Research On Discriminant Analysis For Muti-lead ECG

This paper proposes an electrocardiogram (ECG) pattern classificationmethod for12-lead ECG using Semi-supervised Discriminant Analysis(SDA).The research on discriminant analysis for ECG mainly includes thepre-processing, feature extraction, rank reduction and classification forECG. The pre-processing for raw ECG data includes noise elimination, Rpeak detection and ECG signal segmentation. The ECG feature is extractedby wavelet transform. In rank reduction respect, we introduce the classicallinear discriminant analysis (LDA) method and the concept of SDA. SDAcan make use of the structure information which is contained in unlabeleddata, and can also improve its ability of generalization by using regularizer.So SDA can be used for the rank reduction of ECG feature. At last, supportvector machine is used for ECG feature classification.The experiments show the rank reduction performance of SDA is betterthan other common rank reduction method. And the ECG classificationmethod based on SDA can achieve high classification accuracy.As an application, this paper also implements a remote ECG diagnosticplatform, and tries to apply the classification method in it, so that theplatform can provide the auxiliary decision service for doctors andimproves the quality of the ECG diagnosis. With this platform, it onlycosts a short time from collect the ECG data to get ECG diagnosis.