Rapid estimation of blood velocity and visualization of complex flow patterns are important for clinical use of diagnostic ultrasound. This paper presents real-time processing for two-dimensional (2-D) vector flow imaging which utilizes an off-the-shelf graphics processing unit (GPU). In this work, Open Computing Language (OpenCL) is used to estimate 2-D vector velocity flow in vivo in the carotid artery. Data are streamed live from a BK Medical 2202 Pro Focus UltraView Scanner to a workstation running a research interface software platform. Processing data from a 50 millisecond frame of a duplex vector flow acquisition takes 2.3 milliseconds seconds on an Advanced Micro Devices Radeon HD 7850 GPU card. The detected velocities are accurate to within the precision limit of the output format of the display routine. Because this tool was developed as a module external to the scanner's built-in processing, it enables new opportunities for prototyping novel algorithms, optimizing processing parameters, and accelerating the path from development lab to clinic.
FDA requires that intensity and safety parameters are measured for all imaging schemes for clinical imaging. This is often cumbersome, since the scan sequence has to broken apart, measurements conducted for the individually emitted beams, and the final intensity levels calculated by combining the intensities from the individual beams. This paper suggests a fast measurement scheme using the multi-line sampling capability of modern scanners and research systems. The hydrophone is connected to one sampling channel in the research system, and the intensity is measured for all imaging lines in one emission sequence. This makes it possible to map out the pressure field and hence intensity level for all imaging lines in a single measurement. The approach has several advantages: the scanner does not have to be re-programmed and can use the scan sequence without modification. The measurements are orders of magnitude faster (minutes rather than hours) and the final intensity level calculation can be made generic and reused for any kind of scan sequence by just knowing the number of imaging lines and the pulse repetition time. The scheme has been implemented on the Acoustic Intensity Measurement System AIMS III (Onda, Sunnyvale, California, USA). The research scanner SARUS is used for the experiments, where one of the channels is used for the hydrophone signal. A 3 MHz BK 8820e (BK Medical, Herlev, Denmark) convex array with 192 elements is used along with an Onda HFL-0400 hydrophone connected to a AH-2010 pre-amplifier (Onda Corporation, Sunnyvale, USA). A single emission sequence is employed for testing and calibrating the approach. The measurements using the AIMS III and SARUS systems after calibration agree within a relative standard deviation of 0.24%. A duplex B-mode and flow sequence is also investigated. The complex intensity map is measured and the time averaged spatial peak intensity is found. A single point measurement takes 3.43 seconds and the whole sequence can be characterized on the acoustical axis in around 6 minutes.
This paper presents a new design of a discrete time Delta-Sigma (ΔΣ) oversampled ultrasound beamformer which integrates individual channel apodization by means of variable feedback voltage in the Delta-Sigma analog to digital (A/D) converters. The output bit-width of each oversampled A/D converter remains the same as in an unmodified one. The outputs of all receiving channels are delayed and summed, and the resulting multi-bit sample stream is filtered and decimated to become an image line. The simplicity of this beamformer allows the production of high-channel-count or very compact beamformers suitable for 2-D arrays or compact portable scanners. The new design is evaluated using measured data from the research scanner SARUS and a BK-8811 192 element linear array transducer (BK Medical, Herlev, Denmark), insonifying a water-filled wire phantom containing four wires orthogonal to the image plane. The data are acquired using 12-bit flash A/D converters at a sampling rate of 70 MHz, and are then upsampled off-line to 560 MHz for input to the simulated ΔΣ beamformer. The latter generates a B-mode image which is compared to that produced by a digital beamformer that uses 10-bit A/D converters. The performance is evaluated by comparing the width of the wire images at half amplitude and the noise level of the images. The ΔΣ beamformer resolution has been found to be identical to that of the multi-bit A/D beamforming architecture, while the noise floor is elevated by approximately 6 dB.
