A proposed system using a k-nearest neighbor classifier achieved 97.22% sensitivity and 97.4% specificity for heartbeat detection, and 97.1% sensitivity and 96.9% specificity for classification.
Automatic interpretation of electrocardiography provides a non-invasive and inexpensive technique to analyze the heart activity for different cardiac conditions. The emergence of smartphones and wireless networks has made it possible to perform continuous Holter monitoring on patients or potential patients. Recently, much attention has been paid to the development of the monitoring methodologies of heart activity, which include both the detection of heartbeats in electrocardiography and the classification of types of heartbeats. However, many studies have focused on classifying limited types of heartbeats. We propose a system for classification into 17 types of heartbeats. This system consists of two parts, the detection and classification of heartbeats. The system detects heartbeats through repetitive features and classifies them using a A-nearest neighbor algorithm. Features such as the QRS complex and P wave were accurately extracted using the Pan-Tompkins algorithm. For the classifier, the distance metric is an adaptation of locally weighted regression. The system was validated with the MIT-BIH Arrhythmia Database. The system achieved a sensitivity of 97.22 % and a specificity of 97.4 % for heartbeat detection. The system also achieved a sensitivity of 97.1 % and a specificity of 96.9 % for classification.
Park et al. (Sun,) conducted a other in Arrhythmia. k-nearest neighbor classifier system was evaluated on Heartbeat detection and classification. A proposed system using a k-nearest neighbor classifier achieved 97.22% sensitivity and 97.4% specificity for heartbeat detection, and 97.1% sensitivity and 96.9% specificity for classification.