Binary Classification for Linear Approximated ECG Signal in IoT Embedded Edge Device

Abstract

Abnormal beat detection in electrocardiogram (ECG) signal is an important research subject. Abnormal beat detection can be used effectively for adaptive signal compression according to normal/abnormal beat, and it enable to save time and cost of arrhythmia diagnosis by providing the detected abnormal beats to cardiologist. However, the fiducial point detection for feature value extraction has low reliability and is difficult to implement in embedded edge devices due to the auxiliary signal acquisition and complex algorithm for detection. In this study, we propose a method that expresses a signal as a small number of vertices using linear approximation and detects an abnormal beat quickly and reliably using the feature value of vertices. The proposed method is based on the similar distribution of feature values of the approximate vertices for the same type of beat. As a result of an experiment on a record containing premature ventricular contraction (PVC) whose shape was deformed from a normal beat, we confirmed that the proposed algorithm enable to detect whole abnormal beat correctly.