Ultrasound in Med. & Biol., Vol. 39, No. 6, pp.

10151025, 2013 Copyright 2013 World Federation for Ultrasound in Medicine & Biology Printed in the USA. All rights reserved 0301-5629/$ - see front matter

http://dx.doi.org/10.1016/j.ultrasmedbio.2012.12.016

Original Contribution
HIGH-FRAME-RATE ULTRASOUND COLOR-ENCODED SPECKLE IMAGING OF COMPLEX FLOW DYNAMICS BILLY Y. S. YIU and ALFRED C. H. YU
Medical Engineering Program, The University of Hong Kong, Pokfulam, Hong Kong
(Received 5 August 2012; revised 4 November 2012; in nal form 23 December 2012)

AbstractRealization of ow imaging at high frame rates is essential to the visualization of complex ow patterns with fast-changing spatiotemporal dynamics. In this study, we present an experimental demonstration of a novel ultrasound-based high-frame-rate ow visualization technique called color-encoded speckle imaging (CESI), which depicts ow information in a hybrid form comprising ow speckle pattern and color-encoded velocity mapping. This technique works by integrating two key principles: (i) using broad-view data acquisition schemes like plane wave compounding to obtain image data at frame rates well beyond the video display range and (ii) deriving and displaying both ow speckles and velocity estimates from the acquired broad-view image data. CESI was realized on a channel-domain ultrasound imaging research platform, and its performance was evaluated in the context of monitoring complex ow dynamics inside a carotid bifurcation ow phantom with 25% eccentric stenosis at the inlet of the internal carotid artery. Results show that, using an imaging frame rate of 2000 frames per second (based on plane wave compounding with ve steering angles), CESI can effectively render ow acceleration and deceleration with visual continuity. It is also effective in depicting how stenosis-related ow disturbance events, such as ow jet formation and post-stenotic ow recirculation, evolve spatiotemporally over a pulse cycle. We anticipate that CESI can represent a rational approach to rendering ow information in ultrasound-based vascular diagnoses. (E-mail: alfred.yu@hku.hk) 2013 World Federation for Ultrasound in Medicine & Biology. Key Words: Ultrasound ow imaging, Color-encoded velocity mapping, Flow speckle visualization, High frame rate, Complex ow analysis.

INTRODUCTION Ultrasound has long been used clinically for visualization of blood ow dynamics in real-time through the use of the color-ow imaging (CFI) mode (Beach et al. 1997). This modality generally depicts ow information in the form of color-coded pixels that are overlaid on top of a brightness-mode (B-mode) image of the anatomical structure, and the ow pixels can represent either blood velocities or blood signal power as typically derived from Doppler principles (Ferrara and DeAngelis 1997). Despite having emerged as a clinical modality for real-time ow visualization, CFI is still subject to a number of systemic weaknesses inherent in its imaging process. One of the weaknesses is that its visualization performance can be obscured by various artifact sources, such as aliasing and beam-ow angle variations that would lead to false coloring in the image (Evans 2010). These systemic

Address correspondence to: Alfred C. H. Yu, The University of Hong Kong, Pokfulam, Hong Kong. E-mail: alfred.yu@hku.hk 1015

artifacts make it non-trivial at times for inexperienced sonographers to interpret blood ow dynamics using CFI (Tahmasebpour et al. 2005; Torp-Pedersen and Terslev 2008). To improve ultrasounds ow visualization performance, alternate non-CFI techniques have been proposed. For example, it has been suggested that, through the use of microbubble contrast agents, B-mode imaging can be used to visualize blood ow (Liu et al. 2008). This technique, known as echo particle image velocimetry, is capable of displaying ow trajectories at frame rates up to 500 fps in some reported custom hardware (Zhang et al. 2011). Nevertheless, the use of microbubbles can represent an unfavorable factor for the techniques widespread clinical adoption, given that the clinical safety of microbubbles remains a controversial issue at present (Stride and Coussios 2010). Another non-CFI ow visualization technique that has been proposed is B-ow imaging. It is based on the extraction of the ow speckle pattern from inter-frame high-pass ltering of B-mode images through leveraging

1016

Ultrasound in Medicine and Biology

Volume 39, Number 6, 2013

on the decorrelation of blood signals in between frames (Chiao et al. 2000; Lovstakken et al. 2006). This approach has been considered as a more visually intuitive ow imaging mode than CFI (Umemura and Yamada 2001), particularly if velocity coding is included in the image (Lovstakken et al. 2008; Nyrnes et al. 2007). An important sensitivity criterion for B-ow imaging to work properly is that the frame rate needs to be much higher than the temporal rate of blood motion in order for the blood signals to remain coherent between frames while yielding a detectable decorrelation. Specically, frame rates .1000 fps are known to be required to properly visualize complex ow patterns with signicant ow disturbance (Hoskins 2010). However, satisfying such a frame rate requirement can be rather challenging for conventional ultrasound imaging that is typically based on the use of a beamline-based data acquisition paradigm, although beam inter-leaving (Lovstakken et al. 2006) and coded ring schemes (Leavens et al. 2007) can help to boost the frame rate by a few fold. In this article, we present an experimental demonstration of a new high-frame-rate ultrasound ow visualization technique that is intended to render a visually coherent presentation of complex ow dynamics, such as those observed in stenosed arteries. Our new technique, which we refer to as color-encoded speckle imaging (CESI), can achieve high frame rates of greater than 1000 fps through the use of broad-view transmission schemes that are known to be capable of drastically reducing the number of rings needed to acquire each frame (Mo et al. 2003; Sandrin et al. 2001). It is worth noting that broad-view data acquisition schemes have also been applied recently to boost the frame rate of conventional CFI, and their formulations generally involve plane-wave excitation principles (Bercoff et al. 2011; Diagle and Kaczdowski 2009; Hasegawa and

