A Noise-Adaptive Method for Detection of Brief Episodes of Paroxysmal Atrial Fibrillation

The aim of this work is to develop a method for detection of brief episode paroxysmal atrial fibrillation (PAF). The proposed method utilizes four different features: RR interval irregularity, absence of P waves, presence of f-waves and noise level. The obtained features are applied to the Mamdani-type fuzzy inference method for decision-making. The performance was evaluated on one hundred 90 s long surrogate ECG signals with brief PAF episodes (5-30 beats). The robustness to noise in ECGs where noise level in each set is incremented in steps of 0.01 mV from 0 to 0.2 mV was examined as well. When compared to the coefficient of sample entropy, our method showed considerably better performance for low and moderate noise levels (<; 0.06 mV) with an area under the receiver operating characteristic curve of 0.9 and 0.94, respectively. Similar performance is expected for higher noise levels as atrial activity is less used in the detection process. Finally, the results suggest that our method is more robust to false alarms due to ectopic beats or other irregular rhythms than the method under comparison.