A single-layer perceptron to discriminate non-sinus beats in ambulatory ECG recordings

The recognition of non-sinus electrocardiographic complexes (ectopics) in ambulatory ECG recordings is very important for a correct analysis of arrhythmias as well as of episodes of transient ischemia. We present the results obtained by a single layer neural network (perceptron) that is used to classify sinus/non-sinus beats. The neural network is trained by using the "Perceptron rule" and an alternative least-mean squares (LMS) algorithm. Both approaches indicate linear separability between sinus and ectopie beats. The method shows good sensitivity and specificity values.