Kanai 2008; Xu et al. 2010) or point-source ring strategies (Jensen and Nikolov 2004; Nikolov and Jensen 2003). Some of these efforts have shown potential in providing vector ow information (Jensen and Oddeshede 2006; Lu et al. 2006; Udesen et al. 2008). In contrast, the CESI technique to be presented in this work can be considered as a more visually integrative form of high-frame-rate CFI, in that simultaneous visualization of ow trajectories and velocity mapping is provided in the forms of moving speckles and color encoding respectively. Note that ow speckle visualization is realized in CESI by leveraging on decorrelation processing principles similar to those used in conventional B-ow imaging. As such, CESI can also be regarded as a high-frame-rate version of color-encoded B-ow imaging that has been implemented previously using beam inter-leaving principles (Lovstakken et al. 2006). IMAGING METHODS Overview of imaging hardware CESI was developed and realized using a channeldomain imaging research platform available in our laboratory. As illustrated in Figure 1, the platform consists of a programmable, array-based transmit-pulser frontend core (SonixTouch; Ultrasonix Medical Corporation, Richmond, BC, Canada), a multi-channel prebeamformed data acquisition tool (Cheung et al. 2012) and a high-speed processing back-end that leverages graphical processing unit (GPU) technology (So et al. 2011; Yiu et al. 2011). During operation, the platform was used to drive a 128-element linear array transducer with 0.3048 mm pitch (L14-5/38; Ultrasonix) according to our custom-dened transmit-ring sequence, and channel-domain pulse-echo data for all 128 array channels were collected in parallel for each transmit ring

Fig. 1. Schematic of system hardware and data acquisition scheme involved in realizing CESI. Key technical parameters are also shown. During operation, the array scanning front-end sends broad-view transmission pulses into the imaging view (steered plane waves used in this work). Prebeam-formed data would be acquired and streamed to the back-end computing platform for beam forming, signal processing and display rendering.

Color-encoded speckle imaging d B. Y. S. YIU and A. C. H. YU

1017

through the prebeam-formed data acquisition tool. The channel-domain data was then streamed to the back-end computing platform through a universal serial bus version 2.0 link for beam forming and ow processing. Broad-view data acquisition and beam forming To facilitate high-frame-rate data acquisition, we have made use of the plane wave compounding scheme (Bercoff et al. 2011) as depicted in Figure 1. This scheme was implemented on our platform by programming the transmit-ring sequence through the TEXO software development kit (Ultrasonix) to send unfocused pulses (three cycles in duration; 7 MHz center frequency) into the imaging view from different steering angles (achieved by applying beam steering delays only over the array channels; no focusing delays were added). In the current work, ve steering angles were used (transmission was performed in rotational order), and they span the angle range of 210 to 110 (in 5 steps). The pulse repetition frequency (PRF) in between steered plane wave excitations was set to 10 kHz. This setting yielded an effective imaging frame rate of 2000 fps, as each compounded image frame was obtained by coherently summing ve raw images (one from each of the ve steering angles). On the receiver side, channel-domain data were collected synchronously for each transmission event until the onboard data buffer (16 GB) was full. Each channel data vector was acquired at a sampling rate of 40 MHz. As noted earlier (Cheung et al. 2012), the storage capacity of our research platform was sufcient for more than 10,000 pulse-echo events. This capacity allowed us to acquire at least a few pulse cycles of raw data in each experiment when using pulse repetition frequencies in the kilohertz range and an imaging depth of 5 cm (i.e., typical parameters for carotid imaging). After streaming the channel-domain data to the back-end computing platform (equipped with 24 Gb of random access memory to accommodate the large raw data volume), a digital band-pass ltering operation was performed individually on each channel data ensemble before the beam-forming process. Performed using Matlab (R2012a; Mathworks, Natick, MA, USA), this operation served to improve the signal-to-noise ratio of the acquired channel-domain data. Our band-pass lter was dened as a high-order nite-impulse-response (FIR) lter (114th order with 60 dB stop-band suppression; based on an equiripple minimum-order lter design method), and its pass-band was set to 59 MHz for an imaging frequency of 7 MHz. Beam forming was then performed at real-time throughputs using a GPU-based delay-and-sum algorithm that was developed previously (Yiu et al. 2011), and compounding was performed by adding (individually at each pixel position) the respective

