Atrial fibrillation (AF) affects electrophysiological activity, structures, functions, autonomy, and metabolism of human heart. Thus, it is critical to detect AF events early. Heart rate variability (HRV) analysis has been well defined quantitative indicator of clinical cardiac events. Poincare plot is one of the techniques that are used to analyze HRV. Highly heterogeneous Poincare plot patterns exist in arrhythmia subjects, which involve multiple “side lobes” clusters, “islands,” “torpedo,” and dispersed points. To overcome the limitation of conventional Poincare plot features, we developed new Poincare plot features to better represent the heterogeneity of Poincare images. Our method describes clusters and points in the Poincare plot based on their shapes, intensities, and textures by using quantitative image analysis.
To examine the effectiveness of new Poincare plot descriptors in classifying the Poincare plot patterns, we compared the accuracy of three classification models based on conventional HRV features only, new cluster descriptors only, and both conventional and new cluster descriptors. While conventional Poincare plot features showed the accuracy of 78.4 % in classifying Poincare plot patterns, new cluster descriptors improved the classification result with 84.0 % accuracy. When new cluster descriptors were combined with conventional features, the accuracy reached to 99.4 %.
New Poincare plot descriptors also showed their effectiveness in predicting AF episodes. While conventional HRV features could classify the atrial fibrillation and normal sinus rhythm (NSR) with 94.3 % accuracy, our new Poincare plot descriptors improved the accuracy to 98.2 %. The best accuracy of 99.2% was obtained when both HRV features and new Poincare plot descriptors were used.
To illustrate our study outcomes, a web-based application was implemented. The application may provide a useful tool for users to monitor their heart condition. In addition, the application may also serve as a research platform that provides ECG analysis software tools and databases.
In conclusion, the results of our studies suggested that our new descriptors may provide more insight in understanding commonly occurring arrhythmia as well as normal heart rhythms.
