MULTIDIMENSIONAL ANALYSIS TOWARD THE IDENTIFICATION OF ECG NONLINEAR DYNAMICS

Using ECG it is possible to detect the rate and regu-larity of heartbeats and identify possible irregularities to the heart activity. In this paper, a method to clas-sify normal and two types of abnormal ECG signals is introduced. In the first stage of described process, techniques used to extract a number of potential clas-sification parameters evaluated from 2 minutes long ECG signal epochs are described. The extracted pa-rameters can be generally divided into three groups: (i) standard statistical signal parameters, (ii) nonlinear pa-rameters and (iii) specific heart rate variability parame-ters. Two dimensionality reduction algorithms, princi-pal component analysis (PCA) and linear discriminant analysis (LDA), have been employed in order to reduce the size of dataset containing ECG parameters and fol-lowed by a clustering algorithm. The results show the ability of this method to detect different pathologies and to distinguish normal ECG behaviour from patho-logical ones. Furthermore, this approach could be im-plemented in real time applications and embedded in a portable device and this effort is the first step towards the final realization of an automatic solution for the real-time characterization and monitoring of heart sig-nals. Key words Nonlinear indicators, clustering approach, dimen