Applying Novel Time-Frequency Moments Singular Value Decomposition Method and Artificial Neural Networks for Ballistocardiography. (12th December 2006)

• Record Type: Journal Article
• Authors: Akhbardeh Akhbardeh, Alireza Alireza; Junnila Junnila, Sakari Sakari; Koivuluoma Koivuluoma, Mikko Mikko; Koivistoinen Koivistoinen, Teemu Teemu; Värri Värri, Alpo Alpo
• Other Names: Mulgrew Mulgrew Bernard Bernard Academic Editor.
• Is Part Of: EURASIP journal on advances in signal processing. Volume 2007(2007)
• Publication Date: 2006-12-12
• Subjects: Signal processing -- Periodicals; Traitement du signal; Signal processing; Periodicals; 621.3822
• DOI: 10.1155/2007/60576
• Languages: English
• ISSNs: 1687-6172

Abstract

Abstract : As we know, singular value decomposition (SVD) is designed for computing singular values (SVs) of a matrix. Then, if it is used for finding SVs of anm -by-1 or 1-by-m array with elements representing samples of a signal, it will return only one singular value that is not enough to express the whole signal. To overcome this problem, we designed a new kind of the feature extraction method which we call ''time-frequency moments singular value decomposition (TFM-SVD).'' In this new method, we use statistical features of time series as well as frequency series (Fourier transform of the signal). This information is then extracted into a certain matrix with a fixed structure and the SVs of that matrix are sought. This transform can be used as a preprocessing stage in pattern clustering methods. The results in using it indicate that the performance of a combined system including this transform and classifiers is comparable with the performance of using other feature extraction methods such as wavelet transforms. To evaluate TFM-SVD, we applied this new method and artificial neural networks (ANNs) for ballistocardiogram (BCG) data clustering to look for probable heart disease of six test subjects. BCG from the test subjects was recorded using a chair-like ballistocardiograph, developed in our project. This kind of device combined with automated recording and analysis would be suitable for use in many places, such as home, office, and so forth. The results show that the method has high performance and it is almost insensitive to BCG waveform latency or nonlinear disturbance.