False alarm reduction in continuous cardiac monitoring using 3D acceleration signals

In continuous cardiac monitoring through wireless Body Sensor Networks (BSNs) using ECG signals, signal quality can be deteriorated due to several factors, including, noise, low battery power and network transmission problems. Body movements occurring when a subject performs activities of daily living (ADLs) are also major causes of high false alarm rates. This paper presents a hybrid framework for false alarm reduction in continuous cardiac monitoring, where classification models constructed using machine learning algorithms are used for labeling input signals and a rule-based expert system is used for combining the classification results into make a final decision. From their extracted low-level features, ECG signal portions are labeled with heartbeat types and also signal quality levels. Meanwhile, low-level features from 3D acceleration signals are used for predicting types of activities. Taking signal quality levels and activity types into considerations, the rule-based expert system then determines whether abnormal ECG portions should trigger alarms or should be ignored. The proposed framework is validated using two datasets: one is obtained from the MIT-BIH arrhythmia database and the other is acquired from 10 subjects while they are performing ADLs. The results of the experiments demonstrate that our proposed framework can reduce false alarm rates in continuous cardiac monitoring and potentially assist physicians in diagnosing a vast amount of data acquired from wireless sensors.