Cardiac rhythm tracking in electrocardiographic (ECG) signals is one of the essential tasks in heart disease diagnosis. A morphological derivative transform-based singularity detector (MMDTSD) technique is developed for the detection of singular points in the finite non-derivative signal and applied for atrial and ventricular events tracking in ECG. The MMDTSD technique is conducted by substituting the traditional derivative with a multiscale morphological derivative. The cardiac characteristics in ECG can be considered as corresponding to singular points and they can be reliably detected by the proposed MMDTSD technique. Test results indicate that the proposed technique can enhance the capability of arrhythmia detection in cardiac instrumentation system.

A modified Learning Vector Quantization (MLVQ) neural network is employed to develop an unsupervised ECG beat classifier. In order to improve the performance of the classifier for application to ECG signals, three modifications are made on the original LVQ: finding the clustering numbers in an unsupervised way by appending two counters, decreasing the inaccuracy caused by the imprecise input features of classifier using multiple assignment of datum, and finding the global optimal classification of input data using double objective functions. This unsupervised classifier is tested with selected ECG time series and experimental results show that the proposed technique offers a great potential in the unsupervised classification of ECG beats.