Research On Characterization Of Atrial Fibrillation And Auxiliary Clinical Diagnosis Based On Surface Electrocardiogram

Atrial fibrillation (AF) is the most common sustained arrhythmia and difficultly solved cardiac disease. Currently, our country is one of the most quantities of AF patients. With the development of life and population becoming old, the episodes of AF are increasing year by year and AF will be one of the most popular heart diseases in 21st century. Unlike ventricular fibrillation (VF), which is invariably fatal if it is not interrupted, AFs can impair hemodynamics, destroy the cardiac function and especially increase the possibility of brain stroke by thromboembolic formation. Until now, AF treatment maybe viewed as being based on trial and error, since no test is able to predict the natural history of the arrhythmia and its response to treatment. So, the effect of therapy is very low.Surface electrocardiogram is a simple and excellent repeatable noninvasive record of cardiac ecletric activities, which has been holding particular action and important usefulness in diagnosis the cardiac function disease. Atrial fibrillatory signals (f waves) are hidden into the surface ECG of AF patient and contain abundant information related to atrial structural features and pathophysiological mechanisms. So, by applying modern signal processing technique to analysis surface ECG, the usable characteristic indices reflecting atrial activity are expected and can afford a rather efficiently noninvasive way for the administrations of AF patients and lead clinician to make individual decisions.The most contributions of this thesis are as follows:Isolating atrial activity from ventriclar activity, usually using some QRST cancellation technique, is the premise and basement of characterization of atrial fibrillatroy waves, In this thesis, two kinds of cancellation method were studied: independent component analysis (ICA) and principal component analysis (PCA). The first one mainly took use of space information of source signals and adapted to multi-lead ECG records which usually tracked persistent AF. The other one mostly was used for processing proxysmal AF (PAF) which was recorded by one-lead systems. The components corresponding to atrial and ventricular sub-spaces were separated respectively by un-correlating the different cardiac beats in different time segments. The two methods set a strong basis for the following works.On the stage of features extracting, two most essential features of atrial fibrillatory signals, main peak frequency (MPF) and instantaneous frequency (IF), are computed. The respective phsiological fundamentals and clinical significances of them were disscussed in detial. On the basis of comparing benefits with disadvantages of various traditional techniques of power spectrum estimation and time-frequency analysis, Hilbert-Huang Transform (HHT) was introduced to estimation the MPF andrelative parameters of AF signals for the first time. Meanwhile, a new technique of time-frequency representations, we named it hybrid linear-quadratic time-frequency distribution was represented. This method combined the virtues of Gabor extension and quadratic Wigner distribution. Then this method was used for extraction the two quantitative attributions: IF and relative educed measures, which reflected the dynamic changes of AF.Spontaneous termination prediction of PAF was researched in detail. The spontaneous terminative action of PAF is commonly a precursor to the development of persistent AF. This prediction could help clinician to intervene in affected individuals in advance and increased the likelihood of self-termination of what would otherwise develop into sustained AF. Because of a little number of samples and individual differences, support vector machines (SVMs) were introduced to classify (or predict) the self non-terminating AF with self-terminating, and AF terminating at once with AF terminating after one minute. Additionally, problems about the parameters optimization of SVM, the choices of input characteristic variable and relative decisions were explored for acquiring highest classification accuracy.It has been proved by a lot of clinical tests and results that the studies contributed by this thesis could not only bring new methods to signal processing technique, but also get better prediction accuracy ratio in differentiating varies types of AF. So, they could be applied to auxiliary clinical diagnoses of AF, and could provide a feasible noninvasive method for understanding mechanisms and effect evaluations of treatment of AF.