Atrial fibrillation detection using spectra of FSD recurrence complex network

Complex network spectra features are proposed to be used by the classifier to classify atrial fibrillation (AF) and normal sinus rhythm (NSR). This novel complex network construction method utilizes the fuzzy symbolic dynamics (FSD) and recurrence complex network to analyze the synchronization of cardiac electrical activity. Firstly, the multi-lead epicardial signals recorded from dogs are transformed into the FSD plot to construct a complex network by using the recurrence complex network theory. Then, network spectra features are extracted. Finally, epicardium signals are classified into AF and NSR patterns by using the fuzzy c-means algorithm (FCM). The method is validated with three sample sets from dogs A, B and C. The samples are NSR and AF segments which are all from dog experiments. The sensitivity (SE), specificity (SP) and accuracy (AC) of the identification for set A are 98.2%, 98.9% and 98.6% respectively, 96.1%, 98.6%, 97.5% for set B, and 97.7%, 98.7%, 98.4% for set C. The experiments indicate that this approach is effective in AF detection.