AUTHOR=Shan Lianfeng , Li Yu , Jiang Hua , Zhou Peng , Niu Jing , Liu Ran , Wei Yuanyuan , Peng Jiao , Yu Huizhen , Sha Xianzheng , Chang Shijie 

TITLE=Abnormal ECG detection based on an adversarial autoencoder

JOURNAL=Frontiers in Physiology

VOLUME=Volume 13 - 2022

YEAR=2022

URL=https://www.frontiersin.org/journals/physiology/articles/10.3389/fphys.2022.961724

DOI=10.3389/fphys.2022.961724

ISSN=1664-042X

ABSTRACT=Automatic detection and alarm of abnormal ECG events play an important role in ECG Monitor System; however, popular classification models based on supervised learning fails to detect abnormal ECG effectively. Thus, we propose an ECG anomaly detection framework based on adversarial autoencoder and temporal convolutional network (ECG-AAE) which consists of three modules (i.e. autoencoder, discriminator and outlier detector). ECG-AAE is modeled only by normal data. ECG sequence signal are mapped into latent feature space and then reconstructed back to ECG. Discriminator and autoencoder carry out adversarial training, which improves generation ability of autoencoder and makes reconstructed data more authentic. Abnormal scores are proposed based on reconstruction error and discriminant score to evaluate normal ECG criteria. In this process, Temporal convolutional network (TCN) is employed to extract features of normal ECG data. Then, our model is evaluated on MIT-BIH arrhythmia database and CMUH database, with the accuracy, precision, recall, F1-score, and AUC of 0.9673, 0.9854, 0.9486, 0.9666, 0.9672 and of 0.9358, 0.9816, 0.8882, 0.9325, 0.9358, respectively. The result indicates the ECG-AAE can detect abnormal ECG efficiently, with its performance better than other popular outlier detection methods such as AE, AnoGAN.