Blood Flow Signal Extraction Methd In Ultrasound Color Flow Imaging

As a non-invasive vascular visualization tool, ultrasound color flow imaging is safe and comparatively cheap. The low frequency and strong clutter, considered as echoes originating from soft tissue, causes the estimated mean frequency to be biased towards to the clutter rather than that of blood flow signal and leads to an increasing risk of misdiagnosis in cardiovascular disease. An accurate and efficient extraction method of blood flow signal is expected. There are two challenging issues to difficult accurate estimation of blood flow velocity. Firstly, the available pulse number of Doppler signal is usually 8-16 to provide a sufficient frame rate. Secondly, due to effect on breath and heartbeat, the clutter is usually wideband signal. The conventional extraction methods of blood flow signal rely to their predetermined parameters which are independent to received ultrasound echoes. Consequently, these non-adaptive methods usually lead to a reduction on estimation accuracy. To break above mentioned constraint, the non-parametric and parametric adaptive blood flow signal extraction methods are discussed in this work.The conventional extraction methods based on high pass filters usually cause blood flow velocity estimation biases since clutter rejection filters often fail to suppress clutter adequately or distort parts of blood flow signals. We propose a parametric estimation framework called relaxation(RELAX) to directly extract the blood flow information from raw ultrasound demodulated signals. RELAX constructs an exponential model to approximate ultrasound echoes and solves its parameters in a decoupled manner. RELAX can reject clutter efficiently and eliminate the effect caused by white noise simultaneously. A series of the simulations shows RELAX achieves a significantly improvement on frequency estimation and variance of blood flow signal over high pass filters. Comparing with parametric methods, RELAX improves estimation accuracy of clutter and low velocity blood flow signal at a feedback manner. When the clutter is wideband signal, parametric methods are constraint with mathematical model and cause estimation bias.To solve the problem that reduction on estimation accuracy is usually observed when clutter is wideband signal for parametric methods, a clutter filter based singular value decomposition Hankel matrix(Hankel-SVD), which simulates the M-E behaviors with only S-E samples was introduced. To solve the problem that parts of blood flow signal are mistakenly suppressed in Hankel-SVD, we built a bilinear Hankel(B-Hankel) SVD framework. It achieves accurate signal decomposition and decreased compute complexity significantly. Simulations results show that B-Hankel filter reduced the estimation bias of the blood flow over the conventional Hankel one. To ease the constraint condition of nonsingular property of autocorrelation matrix, a novel clutter filter design framework based on forward-backward subsequence smoothing(FBSS) as an extension to the prevalent eigen-based method is proposed. In the FBSS algorithm, a fixed-length segment of each S-E sample is considered as a short conventional S-E vector. By moving forward and backward one sample, several vectors of the same-length can be created and considered as a simulated M-E samples. The pulse size needed to guarantees a nonsingular property of autocorrelation matrix in the FBSS framework is three-quarter of that used in the Hankel-SVD method. As a consequence, FBSS leads to a good discrimination between the clutter and blood flow components. Based on the results obtained from a series of simulations, the proposed FBSS method is shown to achieve a superior performance to the state-of-the-art regression and Hankel-SVD filters in estimating the blood flow velocities when pulse number is limited.To solve the difficulty on determination of frequency and eigenvalue threshold in the Hankel framework and eigen-based method, we introduce two adaptive decomposition methods——empirical mode decomposition(EMD) and ensemble EMD(EEMD). To solve the problem that caused by mode mixing and unique weight, was introduced, in our work, the conventional EEMD is extended and a novel multi-variable regression model based EEMD framework is built. Ridge regression and least absolute shrinkage selection operator(LASSO) are introduced at an optimization procedure during ensemble combination. The non-uniform and trail-dependent weights reduce the effect that the outliers usually lead to an overall poor performance. A theoretic result demonstrates that regression EEMD have an ability to solve the mode mixing problem frequently encountered in EMD with adequate noise strength when separating a composite two-tone signal. The compute complexities of these two novel clutter filters with M-E called as R-EEMD and LASSO EEMD depend on the containing EMD algorithm. In a series of simulations, the proposed methods achieve a significant improvement on blood flow velocity estimation over the eigen-based filter of best-fit order, EMD and EEMD based methods.To achieve the application in the real ultrasound equipment, we provide an in-depth analysis of compute complexity for proposed methods. According to clinical experiments on carotid artery data in which clutter is narrowband signal, numerical results verify that RELAX achieves maximum blood-to-clutter energy ratio, and the maximum vascular energy is observed with LASSO EEMD method. Clinical experiments on kidney data show that, RELAX method fails to access to the real echoes of small veins and causes large estimation bias for these vessels. Comparing with the conventional regression filters, B-Hankel SVD and FBSS-eigen methods proposed adaptive methods achieve higher blood flow estimation accuracy and sensitivity The Comparing with all competing methods, LASSO EEMD achieves best hierarchy for large vessel and most vein vessel detail.