Fusing conventional ECG QRS detection algorithms with an auto-associative neural network for the detection of ectopic beats

The onset of a potentially fatal arrhythmia is often preceded by abnormal morphologies in the QRS complex, the main feature in the electrocardiogram. However, these ectopic beats are difficult to detect as their shape is very similar to those found in a normal sinus rhythm. We show how an auto-associative multi-layer perceptron can be trained to detect normal beats only, so that the subtle abnormalities in the shape of ectopic beats become clearly identifiable. Details of how to train the network for use in a clinical environment are given utilising a new parameter, the variance ratio. Results for a study of the combination of algorithms to produce a robust ectopic beat detector are presented. Finally we discuss an on-line implementation for patient-specific adaptability.