The Research On J Wave Diagnosis Techniques Based On Support Vector Machine

The emergence of J wave on the ECG signal indicates that there would occur a series of fatal heart diseases such as malignant arrhythmia, myocardial infarction, sudden cardiac death and so on. How to quickly and accurately detect the J wave and give the appropriate treatment to reduce the mortality of patients is a difficult problem to be solved in the medical field. At present, the judgment of whether J-wave exists in most of the present researches relies only on the experience of the clinical doctors, which not only are time-consuming but can hardly avoid misdiagnoses.SVM has been widely used in the field automatic diagnosis such as arrhythmia, myocardial infarction, epilepsy, children with attention deficit and hyperactivity. Support vector machine is applied to J wave automatic classification in this paper, which broaden the application range of the support vector machine in the medical field. This paper present two algorithms for automatic detection of J wave based on Support Vector Machine and the specific content is as follows:The first algorithm mainly has three steps including the feature selection,dimensionality reduction and classification. In the feature selection process,curve fitting, wavelet transform, HRV analysis, waveform analysis are mainly used. Principal component analysis is applied in the dimensionality reduction.In the training phase of classification, a small number of similar support vectors are found through similarity. In the testing process, similar support vectors are used to retrain the SVM and test its effectiveness. Simulations show that proposed algorithm has high recognition rate when compared with the traditional SVM algorithm.The second algorithm principally include feature selection, feature dimension reduction and the selection of optimal SVM parameters. In feature selection, this paper puts forward the global features and local features based on the first algorithm. By using independent component analysis method, the feature vectors are obtained. In the selection of parameters, this algorithm proposes an adaptive method for kernel function parameter based on the variable step size method. The simulation results show that the algorithm’s overall accuracy can reach 96.1%.