Comparison of atrial rhythm extraction techniques for the estimation of the main atrial frequency from the 12-lead electrocardiogram in atrial fibrillation

Abstract

One of the greatest challenges in analysis of the atrial rhythm from the ECG is to distinguish the atrial component from the large ventricular components. Our aim was to compare three techniques of atrial rhythm extraction from three groups working on this problem. 12-lead ECG data from 7 patients in atrial fibrillation were analysed. For extraction of the atrial rhythm, spatiotemporal QRST cancellation was performed by the Lund group, blind source separation by the Valencia group, and principal component analysis by the Newcastle group. Peak atrial frequency was determined by Fourier transform of the signal with the largest atrial activity. All algorithms were successful in distinguishing the atrial rhythm from the low frequency ventricular rhythm. The mean (range) atrial frequency was 6.5 (5.9 -7.6) H z (Lund), 6.7 (5.7 -7.9) Hz (Valencia) and 6.5 (5.9 -8.2) H z (Newcastle). There were no SigniJicant differences between the atrial frequencies estimated by each of the techniques. IntroductionAtrial fibrillation is a common arrhythmia in old age with increased risk of stroke and death [ 1 I. The arrhythmia is diverse in its presentation and there are a number of different treatments which aim to restore sinus rhythm [2]. Analysis of the atrial rhythm, before, during and after treatment, provides a means by which to quantify the effects of treatment and to assess the effectiveness of treatment on the different characteristic types of atrial fibrillation. One of the greatest challenges in analysis of the atrial rhythm from the ECG is to distinguish the atrial component from the large ventricular components, namely the QRS and T waves.In this collaborative project, our aim was to compare, by analysis of the main atrial frequency, three techniques of atrial rhythm extraction from three groups working on this problem. The Lund group have developed spatiotemporal QRST cancellation, an extension of a technique for subtraction of the QRST waveform from each ECG lead which has been used for a number of years. Blind source separation and principal component analysis have only recently been applied to the problem of atrial rhythm extraction by the Valencia and Newcastle groups respectively, but are well established signal processing techniques in the fields of image analysis and data communications. Methods Data12-lead ECGs of 7 patients with atrial fibrillation were recorded to computer at the Freeman Hospital, Newcastle.The sampling rate was 500 Hz and the amplitude resolution was less than 5 pV. Recording duration analysed was 60 s. The ECGs were copied and distributed to the groups in Lund and Valencia. Ventricular rate was derived for each patient from the mean heart rate over the 60 s recording. Atrial rhythm extraction techniques Spatiotemporal QRST cancellationThe Lund group applied the spatiotemporal QRST cancellation technique [3]. The ECGs were resampled at 1 kHz and filtered using a 0.3 Hz high pass filter to reduce baseline wander prior to QRST cancellation. Beat detection and cross-correlation based beat class...

Mixing Matrix Pseudostationarity and ECG Preprocessing Impact on ICA-Based Atrial Fibrillation Analysis

“…Contraction of the atria during fibrillation, the ventricles in the cardiac cycle or any other relative movement from sources to observations, could violate the assumption of spatial stationarity. The authors consider that these variations do not affect significantly the ICA model in AF: firstly, results of the AA have demonstrated its validity [10,11], secondly, the main atrial frequency of AA obtained using ICA is in agreement with the results obtained via other accepted AA extraction techniques [20]. Finally, the MMPS corroboration gives the definitive support to say that the 12-lead ECG of an AF recording fulfills the instantaneous linear mixing ICA model.…”

“…Considering the spectral mor- phology of AA, with a very pronounced peak, no harmonics and insignificant amplitudes above 15Hz [8,20], it is possible to define a performance extraction index to evaluate the AA extraction quality based on the spectral concentration, that can be defined as [15] SC = 1.17fp 0.82fp…”

Wavelet Domain Blind Signal Separation to Analyze Supraventricular Arrhythmias from Holter Registers

“…Nowadays, there are several techniques that can extract the AA with a good performance-Blind Source Separation [2], Spatio-Temporal Cancellation [3,4]-but poor results are obtained when the number of used reference signals (leads) is less than three, or when the duration of these signals is reduced. On the other hand, classic techniques-Average Beat Subtraction (AVBS) [5]-have developed AA extraction from only one lead, but these systems are very sensitive to the presence of ectopic complexes.…”

Signal processing of intracardiac recordings for the evaluation of propofol effects during atrial fibrillation.

“…However, the fact that the atrial activity has much lower amplitude than ventricular activity and that both are spectrally overlapped makes atrial activity estimation difficult. In the last years several techniques for atrial activity estimation and analysis have been proposed in the literature [78,79,80] Nevertheless, Langley et al compared three of these techniques, namely, spatiotemporal QRST cancellation, blind source separation, and principal component analysis, and determined that there were no significant differences between the atrial frequencies estimated by any of these techniques [81].…”