Evaluating the performance of state of the art algorithms for enhancement of seismocardiogram signals

Abstract

Seismocardiography is a new, low cost and non-invasive method for measurement of local vibrations in the sternum due to cardiac activity. Signals recorded using this procedure are termed as seismocardiogram or SCG signals. Analysis of SCG signals provides information about the functionality of the cardiovascular system. Performing an automatic diagnosis using SCG signals involves the use of signal processing, feature extraction and learning machines. However for such methods to yield reliable results, the digitally acquired SCG signals should be accurately denoised and free from artifacts. In this paper, we evaluate the performance of state of the art algorithms in denoising these signals. In our work, clean SCG signals were corrupted with additive white Gaussian noise and the signals were further denoised. Denoising using wavelet transforms, empirical mode decomposition, adaptive filters and morphological techniques has been considered in our work. Standard metrics: mean squared error (MSE), mean absolute error (MAE), signal to noise ratio (SNR), peak signal to noise ratio (PSNR), cross correlation (xcorr) and CPU consumption time have been computed to assess the performance the aforementioned techniques. From our study it is concluded that wavelet thresholding yields the best denoising and is hence the most suitable method for enhancement of real world SCG signals.