Visualization and analysis of signal averaged high resolution electrocardiograms employing cluster analysis and multidimensional scaling

We describe an algorithm that combines k-means clustering with an adaptive multidimensional scaling procedure. This method allows the on-line visualization of clustering processes. This procedure was applied to the features used in the time domain ventricular late potential (VLP) analysis: duration of the filtered QRS, rootmean square of the last 40 ms and duration of the terminal part of the QRS below 40 /spl mu/V. The algorithm produces 2D-maps of the cluster centers, where the centers were grouped around a straight line and the distances between all cluster centers were preserved. We therefore conclude that this 3-dimensional data set can be embedded into a 1-dimensional abstract feature space. This feature space is spanned more or less by the duration of the QRS-complex.