A kNN Approach for ECG Signal Quality Classification

Abstract

The quality of Electrocardiogram (ECG) signal 
recordings is importantly considered in continuous monitoring systems, especially in monitoring systems using wireless devices, e.g., wireless Body Sensor Networks (BSNs). Patient’s ECG signal recordings with low quality frequently cause in high false alarms in the Cardiac Care Unit. Furthermore, ECG signals acquired from the wireless BSNs while subjects perform activities of daily living (ADLs) can be often deteriorated by baseline drift noises and motion artifacts, occurring from human body movements. Therefore, for improving the performance of continuous monitoring systems using BSNs, low-quality signals should be detected and then should be suppressed from the systems. This paper presents an automatic approach for signal quality classification using a simple instance-based machine learning algorithm, i.e., k-Nearest Neighbor (kNN), and statistical ECG-based features. In data acquisition, a wireless BSN node was used for collecting ECG signals from 10 subjects while performing ADLs. For data annotation, the obtained signals were divided into small segments (each 5 seconds long) and these segments are annotated with good-quality and bad-quality labels depending on their signal quality levels. The average evaluation results of signal quality classification are 96.87%, 84.79%, and 98.44%, for accuracy, sensitivity, and specificity, respectively.