Automated detection of atrial fibrillation using fourier-bessel expansion and teager energy operator from electrocardiogram signals

Abstract

This work presents a new method for detection of atrial fibrillation using predictors derived from Fourier-Bessel (FB) expansion and Teager energy operator (TEO) which are applied strategically on electrocardiogram (ECG) signals. The proposed method begins by extracting a set of direct and indirect predictors. The direct predictors are computed from pre-processed ECG signals themselves. A part of indirect predictors are computed from (a) RR-interval and heart rate (HR) signals, and (b) FB expansion along with its spectrum applied on RR and HR signals. The rationale of using FB expansion is that the clinical information is found to be more evident in the FB coefficients (FBC) and their spectrum than that of RR and HR signals themselves. In the same line of thought, TEO is applied on pre-processed ECG, RR-interval, HR signals, said FBC and their spectrum to obtain the other part of predictors. In all, 47 predictors are computed and subsequently they are fed to an ensemble system of bagged decision trees for classifying the ECG recordings. When evaluated with 2017 PhysioNet/CinC Challenge dataset (Phase II subset), the experimental outcomes demonstrate the F1 scores of Normal, AF and other classes as: 90.89 %, 80.07%, 72.24% respectively with overall F1 score of 81% for the hidden test data.