| Abstract [eng] |
Atrial fibrillation (AF) is the most common arrhythmia in clinical practice. Individuals with atrial fibrillation have an increased incidence of various complications (i.e., stroke) and increased mortality. Atrial fibrillation is a progressive disease, therefore, a pattern of self-terminating paroxysmal AF (PAF) episodes depends on the stage of the disease. In order to diagnose PAF in the initial stage, there is a need for technologies that could ensure unobtrusive long-term monitoring. Long-term monitoring of PAF progression would allow characterization of temporal distribution of PAF episodes and could be useful for finding causal relationships between PAF pattern and complications. The aim of this work is to introduce parameters for an objective evaluation of temporal PAF episodes distribution and to investigate these parameters with real and simulated PAF patterns. In this work, real PAF patterns from public Physionet databases were analyzed. However, there is a lack of long-term annotated PAF pattern, so two models were created. The first model simulates four different PAF progression patterns (duration – 1 year), and the second simulates four PAF patterns (duration – 24 h), which are common in real patterns. In addition, four metrics for the objective evaluation of temporal PAF patterns were introduced (PAF burden, intensity, aggregation and Gini coefficient). Four metrics were investigated with the real and simulated PAF patterns. The results with real PAF patterns show that the aggregation metric is capable of discriminating among different PAF patterns. While the results from the simulated PAF patterns show that the aggregation brings information about PAF progression stage and that the presented model of PAF patterns is suitable for simulating individual PAF patterns which are equivalent to the real ones. In addition, the aggregation metric was investigated in more detail. |