image values from the ve steered beam-formed frames in the group (executed on the GPU as well). Every compounded data frame (i.e., the set of analytical samples after beam forming and compounding) was 128 3 255 pixels in dimensions (axial by lateral), and each pixel was nominally 0.15 3 0.15 mm in size. These data frames were subsequently passed to the ow processing module for signal analysis. Flow signal processing: speckle extraction and Doppler estimation To extract ow speckles and derive velocity estimates as required for CESI, slow-time signal processing was performed on the set of beam-formed data frames. A schematic overview of the steps involved is shown in Figure 2, and it was performed in real time using another in-house GPU kernel that we developed based on parallel processing strategies similar to those in our previous study (So et al. 2011). For each pixel position, its corresponding slow-time ensemble (i.e., the set of beam-formed data samples over multiple frames) was rst passed through a clutter lter. This step is crucial for ow speckle extraction because it serves to suppress unwanted tissue echoes that remain highly correlated in between frames (as opposed to blood scatterers that would lead to slow-time signal decorrelation as they traverse through the sample volume). In terms of its implementation, FIR high-pass ltering has been deemed as suitable because its linear time-invariant ltering characteristics would ensure that the ltering effects are the same over all slow-time ensembles and thereby help to preserve the spatial continuity of the ow speckle pattern (Lovstakken et al. 2006). Such an operation was realized as a high-order lter in this work using Matlab (169th order with 100 dB stop-band suppression; based on equiripple minimum-order lter design method), and its cutoff frequency was set to 100 Hz (empirically chosen via ad hoc evaluation of clutter suppression performance). Note that a high lter order can be used here because a large number of slow-time data samples (essentially equal to the frame rate on a per-second scale) were made available by broad-view data acquisition. This is not realizable in conventional CFI and color-encoded B-ow imaging whose slow-time ensemble size is usually kept to less than 20 samples to satisfy real-time imaging constraint (low lter orders must therefore be used instead). After slow-time clutter ltering was completed, computation of the velocity ensemble and the ow speckle ensemble for each pixel position was done by performing Doppler estimation on the ltered slow-time ensemble corresponding to that position. Specically, for velocity estimation, each sample in the velocity ensemble was obtained by: (i) dening a data window that was centered

1018

Ultrasound in Medicine and Biology

Volume 39, Number 6, 2013

Fig. 2. Flow signal processing framework used in CESI. For the slow-time ensemble of each pixel position, clutter ltering is rst performed. Next, ow speckle is extracted in the form of mean-squared magnitude, and mean ow velocity is estimated using the lag-one autocorrelation algorithm (an individual estimate is obtained over each sliding window position). Once this process is performed at all positions along the slow-time axis and for all pixels, alpha compositing is applied to form a set of color-encoded speckle images.

about the respective time point in the ltered ensemble (Fig. 2), (ii) computing the mean velocity over the slow-time window using the well-known lag-one autocorrelation algorithm (Evans 2010) and (iii) converting the velocity estimate into a color code based on the hot-cold bicolor hue (with maximum value pegged to the aliasing limit without Doppler angle correction: 11 cm/s for our data acquisition parameters). The data window size was empirically chosen as 64 samples, and the estimation process was performed at each position along the slowtime axis (i.e., essentially a sliding window estimation strategy). For the ow speckle ensemble, a similar sliding window strategy was used for estimation, but instead of computing the mean frequency, the mean power was calculated by nding the mean-squared value over each data window. The power estimates were then mapped to a grayscale color code (over a 20-dB dynamic range) to obtain ow speckles for display. Display rendering algorithm Once obtaining the ow speckle map and the colorencoded velocity map by performing ow signal processing on all slow-time ensembles, they were combined for simultaneous display using alpha compositing principles. In particular, each grayscale ow speckle and its corresponding colored velocity code were merged with equal alpha transparency weighting (i.e., alpha 5 0.5), and they were overlaid on top of a B-mode image of the anatomical structure. As such, CESI can be considered as a triplex display mode. Note that the display gain for grayscale speckles and colored velocity codes were

both pegged to the ow signal power. In addition, persistence and median ltering were performed on an ad hoc basis to enhance the visualization quality. EXPERIMENTAL METHODS Overview of ow phantom setup To provide an experimental means of characterizing the performance of CESI, we have constructed a carotid bifurcation ow phantom setup that can simulate complex ow dynamics under stenotic conditions. As illustrated in Figure 3a, during operation, carotid pulsatile ow with a waveform pulse cycle of 72 beats per minute (i.e., 1.2 Hz) was driven into the phantom using a commercial gear pump system (AccuFlow-Q; Shelley Medical Imaging, London, ON, Canada). The pumps systolic ow rate was set to 5 mL/s, and the mean ow rate was measured at 1.5 mL/s (by timing the duration for a xed amount of uid to pass through the phantom outlet). The blood mimicking uid that we used was a nylon particle scatterer solution purchased from the ow pump manufacturer (BMF-US; Shelley Medical Imaging). The uid is known to possess acoustic properties similar to that of blood, and its kinematic viscosity and density were 3.95 3 1026 m2/s and 1037 kg/m3, respectively (Ramnarine et al. 1998). The carotid bifurcation phantom is based on a generalized tuning-fork co-planar vessel geometry that is widely used in computational uid dynamics investigations (Steinman et al. 2000). It is a wall-less model with tube diameters of 6, 3 and 4 mm, respectively, for the

Color-encoded speckle imaging d B. Y. S. YIU and A. C. H. YU

1019

Fig. 3. Overview of bifurcation ow phantom system used for experimental demonstration of CESI. (a) The major components and the operational parameters. (b) A 3-D model of the 25% stenosis carotid bifurcation geometry used for ow phantom construction. B-mode images for (c) long-axis views of the CCA branch and (d) the bifurcation.

