Carlos Vayá

Convolutive Blind Source Separation Algorithms Applied to the Electrocardiogram of Atrial Fibrillation: Study of Performance

The analysis of the surface electrocardiogram (ECG) is the most extended noninvasive technique in medical diagnosis of atrial fibrillation (AF). In order to use the ECG as a tool for the analysis of AF, we need to separate the atrial activity (AA) from other cardioelectric signals. In this matter, statistical signal processing techniques, like blind source separation (BSS), are able to perform a multilead statistical analysis with the aim to obtain the AA. Linear BSS techniques can be divided in two groups depending on the mixing model: algorithms where instantaneous mixing of sources is assumed, and convolutive BSS (CBSS) algorithms. In this work, a comparison of performance between one relevant CBSS algorithm, namely Infomax, and one of the most effective independent component analysis (ICA) algorithms, namely FastICA, is developed. To carry out the study, pseudoreal AF ECGs have been synthesized by adding fibrillation activity to normal sinus rhythm. The algorithm performances are expressed by two indexes: the signal to interference ratio (SIRAA) and the cross-correlation (RAA) between the original and the estimated AA. Results empirically prove that the instantaneous mixing model is the one that obtains the best results in the AA extraction, given that the mean SIRAA obtained by the FastICA algorithm (37.6 plusmn 17.0 dB) is higher than the main SIRAA obtained by Infomax (28.5 plusmn 14.2 dB). Also the RAA obtained by FastICA (0.92 plusmn 0.13) is higher than the one obtained by Infomax (0.78 plusmn 0.16).

Wavelet denoising as preprocessing stage to improve ICA performance in atrial fibrillation analysis

Blind Source Separation (BSS) has been probed as one of the most effective techniques for atrial activity (AA) extraction in supraventricular tachyarrhythmia episodes like atrial fibrillation (AF). In these situations, a wavelet transform denoising stage can improve the extraction quality with low computational cost. Each ECG lead is processed to obtain its representation in the wavelet domain where the BSS systems improve their performance. The comparison of spectral parameters (main peak and power spectral density concentration) and statistics values (kurtosis) proves that the sparse decomposition in the wavelet domain of the observed mixtures reduces Gaussian contamination of these signals, speeds up the convergence and increase the quality of the extracted signal. The easy and fast implementation, robustness and efficiency are some of the main advantages of this technique making possible the application in real time systems as a support tool to clinical diagnostics.

Performance study of convolutive BSS algorithms applied to the electrocardiogram of atrial fibrillation

Atrial Fibrillation (AF) is one of the atrial cardiac arrythmias with highest prevalence in the elderly. In order to use the electrocardiogram (ECG) as a noninvasive tool for AF analysis, we need to separate the atrial activity (AA) from other cardioelectric signals. In this matter, Blind Source Separation (BSS) techniques are able to perform a multi-lead analysis of the ECG with the aim to obtain a set of independent sources where the AA is included. Two different assumptions on the mixing model in the human body can be done. Firstly, the instantaneous mixing model can be assumed in spite of the inaccuracy of this approximation. Secondly, the convolutive model is a more realistic model where weighted and delayed contributions in the generation of the electrocardiogram signals are considered. In this paper, a comparison between the performance of both models in the extraction of the AA in AF episodes is developed by analyzing the reults of five distinct BSS algorithms.

Derivation of atrial surface reentries applying ICA to the standard electrocardiogram of patients in postoperative atrial fibrillation

In this study a set of patients undergoing cardiac surgery, that developed postoperative atrial fibrillation, were selected to verify if the information available on the atrial surface can be derived with the only use of body surface recordings. Standard electrocardiograms were obtained and processed by independent component analysis (ICA) to extract a unified atrial activity (AA) that takes into account the atrial contribution from each surface lead. Next, this AA has been compared with internal recordings. Main atrial frequency, cross-correlation between power spectral densities and spectral coherence have been obtained in this study. Results show that information provided by surface ICA-estimated AA allows to derive atrial surface reentries in AF patients, thus improving the noninvasive knowledge of atrial arrhythmias when internal atrial recordings are unavailable.

