Skip to Main Content
The present study discusses two different training techniques for electrocardiogram (ECG) beat detection algorithms. The first technique is a patient specific training method which uses data from the patient's ECG signal to train the beat detector. The second technique is more generic as opposed to patient specific and uses ECG information from a database consisting of a number of patient records to train the detector. Four different beat detection algorithms were considered to facilitate the evaluation of the influence of the training techniques in relation to beat detection performance; a non-syntactic approach, a cross-correlation (CC) approach, a multi-component based CC technique and a multi-component based neural network (NN) technique. An ECG database containing approximately 3000 annotated beats was used for training and test. Superior results were attained with the patient specific training technique. The performance of the two multi-component based classifiers were increased by up to 22% for P-wave and T-wave detection for the patient specific training approach compared to the generic training approach.