This paper presents 3D vector flow images obtained using the 3D Transverse Oscillation (TO) method. The method employs a 2D transducer and estimates the three velocity components simultaneously, which is important for visualizing complex flow patterns. Data are acquired using the experimental ultrasound scanner SARUS on a flow-rig system with steady flow. The vessel of the flow-rig is centered at a depth of 30 mm, and the flow has an expected 2D circular-symmetric parabolic profile with a peak velocity of 1 m/s. Ten frames of 3D vector flow images are acquired in a cross-sectional plane orthogonal to the center axis of the vessel, which coincides with the y-axis and the flow direction. Hence, only out-of-plane motion is expected. This motion cannot be measured by typical commercial scanners employing 1D arrays. Each frame consists of 16 flow lines steered from -15 to 15 degrees in steps of 2 degrees in the ZX-plane. For the center line, 3200 M-mode lines are acquired yielding 100 velocity profiles. At the center of the vessel, the mean and standard deviation of the estimated velocity vectors are (vx, vy, vz) = (-0.026, 95, 1.0)±(8.8, 6.2, 0.84) cm/s compared to the expected (0.0, 96, 0.0) cm/s. Relative to the velocity magnitude this yields standard deviations of (9.1, 6.4, 0.88) %, respectively. Volumetric flow rates were estimated for all ten frames yielding 57.9±2.0 mL/s in comparison with 56.2 mL/s measured by a commercial magnetic flow meter. One frame of the obtained 3D vector flow data is presented and visualized using three alternative approaches. Practically no in-plane motion (vx and vz) is measured, whereas the out-of-plane motion (vy) and the velocity magnitude exhibit the expected 2D circular-symmetric parabolic shape. It shown that the ultrasound method is suitable for real-time data acquisition as opposed to magnetic resonance imaging (MRI). The results demonstrate that the 3D TO method is capable of performing 3D vector flow imaging.
This paper presents a recursive approach for parametric delay calculations for a beamformer. The suggested calculation procedure is capable of calculating the delays for any image line defined by an origin and arbitrary direction. It involves only add and shift operations making it suitable for hardware implementation. One delaycalculation unit (DCU) needs 4 parameters, and all operations can be implemented using fixed-point arithmetics. An N-channel system needs N+ 1 DCUs per line - one for the distance from the transmit origin to the image point and N for the distances from the image point to each of the receivers. Each DCU recursively calculates the square of the distance between a transducer element and a point on the beamformed line. Then it finds the approximate square root. The distance to point i is used as an initial guess for point i + 1. Using fixed-point calculations with 36-bit precision gives an error in the delay calculations on the order of 1/64 samples, at a sampling frequency of fs = 40 MHz. The circuit has been synthesized for a Virtex II Pro device speed grade 6 in two versions - a pipelined and a non-pipelined producing 150 and 30 million delays per second, respectively. The non-pipelined circuit occupies about 0.5 % of the FPGA resources and the pipelined one about 1 %. When the square root is found with a pipelined CORDIC processor, 2 % of the FPGA slices are used to deliver 150 million delays per second.
The image quality in medical ultrasound scanners is determined by several factors, one of which is the ability of the receive beamformer to change the aperture weighting function with depth and beam angle. In digital beamformers, precise dynamic apodization can be achieved by representing that function by numeric sequences. For a 15 cm scan depth and 100 lines per image, a 64-channel, 40 MHz ultrasound beamformer may need almost 50 million coefficients. A more coarse representation of the aperture relieves the memory requirements but does not enable compact and precise beamforming.
Previously, the authors have developed a compact beamformer architecture which utilizes sigma-delta A/D conversion, recursive delay generation and sparse sample reconstruction using FIR filters. The channel weights were here fixed. In this paper, a compact implementation of dynamic receive apodization is presented. It allows precise weighting coefficient generation and utilizes a recursive algorithm which shares its starting parameters with the recursive delay generation logic. Thus, only a separate calculation block, consisting of 5 adders and 5 registers, is necessary. A VHDL implementation in a Xilinx XCV2000E-7 FPGA has been made for the whole receive beamformer for assessing the necessary hardware resources and the achievable performance for that platform. The code implements dynamic apodization with an expanding aperture for either linear or phased array imaging. A complete 32-channel beamformer can operate at 129.82 MHz and occupies 1.28 million gates. Simulated in Matlab, a 64-channel beamformer provides gray scale image with around 55 dB dynamic range. The beamformed data can also be used for flow estimation.
Modern diagnostic ultrasound beamformers require delay information for each sample along the image lines. In order to avoid storing large amounts of focusing data, delay generation techniques have to be used. In connection with developing a compact beamformer architecture, recursive algorithms were investigated. These included an original design and a technique developed by another research group. A piecewise-linear approximation approach was also investigated. Two imaging setups were targeted -- conventional beamforming with a sampling frequency of 40 MHz and subsample precision of 2 bits, and an oversampled beamformer that performs a sparse sample processing by reconstructing the in-phase and quadrature components of the echo signal for 512 focal points. The algorithms were synthesized for a FPGA device XCV2000E-7, for a phased array image with a depth of 15 cm. Their performance was as follows: (1) For the best parametric approach, the gate count was 2095, the maximum operation speed was 131.9 MHz, the power consumption at 40 MHz was 10.6 mW, and it requires 4 12-bit words for each image line and channel. (2) For the piecewise-linear approximation, the corresponding numbers are 1125 gates, 184.9 MHz, 7.8 mW, and 15 16-bit words.