Poincaré plots of time-frequency parameters applied to the prediction of Atrial Fibrillation termination

Atrial Fibrillation (AF) episodes are commonly encountered in the daily clinical practice and cardiologists have often to face the difficulty of classifying between terminating and non-terminating AF episodes. Given that in these critical situations a decision must be made with the utmost urgency, it would be desirable to have a visualization tool of easy interpretation that could provide a fast and reliable prediction of AF episode evolution. In this essay, a method based on Poincare plots and time-frequency analysis is presented as a new technique of AF diagnosis.

Atrial activity enhancement by blind sparse sequential separation

Blind source separation (BSS) has been probed as one of the most effective techniques for the atrial activity (AA) extraction in supraventricular tachyarrhythmia episodes like atrial fibrillation (AF). In these situations, the registered episodes with only a few leads are noisy and time-varying and previous stages for sparse separation have been demonstrated as necessary. Including wavelet transform de-noising and natural gradient algorithm for the BSS system can improve the extraction quality with low computational load. Synthetic signals have been used to test the proposed technique in different noisy cases. The obtained cross-correlation coefficients with sparse sequential separation between the extracted signal and the original ones exceed the 94% in contrast with the obtained results using standalone BSS method. The easy and fast implementation and the minimum required reference recordings of the same ECG are some of the main advantages of this technique and make the application in real time systems possible

Epicardial atrial activation assessment from the surface ECG in atrial fibrillation

Atrial fibrillation (AF) is one of the most common complication of cardiothoracic surgery affecting from 30% up to 60% of the patients. In this study 15 patients undergoing cardiac surgery, that developed postoperative atrial fibrillation, were selected to assess if the information available in epicardial recordings can be recovered with the only use of body surface recordings. Surface ECGs were processed by independent component analysis (ICA) to extract a unified surface atrial activity (AA) that takes into account the atrial contribution of each lead. Next, the estimated AA has been compared with epicardial recordings, the spectral cross-correlation being 85.34plusmn11.08% (range 60.57-97.31) and the average spectral coherence 70.10plusmn9.46% (range 54.92-83.95). Therefore, this study has assessed that information provided by the surface ICA-estimated AA is a valid and useful tool to analyze the properties of atrial activation patterns in AF patients, thus allowing to improve the information about atrial arrhythmias in those patients where epicardial recordings are unavailable

Complex detection and subtraction via wavelet, a new atrial activity extracting algorithm

In this paper, a new technique for extracting the Atrial Activity (AA) using a single-lead from surface ECG and based on Wavelet transform and adaptive filtering, is presented. Firstly, the fiducial points of each beat are detected using a Discrete Wavelet Transform (DWT). In the second stage, the dominant frequency (Fp) of the f waves segments is calculated, allowing the application of an adaptive filtering. Averaging this signal with a median complex based on Template Matching and Subtraction cancellation technique (TMS) results a signal where AA is minimum. Finally, a subtraction between the original lead and the averaged signal produces a residual signal which contains the expected AA. The presented results show that Complex Detection and Subtraction via Wavelet (CDSW) can be a highly efficient tool for the study of atrial arrhythmias in those systems with reduced number of leads, like Holter recording systems

Feasibility and performance of methods based on statistical signal processing to study atrial fibrillation

In order to use the ECG as a tool for atrial fibrillation (AF) analysis, we need to separate the atrial activity (AA) from other cardioelectric signals. In this matter, some statistical signal processing techniques, such as Blind Source Separation (BSS), are able to perform a multi-lead statistical analysis of the ECG with the aim to obtain a set of independent sources where the AA is included. BSS techniques can be divided in two groups depending on the mixing model. Firstly, in algorithms based on Independent Component Analysis (ICA) instantaneous mixture of the sources is assumed. Secondly, in convolutive BSS (CBBS) algorithms the more realistic case of weighted and delayed contributions in the generation of the observed signals is considered. In this paper, a comparison between the performance of ICA algorithms and CBSS algorithms in the extraction of the AA in AF episodes is developed