common carotid artery (CCA), external carotid artery (ECA) and internal carotid artery (ICA). To facilitate an evaluation of the ability of CESI to visualize complex ow dynamics, a 25% eccentric stenosis was created at the inlet to the ICA branch of the phantom, and this was dened based on the criteria set forth by the North American Symptomatic Carotid Endarterectomy Trial (Smith et al. 1996). The phantoms tissue-mimicking material was a polyvinyl alcohol (PVA) cryogel whose acoustic and mechanical properties were tuned to be within the same range as that of human tissues. Details of its fabrication process will be described in the next subsection. With our phantoms vessel geometry and ow rate, it can be shown that the mean velocity in the CCA branch was equal to 5.3 cm/s based on established formulas (Peacock et al. 1998). In addition, the peak CCA systolic velocity (arising at tube center; at most twice the average systolic velocity assuming parabolic prole) was approximately 35.4 cm/s, thereby yielding a maximum oscillatory velocity component of 30.1 cm/s (i.e., difference between peak systolic velocity and mean velocity) in the pulsatile ow prole. Correspondingly, the mean and peak Reynolds numbers in the CCA were estimated to be equal to 80 and 537, respectively, for the given viscosity of our blood mimicking uid. These values

are smaller than the critical Reynolds number of 1817 for turbulence to emergea threshold that can be readily derived from an empirical relation (equation 6 in Peacock et al. 1998) incorporating the Womersley parameter and the Strouhal number (4.14 and 0.012, respectively, for our specied pulse frequency). As such, our phantom setup can be considered to exhibit laminar ow conditions in general. Stenosed bifurcation phantom: fabrication procedure The carotid bifurcation phantom was fabricated inhouse via an investment casting approach that is well regarded as a suitable framework for creating complex ow phantoms (Poepping et al. 2002; Watts et al. 2007). Using a computer-aided design software (SolidWorks; Dassault Syst emes, Waltham, MA, USA), we rst generated a model of the stenosed carotid bifurcation (Fig. 3b) based on our nominal diameters of CCA, ECA and ICA (6, 3 and 4 mm) and the stenosis geometry (25% blockage stemming from the outer wall of the ICA inlet). A rigid master pattern of this model was then produced on a stereolithography machine (Eden 350V; Objet Geometries, Rehovot, Israel) using a photopolymer that becomes rigid upon curing (FullCure 850; Objet Geometries). With the master pattern, a negative mold of the stenosed bifurcation was subsequently formed

1020

Ultrasound in Medicine and Biology

Volume 39, Number 6, 2013

using silicone elastomer (Rhodorsil RTV 573; Rhodia Silicones, Braeside, Victoria, Australia), and the steps involved are the same as those described elsewhere (Poepping et al. 2002). Next, a bifurcation replica was obtained by pouring melted injection wax (ACCU Flakes 14079; Kerr Corporation, Orange, CA, USA) into the open channels of the negative mold and letting the wax cure at room temperature. Before pouring in the wax, a string of 150-mmdiameter nylon wire (with knots tied 3 cm apart) was threaded into the mold inlet to serve as a guidewire for removing the replicate core later. To construct a ow phantom from the bifurcation replica, we rst took the nylon-threaded wax core out of the negative mold and mounted it onto an acrylic box container measuring 245 3 30 3 50 mm in dimensions (long axis of bifurcation was along the box length, and all the vessel branches were aligned to the vertical plane). In accommodating the three branches of the bifurcation wax core at the two ends of the box, quickdisconnect tting connectors (EW-06361-61; ColeParmer, Vernon Hills, IL, USA) were placed at the inlet and outlet sides of the box. Subsequently, the tissuemimicking material was cast; it was composed of a PVA mixture containing 88.7% distilled water, 10% PVA (341584; Sigma-Aldrich, St. Louis, MO, USA), 1% silicon dioxide (S5631; Sigma-Aldrich) and 0.3% potassium sorbate (85520; Sigma-Aldrich). To perform this step, the uidic PVA mixture was rst poured at room temperature into the container with the mounted bifurcation wax core, and the entire box underwent three freeze-thaw cycles (freeze at 220  C for 8 h followed by thawing at 4  C for 16 h) to set the PVA cryogel. Note that three freeze-thaw cycles were chosen for this work because the resulting acoustic and mechanical properties of the PVA cryogel, as determined by others earlier (Fromageau et al. 2007), would be similar to those of human tissues (Hoskins 2007). We have conrmed the acoustic properties of our PVA cryogel using a wellestablished insertion-loss measurement technique as described by others elsewhere (Zell et al. 2007). The acoustic speed and attenuation coefcient were 1518 6 12 m/s and 0.24 6 0.01 dB/(cm,MHz), respectively, for nine trials with test slabs whose thickness ranged between 2.58.0 cm (data not shown). To complete the phantom fabrication, the bifurcation wax core was removed by pulling on the nylon guidewire from all three ends to fracture the wax, with the knots on the wire serving as snapping points. The wax pieces were then slowly withdrawn from the phantom, and the resulting wall-less ow channels were rinsed with distilled water. The bifurcation phantom was subsequently connected to the ow pump for experimentation.

