Exploiting locality based Fourier transform for ECG signal diagnosis

Computerized analysis of electrocardiography (ECG) signals can aid early detection of heart disorders. For analysis, physicians use signal changes evident in ECG signals, such as, time intervals, shape of peaks, predefined peak compositions, etc., all of which require ECG signal information. The most important part of computerized ECG signal classification is to produce the discriminative features properly to identify heart disorders. Proposed locality based Fourier transform based method is a powerful solution for feature extraction which exploits local information embedded in the signal. Extraction of proper set of features is critical for successful execution of the local Fourier transform. In this paper, we present experimental results with different combination of Fourier coefficient obtained from MIT-BIH database. We compare our method with similar approach obtained by principal component analysis (PCA) features to prove classification accuracy.