The use of Sequential RR Distributions to detect atrial fibrillation episodes in very long term ECG monitoring

The Sequential RR Distribution (SRRD) is introduced as a diagnostic tool for AF detection and classification. SRRD is obtained by computing consecutive RR histogram distributions in successive temporal windows and plotting them prospectically The validation of SRRD for AF detection was performed using the MIT AF database. A interactive graphic interface was developed to navigate in the SRRD and to manually annotate the onset and offset of the AF episodes. Two expert cardiologists were trained to evaluate the SRRD using an home-made database. They were asked to annotate AF events in the MIT database using RR distributions (without accessing the ECG). The results were: episodes sensitivity 97%, episode P+ 78%, duration sensitivity 98%, duration P+ 95%. These results show that sequential RR histogram distributions are accurate enough to allow the detection of AF events without the need of viewing the ECG signal.