Imaging experiment protocol Using our channel-domain imaging research platform, we have generated CESI volumes of our phantom when it is in operation. The steps involved are described in the Imaging Methods section, and the relevant imaging parameters are given in Figure 1. For reference, B-mode images were also captured from the SonixTouch scanners clinical mode in the same imaging view (see Fig. 3c, d for long-axis views of the CCA branch and at the bifurcation). In addition, to provide comparative insights for the CESI results, Doppler spectrograms were computed over selected sample volumes in the imaging view using the same raw slow-time data sets. Note that, because broad-view data acquisition schemes were used, Doppler spectrograms can be computed at multiple arbitrary-sized sample volumes simultaneously because slow-time data for all pixel positions of the imaging view were acquired in parallel, as demonstrated by others recently (Diagle and Kaczdowski 2009; Xu et al. 2010). For the Doppler spectrograms that we computed, 128-sample sliding windows were used for the short-time Fourier transform operation, and the data overlap in between spectral estimates was 93.75%. In addition, the nominal beam-ow angle used for velocity conversion (via the Doppler equation) was dened as the angle between the transducer and the vessel lumen adjacent to the sample volume position (determined from the reference B-mode images). RESULTS CESI allowed tracking of pulsatile ow at high temporal resolution Simultaneous visualization of ow speckles and ow velocities at high frame rates, as provided by CESI, was generally found to be effective in tracking pulsatile ow patterns. As a demonstration of this nding, Supplemental Movie 1 shows a cineloop for a long-axis view of the CCA in our bifurcation phantom taken at 2000 fps (playback frame rate: 60 fps). Note that the ow prole in this case is monophasic, as typical of carotid ow. As can be observed in the cineloop, the moving ow speckles consistently depicted ow acceleration and deceleration during the systolic and diastolic phases of the pulse cycle respectively. This specklebased visual rendering of ow dynamics is augmented by the color encoding that showed brightening and darkening (Fig. 4a, b) in the vessel region depending on the pulse-cycle phase. Visual continuity of ow speckles: rendered by high inter-frame correlation The smooth rendering of ow dynamics shown in the cineloop of Supplemental Movie 1 seems to be

Color-encoded speckle imaging d B. Y. S. YIU and A. C. H. YU

1021

Fig. 4. Representative CESI frames for the CCA branch of the carotid bifurcation phantom. (a) The image appearance during systole and (b) diastole of the pulse cycle. (c) The inter-frame correlation coefcient over a few pulse cycles, averaged over ve sample volumes denoted in (a). Error bars denote standard error of measurement, shown once every 100 frames. (d) Doppler spectrogram for an 8 3 1-mm sample volume that encompasses the entire vessel lumen. See also Supplementary Movie 1.

facilitated largely by the high inter-frame correlation of ow speckles. As shown in Figure 4c, the inter-frame correlation coefcient for pixels within the CCA vessel was close to 0.9 for most of the pulse cycle (each estimate was the average over ve independent estimates for the 1 3 1 mm windows depicted in Fig. 4a). During ow acceleration, there is a corresponding drop in correlation coefcient because of the increased scatterer speed at these time points, but it is still maintained at a relatively high level of 0.7, which helps to maintain the visual continuity of the ow speckle pattern. High-frame-rate visualization of ow jets and ow recirculation When applied to examine ow dynamics in the carotid bifurcation, CESI was found to be effective in providing smooth visualization of how the ow pattern varies spatially within the vessel lumen and evolves temporally over the pulse cycle. An example corresponding to such a nding is given in Supplemental Movie 2 for a long-axis co-planar view of our carotid bifurcation phantom with 25% eccentric stenosis at the ICA inlet. From this cineloop, visualization of two ow disturbance events that are characteristic of carotid stenosis can be noted as depicted in Figure 5a. The rst event is the ow jet stemming from the stenosis site at the ICA inlet (i.e., at the entrance to upper branch). Its progression from initial formation (during systole) to dissipation downstream from the stenosis (during diastole) can be continuously traced throughout the CESI cineloop. The second event worth noting is the post-stenotic ow recir-

culation in the ICA carotid bulb. As can be observed, CESI has provided a high-frame-rate rendering of how this recirculation evolves spatiotemporally over the pulse cycle. The presence of such ow reversal, highlighted in the CESI cineloop by the apparent change in color encoding (red hue for the forward ow jet vs. blue hue for the ow reversal region), is known to be an indication for the formation of a ow separation zone in this part of the vasculature. Flow speckles and color encoding: complementary role in ow visualization From a visual interpretation standpoint, the ow speckles and the color encoding provided by CESI seem to complement each other in rendering the ow conditions in our bifurcation phantom investigation. For example, for the ow jet observed in Supplemental Movie 2, the spatiotemporal progression of ow speckles has helped to depict that ow is traversing toward the distal end of the ICA branch; on the other hand, the color encoding has served to highlight the ow jets spatial location, even while velocity aliasing was observed at certain time points (Fig. 5b, c). Similarly, in visualizing ow in the ICA carotid bulb, the ow speckle trajectories have shown the presence of recirculating ow within this region, whereas color encoding has provided visual emphasis through rendering a distinctly different color hue with respect to the nearby ow jet (red vs. blue). It is worth noting that the complementary relationship between ow speckles and color encoding becomes critical in interpreting the ow pattern in the ECA (i.e.,

1022

Ultrasound in Medicine and Biology

Volume 39, Number 6, 2013

Fig. 5. Summary of key features depicted by CESI for the bifurcation segment of the carotid ow phantom. Note that there is a 25% stenosis at the inlet of the ICA (upper branch). The features shown here correspond to observations deduced from Supplemental Movie 2, and they include (a) instances of disturbed ow (ow jet and recirculation) and (b) instances of color encoding artifacts (aliasing and angle dependence). As reference, Doppler spectrograms (after phase unwrapping and beam-ow angle correction) are also shown for 1 3 1-mm sample volumes placed in (c) ICA ow jet, (d) ECA distal end and (e) ECA proximal section. Locations for these three sample volumes are labeled in (b).

