Predicting spontaneous termination of atrial fibrillation with time-frequency information

Abstract:
Our approach to predicting spontaneous termination of atrial fibrillation (AF) is based on parameters derived from the time-frequency distribution of the residual ECG signal, resulting from spatiotemporal cancellation of QRST complexes. Each consecutive, logarithmically scaled spectrum of the time-frequency distribution of the residual ECG is modeled as a frequency shifted and amplitude scaled version of a spectral profile. Parameters characterizing AF, such as fundamental frequency, amplitude, and waveform of the fibrillation waves, are then extracted from the spectral profile. The waveform of the spectral profile is used to distinguish AF from sinus rhythm and noise, so that occasional episodes of sinus rhythm and noise can be excluded. Using these parameters, 95% of the ECGs from the training set, and 90% from the test set was correctly classified as non-terminating and terminating AF respectively.