Application of an Improved Fisher Criteria in Feature Extraction of Similar ECG Patterns

The abnormal changes of myocardial infarction (MI) appeared in Electrocardiogram (ECG) are low-level signals. The patterns to represent MI ECGs are usually extremely similar between different classes. In addition, the using of conventional 12-lead ECG generates large amounts of time-series data. Conventional Linear (Fisher) discriminant analysis (LDA) faces the problems of singular matrix and limited number of the extracted features. An improved Fisher criteria (IFC) based method was employed to discriminate ECG's in current study. The singular matrix problem could be overcome, and more features could be extracted at the same time. The data in the analysis including healthy control (HC), MI in early stage (MIES) and acute MI (AMI) were collected from PTB diagnostic ECG database. The results show that the proposed method can obtain more effective features, and classification accuracy based on IFC can be improved than that of conventional LDA based method.