Modelling and Quality Assessment of Atrial Fibrillatory Waves

Sammanfattning: This doctoral thesis revolves around the analysis of electrocardiogram (ECG) signals during atrial fibrillation (AF). Special emphasis is put on the atrial fibrillatory waves, sometimes called the f-waves, which is the ECG component reflecting the electrical activity of the atria. The thesis comprises an introduction and five papers that introduce and apply methods on ECG-based analysis of AF. Paper~I deals with modelling of the relationship between atrial and ventricular activity while papers~II--V deal with the modeling, analysis and quality assessment of the f-waves, the atrial activity component of the ECG.Paper~I presents a novel statistical dual pathway model of the atrioventricular (AV) node during AF. The model accounts for pathway switching, meaning that atrial impulses may alternate between arriving at the slow and the fast pathway, even if the preceding impulse did not cause a ventricular activation. Comparison between the present model, defined by four parameters, and a reference model, defined by five parameters, does not reveal any difference in modelling capability. However, parameter estimates of the present model exhibit considerably lower variation, a finding that may be ascribed to the reduction of model parameters. Paper~II proposes an f-wave signal quality index (SQI). The SQI is computed using a harmonic f-wave model which allows for variation in frequency and amplitude. Unlike the noise level estimator used for comparison, the f-wave SQI reflects signal quality adequately also when the spectral content of the noise overlaps with that of the observed f-waves. The SQI is shown to be highly associated with f-wave presence, obtaining considerably smaller values when computed from non-AF signals, which is exploited to improve the performance of an AF detector.Paper~III investigates the signal quality aspects of 24h tracking of the dominant atrial frequency (DAF), using the f-wave SQI from Paper~II. The use of the SQI reveals that 40\% of all 5-s signal segments of the database should be excluded due to poor quality, with the recordings of some patients being completely removed. Removal of the noisy segments reduced the variation of the DAF trend during both day- and night-time . A decrease in signal quality is observed during veloergometry exercise, with the quality restored immediately afterwards. Paper~IV investigates the f-wave changes occurring during pulmonary vein isolation, a treatment option for AF patients. Three f-wave parameters, derived from the harmonic f-wave model from Paper~II, are included -- the DAF, the f-wave amplitude, and a novel regularity parameter named the phase dispersion. All three f-wave parameters correlate with clinical characteristics, but none of them can predict AF recurrence. However, the DAF decreases significantly during the procedure. Paper~V investigates the reproducibility of f-wave parameters, including the three from paper~IV as well as the spectral organization index and spatiotemporal variability, which have been included in previous studies on f-wave analysis. For each parameter, the variance ratio between the inter- and intrapatient variance, is computed, a larger ratio corresponding to better parameter stability and reproducibility. A substantial difference in inter- and intrapatient variation is found among different parameters, with the DAF and f-wave amplitude obtaining considerably larger variance ratios than the rest.In summary, this thesis presents and evaluates tools for ECG-based AF analysis with special attention on robustness and quality control. The SQI presented in Paper~II is applied in Paper~III--V, and it is concluded that some kind of quality assessment should be considered in all future studies involving f-wave analysis.