Automatic electrocardiogram signal quality assessment in continuous wireless monitoring

This paper presents an automatic signal quality assessment method for continuously monitoring electrocardiogram (ECG) signals using wireless sensors attached to human bodies, with particular attention being given to ECG signals captured while the subjects are performing daily routine activities. In this study signal recordings from three databases are used: two ECG databases acquired using wireless body sensor networks from young subjects and elderly subjects during their daily routine activities, and the Massachusetts Institute of Technology - Boston's Beth Israel Hospital arrhythmia database. From these databases, ECG signals are divided into small segments, each 5 seconds long, and are then labelled with two levels of quality, i.e. 'low-quality' and 'high-quality'. For feature extraction, two levels of statistical features are employed: (i) window-based temporal features and (ii) segment-based features. The latter are derived from statistical values of the window-based temporal features and ECG signal amplitudes. A correlation-based feature selection algorithm is applied to find an optimal set of features. For signal quality classification, four machine-learning-based classification algorithms, i.e. Instance-based Learning, Decision Tree, Multilayer Perceptron and Rule Induction, are compared.