Time-frequency analysis of atrial fibrillation

Abstract: This licentiate thesis, containing two papers, is in the field of biomedical signal processing with main focus on signal processing of the ECG in patients with atrial fibrillation (AF). The two papers deal with different aspects of characterization of the f waves during atrial fibrillation. In the fist paper several measures designed to extract features of the f waves are evaluated with respect to their ability to predict spontaneous termination of paroxysmal AF. The fibrillation frequency, the variance of the fibrillation frequency, and the harmonic pattern of the f waves was found to differ significantly between terminating and non-terminating AF. In an independent test set, 90\% of the signals were correctly classified using these measures. In the second paper, a method for improving robustness to noise when tracking the fibrillation frequency is presented. The method, which is based on an hidden Markov model (HMM), is evaluated using simulated AF mixed with different types of real noise obtained from ECG. The results show that the use of a HMM improves performance considerably by reducing the RMS error associated with frequency tracking: at 4~dB signal-to-noise ratio the RMS error drops from 0.2~Hz to 0.04~Hz.