Development of new cluster descriptors for image analysis of Poincaré plots

Abstract

Temporally changing heterogeneous Poincaré plot patterns were reported in arrhythmia subjects who show multiple clusters. Since conventional descriptors rely on the fitting of an ellipse to the plot, they may have limitation in describing multiple clusters accurately. Previously, we reported on the dynamic changes of Poincaré plot patterns observed during the transition from normal sinus rhythm to atrial fibrillation episodes. In the current study, by treating the Poincaré plot as intensity surface, we described the shape, size, intensity, count, and texture of clusters in the Poincaré plot. Multiclass logistic regression analysis classified Poincaré plots into torpedo, island, and multiple side lobe patterns with the average accuracy of 99.4 %. Our results suggested new descriptors may provide better analysis tools of commonly occurring arrhythmia as well as normal heart rhythms.