A probability density function method for detecting atrial fibrillation using R-R intervals

A probability density function (PDF) method is proposed for investigating the structure of the reconstructed attractor of R-R intervals. By constructing the PDF of distance between two points in the reconstructed phase space of R-R intervals of normal sinus rhythm (NSR) and atrial fibrillation (AF), it is found that the distributions of PDF of NSR and AF R-R intervals have significant differences. By taking advantage of their differences, a characteristic parameter k(n), which represents the sum of n points slope in filtered PDF curve, is put forward to detect both 400 segments of NSR and AF R-R intervals from the MIT-BIH Atrial Fibrillation database. Parameters such as number of R-R intervals, number of embedding dimensions and slope are optimized for the best detection performance. Results demonstrate that the new algorithm has a fast response speed with R-R intervals as short as 40, and shows a sensitivity of 0.978, and a specificity of 0.990 in the best detecting performance.