Optimal adaptive clutter filtering in color flow imaging

Regression filters provide an efficient method for clutter rejection. The clutter is removed by subtracting the projection of the N-dimensional signal vector into a K-dimensional subspace containing the low frequency clutter signal. With a predefined set of basis vectors, e.g. the Legendre polynomials, the filter cut-off frequency can be adjusted by varying K. In this study the correlation matrix of the clutter signal is estimated by spatial averaging. The eigenvectors corresponding to the K largest eigenvalues are an optimal basis for the clutter space, and the corresponding filter have maximum clutter attenuation. Since stationarity is not assumed, the filter will also adapt to nonlinear tissue movement. Nonlinear movement often occurs when the operator moves the probe in search for blood vessels, and causes artifacts when a standard clutter filter is used. Digital RF data from the carotid artery was recorded with transmit frequency 5.7 MHz and PRF=5 kHz. Doppler angio images were generated by displaying the power of the filtered signal as gray scale images. In a frame with extensive probe movement, the clutter to blood ratio was measured by comparing the mean power in a region with strong tissue echo and regions inside blood vessels. The signal-to-clutter ratio was significantly improved by the adaptive filter. With N=9 and K=3 the improvement was 12.9 dB