An ECG T-wave anomalies detection using a lightweight classification model for wireless body sensors

Various wearable devices are foreseen to be the key components in the future for vital signs monitoring as they offer a non-invasive, remote and real-time medical monitoring means. Among those, Wireless Body Sensors (WBS) for cardiac monitoring are of prominent help to early detect CardioVascular Diseases (CVD) by analyzing 24/24 and 7/7 collected cardiac data. Today, most of these WBS systems for CVD detection, include only limited automatic anomalies detection, particularly regarding ECG anomalies. Severe CVD, such as Myocardial Infarction or Ischemia, needs to achieve an advanced analysis of ECG waves known as P, Q, R, S and T. In particular, the T-wave and its specific changes. In this paper, we focus on T-wave anomalies detection in a context of WBS. Our study suggests an accurate and lightweight T-wave changes detection model which suits well an ECG monitoring system based on WBS architecture. We performed a comparative study of 7 well-known supervised learning classification models, on real ECG data sets from 7 different leads. We compared the results from both perspectives of classification and processing times. Our results show that the C4.5 Decision Tree technique performs better results with 92.54% Accuracy, 96.06% Sensibility, 55.41% Specificity and 7.41% Error Rate.