lower branch in Supplemental Movie 2). As indicated by the ow speckle trajectories, ow is unidirectional in the ECA throughout the pulse cycle, and it can be conrmed by the Doppler spectrograms in the proximal and distal segments of ECA (Fig. 5d, e; similar ow patterns can be observed). Such a visual rendering has served well to rectify the false color-encoding artifact in this branch (Fig. 5b), whereby both blue and red hues can be observed, respectively, in the proximal and distal segments because of beam-ow angle variations (ow in the proximal side was moving away from the transducer, but in the distal end it moved toward the transducer). Still-frame analysis reveals post-stenotic complex ow patterns Given the high imaging frame rate realized in CESI, still-frame snapshots of the ow pattern at specic parts of the pulse cycle can be obtained to gain technical insights on the ow patterns spatiotemporal dynamics. As an illustration, Figure 6 shows a frame-by-frame analysis of the same eld of view observed in the cineloop of Supplemental Movie 2. One particular nding to be noted is that, in the ICA carotid bulb, the ow direction changed

four times in a single pulse cycle (Fig. 6ad). This nding indicates that ow in this part of the vasculature is quadphasic in nature (i.e., the ow reversal region would form and dissipate twice in every pulse cycle), which can be conrmed by the oscillatory appearance of the Doppler spectrogram for the sample volume placed within the ICA carotid bulb (Fig. 6e). A related observation worth noting is that, as illustrated in Figure 7a, the spatial realm of the ow reversal region was found to change signicantly over the pulse cycle. Figure 7b shows an estimate of the area of the ow reversal region over the pulse cycle (determined by counting the number of blue-hued pixels in the ICA carotid bulb via an edge detection routine and multiplying the result with the nominal pixel area). It can be observed that the largest region area corresponds to the time point with the most signicant reverse ow (determined by inspection of Fig. 6e).

DISCUSSION Summary of contributions It has been the intent of this work to devise a novel way of visualizing ow at high frame rates to facilitate a detailed interpretation of complex ow patterns with

Color-encoded speckle imaging d B. Y. S. YIU and A. C. H. YU

1023

Fig. 6. CESI still-frame analysis of complex ow patterns in the stenosed carotid bifurcation phantom. Representative CESI frames are shown for four time points with these features in the ICA carotid bulb: (a) maximum forward ow; (b) maximum reversed ow; (c) secondary forward ow; (d) secondary reversed ow. These time points are all within the same pulse cycle, and they are depicted in the Doppler spectrogram shown in (e) for a 1 3 1-mm sample volume centered in the ICA carotid bulb, as labeled in (a). See also Supplemental Movie 2.

signicant spatiotemporal variations over a pulse cycle. We have demonstrated that a visually intuitive form of high-frame-rate ow imaging, termed CESI, can be achieved by integrating two key principles: (i) using broad-view data acquisition schemes (plane wave compounding in our case) to obtain image frames (Fig. 1), and (ii) deriving and displaying both ow speckles and velocity estimates from the acquired data (Fig. 2). This novel ow visualization technique developed by our team can be considered as a triplex display mode in which ow trajectories (moving speckles), velocity mapping (color encoding) and anatomic structure (B-mode image) are shown simultaneously.

With frame rates well beyond the video display range (2000 fps in our experiments), CESI was found to be capable of visualizing various fast-changing ow dynamics (both spatially and temporally) in a stenosed carotid bifurcation phantom (Fig. 3) operating with 1.2Hz pulse waveforms that are typical of the human cardiac cycle. This new technique was not only able to track ow acceleration and deceleration (Fig. 4, Supplemental Movie 1); it was also effective in depicting ow jet formation and progression, as well as the development of recirculation zones downstream from the stenosis site (Figs. 6 and 7, Supplemental Movie 2). In facilitating this visualization, the speckle trajectories and velocity mapping

Fig. 7. Spatiotemporal changes in the ow reversal region of the stenosed carotid bifurcation phantom. Determined based on analysis of CESI frames, (a) shows a schematic illustration of the size of ow reversal region at time points corresponding to (1) maximum reversed ow and (2) secondary reversed ow. (b) The area of the ow reversal region over multiple pulse cycles, computed by counting the number of pixels exhibiting reversed ow in the ICA carotid bulb of each CESI frame and multiplying that with the nominal pixel area.

1024

Ultrasound in Medicine and Biology

Volume 39, Number 6, 2013

were found to have roles complementary to each other (Fig. 5). Specically, speckle trajectories can help to identify ow recirculation and ow acceleration as well as to rectify any counter-intuitive artifacts in the ow velocity map (e.g., aliasing in the ow jet). On the other hand, velocity mapping serves well in highlighting regions of fast ow (e.g., ow jet formation) and identifying instances of ow separation (as observed in the region between the ow jet and post-stenotic ow recirculation). Note that these detected ow features are in line with those observed in computational uid dynamics simulation studies (Steinman et al. 2000) and particle image velocimetry validation experiments (Poepping et al. 2010). As such, our CESI results can be generally considered as valid. Perspectives for further development Our current framework can potentially be extended into more advanced forms of complex ow analysis. One particular direction of interest is to take advantage of the ow information that we derived at high frame rates to obtain quantitative indicators of complex ow dynamics, such as ow vorticity and shear rate. This would technically require the computation of velocity vectors, and the feasibility of doing so has been initially demonstrated by making use of ow speckle tracking principles (Udesen et al. 2008). Further effort along this line of research would involve investigating new ways to improve the accuracy and consistency of velocity vector estimation algorithms, and such investigation is crucial given that vorticity and shear rate are essentially spatial derivatives of the velocity vector (Hoskins 2010). Another research direction in relation to our current work is on the need to develop more advanced algorithms to better delineate ow regions for CESI display. This direction is essential to enhance the ow detection performance in vessel regions near the wall where clutter distortions from nearby tissue regions are inherently stronger because the use of broad-view data acquisition schemes would inherently lead to stronger off-axis side lobes (Udesen et al. 2008). The issue is likely to be more signicant in the presence of tissue motion, because the FIR lter used in this work is inherently not adaptive to the clutter characteristics in such a scenario. We anticipate that, to establish the efcacy of CESI in cases with significant tissue motion, more rigorous ltering solutions with adaptive characteristics (Bjaerum et al. 2002) would likely be needed. CONCLUSIONS As a new way of perceiving ow dynamics at high frame rates, CESI is after all a non-invasive ow visualization framework. It may thus represent a more rational

