Atrial Fibrillation Detection using Different Duration ECG Signals with SE-ResNet

Atrial fibrillation (AF) is the most common cardiac arrhythmia that has significant effects on associated morbidity and mortality. Therefore, early detection of AF by electrocardiogram (ECG) is substantial and necessary for effective treatments of AF. Here, we creatively propose a signal processing method for ECG signals and develop a 94-layer deep neural network to classify 4 rhythm signals using 8528 different duration single-lead ECGs from The PhysioNet Challenge 2017. And the model is trained and validated on different databases, respectively. We utilize F1 score to evaluate the performance of our method and obtain the mean F1 score of 85%. Finally, the experiment results demonstrate that our algorithm achieves excellent performance for AF and other arrhythmia detection and is also superior to existing algorithms.