Quadrature Doppler Ultrasound Signal Denoising Based on Matching Pursuits with Different Time-Frequency Dictionaries

The spectral analysis of the Doppler ultrasound signal has been widely used in medical diagnosis. The additional frequency components arising from noise will produce an adverse effect on the subjective study of the spectrogram and its quantitative analysis. To remove noise effectively, matching pursuits with three different time-frequency dictionaries including Gabor atoms (GMP), cosine packets (CMP), and wavelet packets (WMP) are proposed to remove noise from the quadrature Doppler ultrasound signal in this paper. The simulation results have shown that the GMP method can be best adapted to approximate the Doppler ultrasound signal with fewer atoms compared with the CMP and WMP methods, and this approach is also superior to the translation-invariant wavelet transform method. In contrast to the translation-invariant adapted local cosine transform, although the GMP method has achieved approximate performance improvements with higher computation intensity, it offers the potential advantages in pulsed Doppler ultrasound application because it have no requirement for the signal over-sample, which is needed in the traditional transform-based denoising methods in order to obtain a more precise estimation of the noise level.