Atrial fibrillation detection using a hybrid 1D CNN-KNN architecture

Early detection of atrial fibrillation plays a crucial role in the timely prevention, diagnosis, and treatment of cardiovascular diseases. This paper proposes two different network architectures for automated atrial fibrillation detection. In the first architecture, a 1D CNN is used as a feature extractor and classifier. In the second hybrid architecture, a 1D CNN is used only as a feature extractor from ECG time series signals that supply a KNN with the most relevant features for further classification. Experimental results showed that the hybrid architecture achieved remarkable results and outperformed a 1D CNN.