Fully automatic electrocardiogram classification system based on generative adversarial network with auxiliary classifier

A generative adversarial network (GAN) based fully automatic electrocardiogram (ECG) arrhythmia classification system with high performance is presented in this paper. The generator (G) our GAN designed to generate various coupling matrix inputs conditioned on different classes for data augmentation. Our discriminator (D) trained both real and generated ECG inputs, extracted as an classifier upon completion of training GAN. After fine-tuning the D by including patient-specific normal beats estimated using unsupervised algorithm, abnormal G that are usually rare obtain, showed superior overall supraventricular ectopic (SVEB or S beats) ventricular (VEB V MIT-BIH database. It surpassed several state-of-art classifiers can perform similar levels some expert-assisted methods. In particular, F1 score SVEB has been improved up 10% over top-performing systems. Moreover, sensitivity (87%) VEB (93%) detection achieved, which great value practical diagnosis. We, therefore, suggest ACE-GAN (Generative Adversarial Network Auxiliary Classifier Electrocardiogram) be a promising reliable tool throughput clinical screening practice, without any need manual intervene expert assisted labeling.