approach to providing ow information in vascular diagnoses in which ultrasound CFI has traditionally been applied in clinical practice but has not fullled the diagnostic needs. For example, CESI can be used as an in vivo imaging tool to visualize ow disturbance events, such as the formation of ow jets and ow recirculation, as demonstrated in our carotid bifurcation imaging experiment. It would also be of interest to apply this new technique to visualize complex ow dynamics non-invasively inside abnormal vascular compartments, such as ulcers and aneurysms. In addition, as an investigative tool, CESI can be useful in rendering turbulent ow with high Reynolds numbers. Fine temporal resolution in the sub-millisecond range is especially needed in this application because turbulence is marked by transient deviations (i.e., rapid spatiotemporal changes) from a laminar ow prole. Realization of these potential applications might help to expand the overall value of ultrasound in vascular diagnostics, because our new imaging method offers frame rates greater than those achievable with B-ow imaging and provides a more lucid visualization of ow patterns than CFI.
AcknowledgmentsThe authors are grateful to Simon S. M. Lai and Alexander K. K. Poon for their assistance with the phantom fabrication and acoustic property measurements. This work was funded in part by the Research Grants Council of Hong Kong (GRF 785811M) and the Hong Kong Innovation and Technology Fund (ITS/083/11).

SUPPLEMENTARY DATA
Supplementary data related to this article can be found online at http://dx.doi.org/10.1016/j.ultrasmedbio.2012.12.016.

REFERENCES
Beach KW, Keith AC, Primozich JF, Yuan C, Powell K, Phillips DJ, Dunmire B, Plett M, Brown K, Paun M, Strandness DE Jr. Ultrasonic color ow mapping: the visualization of four-dimensional cardiac and vascular ow phenomena using two dimensions and realtime. Ultrasound Med Biol 1997;23:347363. Bercoff J, Montaldo G, Loupas T, Savery D, Meziere F, Fink M, Tanter M. Ultrafast compound Doppler imaging: Providing full blood ow characterization. IEEE Trans Ultrason Ferroelec. Freq Contr 2011;58:134147. Bjaerum S, Torp H, Kristoffersen K. Clutter lters adapted to tissue motion in ultrasound color ow imaging. IEEE Trans Ultrason Ferroelec Freq Contr 2002;49:693704. Cheung CCP, Yu ACH, Salimi N, Yiu BYS, Tsang IKH, Kerby B, Azar RZ, Dickie K. Multi-channel pre-beamform data acquisition system for research on advanced ultrasound imaging methods. IEEE Trans Ultrason Ferroelec Freq Contr 2012;59:243253. Chiao RY, Mo LY, Hall AL, Miller SC, Thomenius KE. B-mode blood ow (B-ow) imaging. Proc IEEE Ultrason Symp 2000;14691472. Diagle RE, Kaczdowski PJ. High frame rate quantitative Doppler ow imaging using unfocused transmit beams. US Patent Application 2009/0326379 A1. Evans DE. Color ow and motion imaging. Proc Inst Mech Eng H 2010; 224:241253. Ferrara K, DeAngelis G. Color ow mapping. Ultrasound Med Biol 1997;23:321345. Fromageau J, Gennisson JL, Schmitt C, Maurice RL, Mongrain R, Cloutier G. Estimation of polyvinyl alcohol cryogel mechanical properties with four ultrasound elastography methods and

