Inter-patient ECG Classification Based on Residual Networks and Active Learning

In this paper, an inter-patient electrocardiogram (ECG) classification algorithm based on residual networks and active learning is proposed. Our model is an end-to-end model. We use Convolutional Neural Networks (CNN) to extract morphological features of each ECG beat and use Long Short-Term Memory (LSTM) to extract temporal features. we apply active learning to select the most informative beats and incorporate them into the training set to improve algorithm performance. When evaluated on the MIT-BIH Arrhythmia Database, the method our proposed gives 97.7% sensitivity and 61.6% positive predictive for ventricular ectopic beat (VEB). For supraventricular ectopic beat (SVEB), our method gives 89.4% sensitivity and 41.5% positive predictive. Considering the high sensitivity in detecting these two pathological classes, this method has potential clinical application.