To search, Click below search items.


All Published Papers Search Service


A Deep Learning Framework for Inter-Patient ECG Classification


Manh-Hung Nguyen, Vu-Hoang-Tran, Thanh-Hai Nguyen, Thanh-Nghia Nguyen


Vol. 19  No. 1  pp. 74-84


The robust automatic ECG classification systems has attracted researchers in recent years due to saving time and minimizing errors for heart clinical predictions. Inter-patient Electrocardiography (ECG) classification has been studied extensively and provided promising results, but remains a difficult task due to the diversity among patients. Moreover, conventional methods relied on feature selection frameworks to select suitable feature-sets for inter-patient ECG classification task. However, the hand-craft features are specifically designed for different purposes and may not characterize the original signal in an optimal way. In this paper, a proposed deep learning framework is applied to obtain learned patient-invariant features for ECG classification. In particular, two constraints are embedded in a unified Siamese structure to handle inter-patient diversity and ECG classification simultaneously. The first one indicates that if two ECG signals are in the same class, the extracted features should be similar even these signals are from two different patients. The second one is that the learned features should support well for the classification task. Experimental results have shown that the accuracy is improved significantly even in inter-patient datasets. Moreover, t-SNE visualization proves that the proposed framework can learn the discriminative features.


ECG classification, Deep learning, Feature Learning, Siamese network.