Introduction: The detection of atrial fibrillation (AF) is important in clinic practices. The irregular RR interval is a main clinical manifestation of AF in electrocardiogram (ECG). However, it is also the characteristic of other kinds of arrhythmia, such as the bundle-branch block. In order to improve the performance of AF detection, we used a combination of RR intervals and PR interval. Experimental results show that our method archived a good performance on the detection of AF.
Method: After filtering the ECG signal, the features based on the PR intervals, RR intervals and RR intervals difference (ΔRR) were extracted. Then the support vector machine (SVM) with radial basis function (RBF) kernel was used to classify the features. The features of PR intervals was calculated based on the probability density function of the the PR intervals’ phase space. Two statistic moments of ΔRR's histograms was used for the detection of AF: median and standard variation. Furthermore, we developed the entropy based features for the RR intervals, ΔRR and histogram of ΔRR through dividing the sample entropy and approximation entropy by the threshold of entropy.
Result: The performance of our method was evaluated on MIT-BIH atrial fibrillation database. The sensitivity and specificity of the proposed algorithm were 98.43% and 98.51% respectively.
Conclusion: The experimental results show the proposed method has a potential to run in the portable electrocardiograph or AF monitoring device. In the future, we will adopt deep learning technique to improve the performance of AF detection.