| Zugriffsnummer | 19989 |
| Dokumenttyp | Zeitschriftenartikel |
| Sprache | Englisch |
| Titel | Clustering method for evaluation of beat-to-beat variability in high resolution ECG/MCG signals |
| Autor(in); Institution |
Steinhoff, Uwe; 8.21, Bioelektrizität und -magnetismus, PTB-Berlin
Lewandowski, P.; Inst. of Biocybernetics and Biomedical Engineering, PAS, Warsaw, POLAND
Burghoff, Martin; 8.21, Bioelektrizität und -magnetismus, PTB-Berlin
Maniewski, R.; Inst. of Biocybernetics and Biomedical Engineering, PAS, Warsaw, POLAND
Trahms, Lutz; 8.21, Bioelektrizität und -magnetismus, PTB-Berlin
|
| Quelle/Jahr | Biocybernetics and Biomedical Engineering: 19 (1999), 3, 93 - 102 |
| ISSN | 0208-5216 |
| Verlag | Warszawa: Panstwowe Wydawn. Naukowe |
| Freie Schlagworte | Magnetocardiography ; Signal-averaged ECG ; Signal classification |
| Zusammenfassung | Averaging of heart beat signals, yielding the so-called signal averaged electro- or magnetocardiogram (SAECG/SAMCG), is a widely used technique to improve the signal to noise ratio. Usually, heart beats to be averaged are selected from a measured sequence of beats, excluding abnormalities like extrasystoles or artifacts. A common method to achieve the best selection and alignment of the signals is matched filtering. A typical beat is defined as a template. Then the correlation between all possible signal intervals with this template is calculated. All beats with correlation above a certain threshold value are averaged. A critical point of this procedure is the selection of the template. This selection can be done by an operator, leading to a subjective selection depending on the experience or personal preferences of the operator. Also an automatic template selection is possible; criteria for selecting a beat as a template are for example the noise level or the breathing state. More advanced algorithms create a template by preliminary averaging of heart beats. Template based pattern matching provides in any case only a limited view of the data. Structural features not represented by the template are lost, because each beat is compared only with the template and not with other beats containing different structure features. In order to overcome this limitation a new approach is proposed, that avoids the use of a signal template. Instead a crosscorrelation coefficient between each heart beat in the sequence with each other beat is calculated from the multichannel signal. Although this crosscorrelation coefficient can be seen as an euclidian distance measure between two beats, it is not possible to find a common euclidian distance measure for the combination of all beats with each other. The new aproach is to transform the matrix of crosscorrelation coefficients into a similarity matrix, that can be represented as a binary network graph. Then the problem of finding groups of similar heart beats can bE described as the clique covering problem, well-known from graph theory. A Reactive Local Search (RLS) algorithm is used to find clusters of maximal size in the network graph, thus automatically grouping the heart beats into clusters. It is shown that this method does not only improve the averaging procedure, but also gives useful information about the physiological processes that alter the pattern of the heart beat signals. |
