PCA AND KPCA OF ECG SIGNALS WITH BINARY SVM CLASSIFICATIONAuthor:Maya Kallas, Clovis Francis, Lara Kanaan, Dalia Merheb, Paul Honeine, Hassan AmoudSource:Signal Processing Systems (SiPS), 2011 IEEE WorkshopAdvisor:Yin-Fu HuangStudent:YU-HSIEN CHO
OUTLINEI. AbstractII. IntroductionIII. MethodsIV. Experiment and ResultV. Conclusion
I. Abstract Cardiovasculardiseases remain the primary cause of death around the world. In Lebanon, 63.13% of men and 56.57%of women aged 50 years and above die from heart diseases. For effective diagnostics, the study of the ECG signal must be carried out for several hours. Automatic detection and classification of electrocardiogram (ECG) signals are important for diagnosis of cardiac irregularities.
II. Introduction In this paper, we propose to combine the Support Vector Machines used in classification on one hand, with the Principal Component Analysis used in order to reduce the size of the data by choosing some axes that capture the most variance between data. On the other hand, with the kernel principal component analysis where a mapping to a high dimensional space is needed to capture the most relevant axes but for nonlinear separable data. The efficiency of the proposed SVM classification is illustrated on real electrocardiogram dataset taken from MIT-BIH Arrhythmia Database.
Training data sets:MIT-BIH arrhythmia database inputs:10 normal signals, 10 PVC signals, 14 LBBB signals 4Classes: TP:abnormalabnormal , FP:normalabnormal FN:abnomalnormal , TN:normalnormal
III. Methods PCA feature extraction: The analyzed signal is a single lead ECG which has been converted into a data matrix X. where M is the number of beats Step1. Getting the data matrix X Step2. Subtract the mean Step3. Calculate the covariance matrix Step4. Calculate the Eigenvectors/Eigenvalues of the covariance matrix
Step5. Choosing principal components and forming feature vectorStep6. Deriving the new data set :Final Data = (row feature vector) ⋅ (row data adjust)Step7. Calculate the reconstruction
KPCA feature extraction : The basic idea of KPCA is to map the original data into a high dimensional space via a specific function and then to apply the standard PCA algorithm on it. Since the ECG signal has a lot of nonlinear structures and since the PCA extracts only the linear ones, we adopted KPCA to extract the nonlinear components.
IV. Experiment and Result The kernel functions used are the linear, polynomial and the Gaussian, where x and y refer to vectors of x i and y i components respectively. Table 1 shows the parameters considered as best models.kernel parameters σ=0.8,regularization constant C =100
Experiment and Result Resultsof applying the Gaussian classifier on the data is presented in Figure 1.
Experiment and Result As we can see in Table 2, the results achieved by the Gaussian classifier were better than those achieved by the SVM linear and the SVM-poly.
Experiment and Result PCA uses linear transformation to convert a large amount of correlated variables to a smaller number of uncorrelated principal components that preserve most of the useful information. See the Figure 2
Experiment and Result Figure3 illustrates the results of applying the binary SVM combined with KPCA. So we can conclude that for binary ECG classification the most convenient method is combining SVM with KPCA.
Experiment and Result Thebest results obtained in SVM by using the original input without feature extraction, and by using PCA (σ=10, C =40) and KPCA (σ=0.5, C =10) feature extraction are given in table 3.
V. Conclusion Electrocardiogram (ECG) supervising is the most important and efficient way for preventing heart attacks. Our work presents an integrated classification method, which combines the Support Vector Machine (SVM) with either the Principal Component Analysis (PCA) or the Kernel PCA (KPCA) for classification of different types of cardiac abnormalities. Our experiments show that binary SVM combined with feature extraction using PCA or KPCA greatly improves the quality of classification than that without feature extraction.