Independent Component Analysis Based Research On Extracting Atrial Fibrillation Signals

Independent Component Analysis(ICA) is a multi-channel signal processing method derived from Blind Source Separation which elicits the hidden component of random variables and signals,helps to achieve the enhancement and analysis signals.The purpose of ICA is to recover the N unknown source signals from the M mixed signals without any other prior information.This thesis faciliates the ICA into AF (Atrial Fibrillation) signal.AF is the most common arrhythmia, that is currently, one of the most difficult to capture the heart diseases in the field of arrhythmia. On the surface of AF patients,it has signal including reflecting atrial activity, that is, atrial fibrillation waves (f waves).It contains a wealth information of the atrial characteristics structure and the physiological, pathological conditions, which are clinically significant Potential applications. Therefore, this thesis faciliates the ICA to extract the f-wave, in order to provide credible data for later clinic diagnoses.The main emphases of this thesis could be concluded into following steps:①Based on the definition and hypothesis of ICA, we firstly analyse its basic mathematic model[1]; its basic principles; previous data processing: patterns filter[2], meaning and whiting; independence criterion and its optimized principles; I also describe common algorithms of ICA[3];②I pre-process the signal using the methods below: meaning and whitening to remove the correlation between the datas.Then use FICA into this ECG signal separation to extract features of AF signals. I process the distilled signals. Using combinating patterns filter to remove baseline drift and high-frequency noise and using wavelet threshold[5] to de-noise the AF signals and finally acquire excellent clean ECG signals;③Making qualitative analysis of the experimental results, firstly I draw the atrial fibrillation signal histogram in order that AF is sub-Gaussian signal, which is in line with the ICA assumptions. Then I make AF spectrum estimated map, by it, I find the main peak of the AF-wave frequency in the 3-10Hz[4].The data has important significance for the study of the atrial fibrillation pathology. Finally, in order to illustrate FICA algorithm in the extraction of the AF signal characteristics better than other algorithms,using icalab toolbox to draw the separation of graphics in the other algorithms; ④I make a summary for this paper, and description the bright spots and improvements in this article.