Propagation of electrical atrial activity (AA) during atrial fibrillation (AF) is a process characterized by different short-and long-term recurrence behaviours. Two antithetical (not mutually exclusive) hypotheses were proposed to noninvasively describe this nonstationary behaviour. The first hypothesis (H1) assumes a process with stationary spatial properties of AA propagation, but time-varying frequency properties, and vice versa for the second (H2). Based on H1 and H2, two phenomenological models were proposed, both able to replicate observations on AF patients, and a novel measure was introduced to assess the spatial variability of AA propagation (SVAAP) over short and long AA segments. Validity of the models was tested by looking at the relation between SVAAP-short and SVAAPlong on real observations from AF patients (high-density body surface potential maps recorded in 75 patients affected by persistent AF). H1 is confirmed if SVAAP-short is approximately equal to SVAAP-long. H2 if SVAAP-short is less than SVAAP-long. Results confirmed H2, showing that AA propagation during AF has strong nonstationary spatial properties. This could suggest new parameters to characterise AF substrate and predict therapy outcome.

Atrial flutter circuit localization is usually determined during a catheter ablation procedure. Knowledge of this information beforehand could aid clinicians assess and plan the operation in advance to improve efficacy. Variability as a marker to discriminate localization noninvasively was suggested in the literature, and evaluated by one group. However, variability may originate from respiratory motion which may affect right and left AFL differently. This is hypothesized to be the reason for difference in right and left AFL variability. To address this, we analyzed the effect of removing respiratory motion influence from f wave observations on classification accuracy. ECG records from patients with AFL were processed using a novel approach: respiratory motion was estimated and removed in order to recover variability intrinsic to AFL. Vectorcardiographic loop parameters were estimated from each observation and statistical measures of the set of parameter values were calculated before being fed into a classification algorithm. The results show that f waves are negligibly affected by respiratory motion. It is also possible to discriminate circuit localization with good accuracy.

Accurate detection of f waves during atrial flutter is difficult. However, f waves contain information on the pathology and are useful for non-invasive diagnosis. The setup and difficulties of f wave detection lends itself to the use of statistical signal detection techniques. Real-life constraints can be modeled in the signal observation using several parameters in order to produce signal detectors with good performance. Several detectors were developed and tested using real 12-lead ECG recordings with manually annotated f wave markers available. At the end, a simple detector is obtained with relatively good detection performance (AUC = 0.89, (Se, Sp) = (0.84, 0.81)) and a threshold is available for use in automatic detection of f waves.

No abstract

The use of the ECG as a diagnostic tool for analyzing supraventricular arrhythmia avoids invasive procedures prior to clinical intervention. Analysis of atrial activity may indicate properties of the underlying arrhythmia. In atrial flutter, there is a high synchroneity between atrial and ventricular activity. Statistical and classical separation methods are unlikely to work due to this dependency, impeding the extraction of the atrial activity. We propose a method using least square polynomial estimation to model the T wave component and subtract it from overlapped fwaves. After pre-processing, an initial f-wave was carefully segmented, representing pure atrial activity. Subsequent f-waves were detected and corrected by subtracting the T wave, modeled as a sum of weighted polynomials which minimize the least square error. To evaluate the method, vectorcardiographic loops were obtained using the inverse Dower transform. Loop parameters were obtained from eigenvectors and eigenvalues issued from principal component analysis. Using degree 3 polynomials allowed recovery of overlapped f-waves with significant improvement in 3 of 4 parameters (p < 0.05). f-waves overlapped by T waves can thus successfully be recovered using this method.