Two-stage detection method of supraventricular and ventricular ectopic beats based on sequential artificial features and heartbeats

Abstract

Early detection of Supraventricular ectopic beats (SVEBs) and Ventricular ectopic beats (VEBs) is particularly important for preventing dangerous heart diseases. Researchers proposed many automatic detection methods. However, most of these methods do not utilize the contextual information of artificial features among neighbor heartbeats. Our method consists of two stages. The first stage aims to classify all heartbeats into the ectopic class and non-ectopic class. The ectopic class includes SVEBs and VEBs. The ectopic class includes other heartbeats. Firstly, we extracted Sequential Artificial Features (SAFs) from every 20 neighbor heartbeats, including RR intervals, skewness, existence flags of P-waves and T-waves. Secondly, we developed a framework to implement classification by using SAFs and heartbeats. This framework integrates four independent networks, three of them are used to implement the same classification but have different classification performances, the remaining one is used to fuse outputs of the other three networks. In the second stage, we designed a network to further classify ectopic heartbeats into SVEBs and VEBs. We validated the proposed method on the MIT-BIH Arrhythmia Database with the inter-patient model. For the ectopic class and non-ectopic class, the F1-score reaches 90.82 and 98.96 respectively. For SVEBs and VEBs, the F1-score reaches 80.46 and 95.10 respectively. Experimental results demonstrated that our method and SAFs are effective for SVEBs and VEBs classification.