Automated localisation and classification of abnormal beats in electrocardiograms using parsimonious wavelet analysis

We report on a work in progress aiming at automatically analysing electrocardiograms with the help of wavelet bases. The objective we have fixed was to introduce a parsimonious and computationally efficient method --- that can be ultimately incorporated on standard ECG recording devices --- to automatically localise points of interest and decide whether they are normal or not and, for abnormal points, assign them a given class of abnormality. The task we have already achieved is the localisation and identification of normal QRS complexes. Identification and localisation of other features, like premature atrial or ventricular beats, fibrillation, noise, bundle branch block beats, ectopic beats, etc. is in progress but the so far obtained preliminary results are very encouraging.