Feature-based segmentation of ECG signals

Automatic segmentation of ECG signals is important in both clinical and research settings. Past algorithms have relied on incorporation of detailed heuristics. Here, the authors propose a segmentation technique based on the best local trigonometric basis. They show by means of real data examples that the entropy criterion which achieves the most parsimonious representation of a signal results in an overly-fine segmentation of the ECG signal, and thus establish the need for a more comprehensive criterion. The authors introduce a novel best basis search criterion which is based on a linear combination of the entropy measure and a local measure of smoothness and curvature. They tested the algorithm on the MIT-BIH arrythmia database.