Cardiology knowledge free ECG feature extraction using generalized tensor rank one discriminant analysis

Abstract

Applications based on ElectroCardioGram (ECG) signals feature extraction and classification are of a major importance to autodiagnosis of heart diseases. Most of the existing work on ECG classification method only targets one or two-leads ECG signals. This limitation results from the unavail- ability of real clinical 12-leads ECG data that would help training the classi- fication models. Through this work, we propose a new tensor based scheme, which is motivated by the lack of effective feature extraction method for di- rect tensor data input. In this scheme, an ECG signal is represented by a 3rd-order tensors in the spatial-spectral-temporal domain after using Short Time Fourier Transformation (STFT) on the raw ECG data. To overcome the limitations of the Tensor Rank One Discriminant Analysis (TR1DA) inherit- ed from the Linear Discriminant Analysis (LDA), we introduce a Generalized Tensor Rank One Discriminant Analysis (GTR1DA). Note that this approach takes into consideration the distribution of the data points near the classifica- tion boundary in order to calculate better projection tensors. The experiment results show that the proposed method performs better in terms of classifi- cation accuracy compared to other vector and tensor based methods. Finally GTR1DA features a better convergence property than the original TR1DA.