 s ECG segments with Naive Bayes classifier. The classifier obtained an accuracy of 99.958% with model size of 1.5 kB and processing time of 2.13 µs. In comparison to previous studies, the evaluation results demonstrate the superior sensitivity, specificity, accuracy, and speed of this newly developed AF detection method with low computational complexity. It is clear from the experimental results that the proposed methodology is highly suitable for implementation in real-time health monitoring systems.