Fusion of Ultrasound Blood Flow And Color Flow Imaging

Zhengjuan Fan, Chaowei Tan, Dong C. Liu

Introduction

Color flow imaging (CFI) is an important and widely-used noninvasive technique for Doppler ultrasound blood flow detection. Conventional ultrasound CFI uses auto-correlation function with temporal lag equal to one to estimate the velocity component along the beam direction at each pixel within the region of interest. Then the velocity estimates are color coded and displayed on the top of the gray scale tissue image. Since CFI is based on the Doppler principle, it only allows CFI to detect the velocity component along the ultrasound beam, not to provide other directions flow velocity information. Recently, Lasse Lovstakken et. proposed blood flow image (BFI), which can provide qualitative information of the blood flow distribution and movement in any direction of the image.

The main purpose of this paper focuses on the implementation of vascular imaging algorithms as a fusion of CFI and BFI imaging. We have made the following improvements: using regression filters instead of FIR filters as wall filtering to increase the frame rate in BFI; optimizing the implementation of the regression filtering as an effective way to remove clutter/tissue signals both in CFI and BFI, taking flexible dynamic compression strategies to get better image in BFI; and proposed color-coding methods to combine the results from CFI and BFI to achieve good visualization of blood flow both in quantitative and qualitative.

Overview of Methods

Data acquisition

The signal data acquisition is basically the same as that in the conventional CFI: an axial line of velocity estimate is calculated by forming 6 to 24 lines at the same direction. The set of 6 to 24 lines is regard as an ensemble and the number of lines will be referred to as the ensemble size (Ne). The number of sample point along the axial line is Ns, the number of scan lines is noted as Nl. The flowchart of the proposed implementation is shown in Fig. 1.

High Pass Filtering

High pass filter is necessary to filter out the blood flow signal from stationary or slow moving tissue. In the high pass filtering block, we use two kinds of filtering technique and compare their performance: the finite impulse response filter (FIR) and the regression filter.

  • FIR filtering

Based on Bjarum's method, the minimum FIR filtering is utilized as the follows,

where L the number of taps of FIR filter, the filter’s impulse response h(l) is not symmetric, to reduce the estimator variance, the flow parameter should be estimated based on two output vectors filtered in the forward and backward direction. s(m) notes the output of the FIR filtering. The valid number of filter output samples that could be used is Nw = Ne – L + 1.

  • Regression filtering

Regression filter is based on the assumption that their inputs are treated as polynomial functions in the time domain, and then the Doppler signal contributed from the blood-flow could be retrieved as a residual from a regression process. We can use a given order polynomial to approximate the signal scattered from the tissue. Subtract the given order polynomial from the inputs and then get the output of regression filtering. Mathematically, we can write.

where x(n) and y(n) represent the input and output of the filter and, ai is the i-th coefficient of the approximate polynomial. Apply the standard least-squares fitting to calculate ai, as follows,

To improve computational efficiency and engineering applications, the regression filtering procedure could be simplified as follows,

The frequency responses of several FIR filters and regression filters for comparison in Fig. 2,

Envelope Detection

The envelope detection method is basically the same as in conventional B mode. This block will produce several frames speckle images (for FIR filtering, Nw frames, for regression filtering, Ne frames), that provide the blood flow movement by tracked from frame to frame.

Dynamic Compression

1. Compress high-bit data to low-bit data representation and bring weak signals to visible grey level for display. The standard form of the log compression is as, P = D×log(A) + G, where D controls the dynamic range, G is the compression gain, A and P is the input and output. Assuming a d dB dynamic range and additional g dB compression gain to bring signals at –g dB grey level to the full brightness scale of the display (d>0, g>0), the following equations could be obtained,

2. Because B-mode image data and speckle images data (the outputs of BFI) have different ranges. We must set different dynamic compression parameters to ensure both weak and strong echoes can be visualized. For B-mode image, compress 15-bit B-mode data to 11-bit data representation; for speckle image, compress 11-bit data to 8-bit data representation. And, set D = 60 dB, G = 0 dB in our test.

Display

The previous processing could produce several frames of speckle images (Nw frames for FIR filtering; Ne frames for regression filtering) that provides qualitative information of the blood flow distribution and movement in any direction of the image (B mode image provides the tissue echo magnitude information, color flow image provides quantitive information of the blood flow velocity component along the beam line). And then, two color-coding methods to combine the quantitative information of CFI and the qualitative information of BFI for visualization.

  1. Based on the HSV color scheme, set the value of H and S components of result image using the value of H and S components of CFI image; set the value of V component using the amplitude of the BFI speckle image. The flow chart is shown in Fig.3.

  2. Construct a two dimensional color look-up table: in its lateral direction, there are 64 columns that represent 64 levels of different qualitative information from BFI; in its axial direction, there are 256 rows that represent 256 levels of quantitative information from CFI.

Experiment and Results

The signal data is acquired from a healthy carotid artery in a digital ultrasound scanner using the following parameters: 85 san lines (Nl), 508 samples (Ns) along the san line, and an ensemble size (Ne) of 16.

Based on the above High Pass Filtering section, we choose a third order regression filter and a cut frequency ten FIR filter for test and work for CFI and BFI purposes. Compare Fig.4b and Fig.4c, the regression filter can reserve lower velocity information than FIR filter. What's more, for FIR filter, the number of samples after the high-pass filtering process is reduced to Nw; for regression filter, the number of samples is still Ne. So the frame rate of speckle images in BFI is improved by regression filtering instead of FIR filtering.

The output of combing BFI and CFI is a sequence of images. We provide three original images at different positions (first, middle and last) of the sequence. Fig.5 and Fig.6 show the outputs of the first (HSV mode) and the second (2D color map) display methods, respectively. Comparing Fig.5 with Fig.6, the 2D color map method provides better visualization of the blood flow information.

Conclusion

    1. Presenting the implementation of combining the output of CFI and BFI in vascular imaging to provide qualitative and quantitive blood flow information. Helpful for doctor to make more accurate diagnose.

    2. It would be possible to estimate the blood velocity in any direction by other techniques, such as compound Doppler scanning technique, two-dimensional speckle pattern tracking technique, and lateral coherent processing technique.

    3. It also would be possible to integrate the output of BFI with other ultrasound imaging modes, such as Power Doppler mode.

    4. It would be applied in other clinical applications, such as cardiac imaging, abdominal imaging.