The Study Of Ventricular Fibrillation Nonlinear Detection

Ventricular Fibrillation is one kinds of malignant arrhythmia. It may cause suddencardiac death (SCD) without any sign before it. According to statistics by AHA(American Heart Association), there are about 30 percent of people died of a suddendeath, who have no or never have heart disease. Almost all of them died outside ofhospital and the opportunity of subsistence rate can be improved if the disease can bedetected ahead and then treatment was carried out immediately. Therefore, it is animportant topic of forecasting a sudden death and life-threatening ventriculararrhythmia. In this paper, we make the detection of ventricular fibrillation technologythe main content.According to studies, heart dynamic is complex and nonlinear, so we can analyseselectrocardiograph (ECG) effectively by using nonlinear dynamics and nonlinearmathematical model. In this paper, we illustrated several classical VF detectionmethods and proposed a nonlinear algorithm which is based on ECG characteristic andsupport vector machine (SVM) theory. The new algorithm was realized as following:firstly, reprocess ECG signal; secondly, extract features of ECG signal in everywindow by using 4s sliding window technology; lastly, input these features into binaryclassification support vector machine and output results of classification. The data setswere taken from the BIH-MIT database and the CU database. Here, we classify VF andnon-VF signal and do not discriminate VF and ventricular tachycardia (VT).For the new algorithm, three algorithms were used to extract ECG signal featureswhich are threshold crossing intervals (TCI) algorithm, time delay (TD) algorithm andHilbert transform (HILB) algorithm. For the three algorithms, we make someimprovements. As for TCI algorithm, it was improved in two aspects: the length ofsliding window and the threshold. The improved TCI algorithm utilized absolute threshold and the technology of 4s length sliding window , which can make thedetection and classification much better by increasing characteristic gap betweenventricular fibrillation and non-ventricular fibrillation. It calculated average thresholdcrossing interval TCI value of the middle 2s in every 4s-length sliding window andmake the TCI value as feature parameter of ECG signal. For TD algorithm, the lengthof sliding window was set to 4s instead of 8s, which can enhance the real-time qualityand reduce the complexity of the new algorithm. For Hilbert transform algorithm, itwas the first time that Hilbert transform algorithm was used to extract ECG signalfeatures and we gave a detail introduction in this paper. In all, by doing the threealgorithms some improvement to achieve more real-time, we can effectively extractcharacteristics of ECG signal.The technique of SVM is a new machine learning technique developed basing onstatistical learning theory. It was proposed essentially for classification problems oftwo classes, including linearly separable cases and non-linearly separable cases. It hasgreat advantages in processing problems, such as classification and pattern recognition.We utilized kernel function support vector machines for nonlinear classification. Thekernel function was chose as radial basis function (RBF), which is used widely and hasless parameter. The support vector machine methods were simulated on SVM packageof Lu Zhenbo (SVM_luzhenbo&LS_SVMlab) and the penalty factor C and radialwidthσof kernel function were obtained by trial and error method.The new nonlinear algorithm, which is based on ECG signal characteristic andSVM theory, realized the detection of VF signal successfully. By comparing thesensitivity, specificity, positive predictability and accuracy with other well knownmethods, the conclusion was made that this method is superior to other methods. It iseasier to implement and has better performance in real-time execution. Theseadvantages make it more suitable in real time ECG monitoring and defibrillator.