Color-encoded speckle imaging d B. Y. S. YIU and A. C. H. YU comparison with gold standard testings. IEEE Trans Ultrason Ferroelec Freq Contr 2007;54:498509. Hasegawa H, Kanai H. Simultaneous imaging of artery-wall strain and blood ow by high frame rate acquisition of RF signals. IEEE Trans Ultrason Ferroelec Freq Contr 2008;55:26262639. Hoskins PR. Physical properties of tissues relevant to arterial ultrasound imaging and blood velocity measurement. Ultrasound Med Biol 2007;33:15271539. Hoskins PR. Haemodynamics and blood ow measured using ultrasound imaging. Proc Inst Mech Eng H 2010;224:255271. Jensen JA, Nikolov SI. Directional synthetic aperture ow imaging. IEEE Trans Ultrason Ferroelec Freq Contr 2004;51:11071118. Jensen JA, Oddershede N. Estimation of velocity vectors in synthetic aperture ultrasound imaging. IEEE Trans Med Imag 2006;25: 16371644. Leavens C, Burns PN, Sherar MD. Fast B-ow imaging: A method for improving frame rate in golay coded B-ow imaging. IEEE Trans Ultrason Ferroelec Freq Contr 2007;54:22722282. Lovstakken L, Bjaerum S, Martens D, Torp H. Blood ow imaging a new real-time, 2-D ow imaging technique. IEEE Trans Ultrason Ferroelec Freq Contr 2006;53:289299. Lovstakken L, Ibrahim KS, Vitale N, Henriksen ST, Kirkeby-Garstad I, Torp H, Haaverstad R. Blood ow imaging: A new two-dimensional ultrasound modality for enhanced intraoperative visualization of blood ow patterns in coronary anastomoses. J Am Soc Echocardiogr 2008;21:969975. Liu L, Zheng H, Williams L, Zhang F, Wang R, Hertzberg J, Shandas R. Development of a custom-designed echo particle image velocimetry system for multi-component hemodynamic measurements: System characterization nad initial experimental results. Phys Med Biol 2008;53:13971412. Lu JY, Wang Z, Kwon SJ. Blood ow velocity vector imaging with high frame rate imaging methods. Proc IEEE Ultrason Symp 2006;963967. Mo LYL, Ji TL, Chou CH, Napolitano D, McLaughlin GW, DeBusschere D. Zone-based color ow imaging. Proc IEEE Ultrason Symp 2003;2932. Nikolov SI, Jensen JA. In-vivo synthetic aperture ow imaging in medical ultrasound. IEEE Trans Ultrason Ferroelectr Freq Contr 2003;50:848856. Nyrnes SA, Lovstakken L, Torp H, Haugen BO. Blood ow imaging a new angle-independent ultrasound modality for the visualization of ow in atrial septal defects in children. Echocardiography 2007;24:975981. Peacock J, Jones T, Tock C, Lutz R. The onset of turbulence in physiological pulsatile ow in a straight tube. Exp Fluids 1998;24:19. Poepping TL, Nikolov HN, Rankin RN, Lee M, Holdsworth DW. An in vitro system for Doppler ultrasound ow studies in the stenosed carotid artery bifurcation. Ultrasound Med Biol 2002;28:495506. Poepping TL, Rankin RN, Holdsworth DW. Flow patterns in carotid bifurcation models using pulsed Doppler ultrasound: Effect of

1025

concentric vs. eccentric stenosis on turbulence and recirculation. Ultrasound Med Biol 2010;36:11251134. Ramnarine KV, Nassiri DK, Hoskins PR, Lubbers J. Validation of a new blood-mimicking uid for use in Doppler ow test objects. Ultrasound Med Biol 1998;24:451459. Sandrin L, Manneville S, Fink M. Ultrafast two-dimensional ultrasonic speckle velocimetry: A tool in ow imaging. Appl Phys 2001;78: 11551158. Smith RF, Rutt BK, Fox AJ, Rankin RN, Holdsworth DW. Geometric characterization of stenosed human carotid arteries. Acad Radiol 1996;3:898911. So HKH, Chen J, Yiu BYS, Yu ACH. Medical ultrasound imaging: to GPU or not to GPU? IEEE Micro 2011;31:5465. Steinman DA, Poepping TL, Tambasco M, Rankin RN, Holdsworth DW. Flow patterns at the stenosed carotid bifurcation: Effect of concentric versus eccentric stenosis. Ann Biomed Eng 2000;28:415423. Stride EP, Coussios CC. Cavitation and contrast: The use of bubbles in ultrasound imaging and therapy. Proc Inst Mech Eng H 2010;224: 171191. Tahmasebpour HR, Buckley R, Cooperberg PL, Fix CH. Sonographic examination of the carotid arteries. Radiographics 2005; 25:15611575. Torp-Pedersen ST, Terslev L. Settings and artefacts relevant in colour/ power Doppler ultrasound in rheumatology. Ann Rheum Dis 2008; 67:143149. Udesen J, Gran F, Hensen KL, Jensen JA, Thomsen C, Nielsen MB. High frame-rate blood vector velocity imaging using plane waves: Simulations and preliminary experiments. IEEE Trans Ultrason Ferroelec Freq Contr 2008;55:17291743. Umemura A, Yamada K. B-mode ow imaging of the carotid artery. Stroke 2001;32:20552057. Watts DM, Sutcliffe CJ, Morgan RH, Meagher S, Wardlaw J, Connell M, Bastin ME, Marshall I, Ramnarine KV, Hoskins PR, Black RA. Anatomical ow phantoms of the nonplanar carotid bifurcation, part I: Computer-aided design and fabrication. Ultrasound Med Biol 2007;33:296302. Xu C, Choi JH, Comess K, Kim Y. Color Doppler and spectral Doppler with high-frame-rate imaging. Proc IEEE Ultrason Symp 2010; 15621566. Yiu BYS, Tsang IKH, Yu ACH. GPU-based beamformer: fast realization of synthetic aperture imaging and plane wave compounding. IEEE Trans Ultrason Ferroelec Freq Contr 2011;58:16981705. Zhang F, Lanning C, Mazzaro L, Barker AJ, Gates PE, Strain WD, Fulford J, Gosling OE, Shore AC, Bellenger NG, Rech B, Chen J, Chen J, Shandas R. In vitro and preliminary in vivo validation of echo particle image velocimetry in carotid vascular imaging. Ultrasound Med Biol 2011;37:450464. Zell K, Sperl JI, Vogel MW, Niessner R, Haisch C. Acoustical properties of selected tissue phantom materials for ultrasound imaging. Phys Med Biol 2007;52:N475N484.