3-D Ultrafast Doppler Imaging Applied to the Noninvasive and Quantitative Imaging of Blood Vessels in Vivo

Abstract

Ultrafast Doppler Imaging was introduced as a technique to quantify blood flow in an entire 2-D field of view, expanding the field of application of ultrasound imaging to the highly sensitive anatomical and functional mapping of blood vessels. We have recently developed 3-D Ultrafast Ultrasound Imaging, a technique that can produce thousands of ultrasound volumes per second, based on three-dimensional plane and diverging wave emissions, and demonstrated its clinical feasibility in human subjects in vivo. In this study, we show that non-invasive 3-D Ultrafast Power Doppler, Pulsed Doppler, and Color Doppler Imaging can be used to perform quantitative imaging of blood vessels in humans when using coherent compounding of three-dimensional tilted plane waves. A customized, programmable, 1024-channel ultrasound system was designed to perform 3-D Ultrafast Imaging. Using a 32X32, 3-MHz matrix phased array (Vermon, France), volumes were beamformed by coherently compounding successive tilted plane wave emissions. Doppler processing was then applied in a voxel-wise fashion. 3-D Ultrafast Power Doppler Imaging was first validated by imaging Tygon tubes of varying diameter and its in vivo feasibility was demonstrated by imaging small vessels in the human thyroid. Simultaneous 3-D Color and Pulsed Doppler Imaging using compounded emissions were also applied in the carotid artery and the jugular vein in one healthy volunteer.

Introduction

Ultrafast Ultrasound Imaging [1] and more specifically Ultrafast Doppler Imaging have recently allowed for novel applications of ultrasound imaging such as the quantification of blood flow in entire 2-D [2]–[12] and 3-D fields [13], [14] of view. For instance, these techniques were applied to the imaging of blood vessels and of the heart [2], [13], [15] and its microvasculature [16], to the imaging the propagation of brain activation during epilepsy in rats [17], and to the mapping of vascular resistivity in human neonates [18].

Given the complex architecture of vascular networks, 3D imaging is of paramount importance for these applications. Available commercial systems do indeed allow for 3D Power and Color Doppler Imaging but they do not benefit from the important advantages of Ultrafast Doppler Imaging. Indeed, Ultrafast Doppler Imaging allows for the mapping of blood vessels with a sensitivity 25 times larger than with conventional methods [19], while simultaneously allowing for the quantification of the Doppler spectrum at any pixel in the image for the mapping of velocity, flow, and resistivity[18]. Moreover, while strategies based on motorized 1-D arrays have been employed to concatenate multiple 2D Ultrafast Doppler images to track angiogenesis in tumors [20], they do not typically allow for the analysis of the dynamics of blood flow, and, perhaps more importantly, suffers from the poor elevational resolution associated with the use of a lens unless lengthy tomographic approaches are used [14].

Ultrafast Imaging can achieve the unprecedented frame rates necessary for full-field-of-view Doppler processing by emitting unfocused waves such as plane [21]–[23] or circular waves [24]–[27] at the cost of a reduced image quality. It is by coherently compounding such waves with varying propagation directions such as tilted plane waves [22] or circular waves emanating from an array of virtual sources located behind the probe [24], [28]–[31] that high image quality, and thus highly sensitive Ultrafast Doppler Imaging can be achieved. In fact, while standard focusing techniques converge linearly (one additional emission provides one additional line) toward the optimal image as a function of the number of ultrasound transmits, coherently-compounded ultrafast imaging has recently been shown to converge rapidly toward the optimal image [13], [22], [24], i.e., in which optimal focusing is applied in both transmit and receive for each pixel. As much fewer transmit events are needed to achieve a specific image quality, much higher frame rates are achieved. For example, when using standard focusing techniques, 128 transmit events would be needed to form an image containing 128 lines. In comparison, coherent compounding approaches can approximate the optimal image with an error that vanishes for most practical purposes when the number of transmit events exceeds a few tens of transmit events in two dimensions [22], [24].

This fast convergence becomes of even greater importance when scaling image formation to 3 dimensions using a 2-D array. Indeed, we have recently shown that 3D Ultrafast Ultrasound Imaging could be performed using a 2-D phased-array probe and that at 3 MHz, a few tens of transmit events were sufficient were sufficient to obtain B-modes with high enough quality to visualize the structures of the heart in 3D [13]. Moreover, full field-of-view 3D Color Doppler of the entire human heart and carotid in vivo were performed at thousands of volumes per second by using a single transmit event per volume. However, coherent compounding for 3D Doppler applications has not been demonstrated to this day.

In this study, we demonstrate that quantitative vasculature maps can be obtained by performing Ultrafast Doppler Imaging with coherent compounding of plane waves in three dimensions. More specifically, we quantify, for the first time, the 3D Power Doppler volume resolution in phantoms and demonstrate its feasibility in the human thyroid in vivo, and show that voxel-wise, coherently-compounded Pulsed Doppler Imaging and Color-Doppler Imaging can be performed in the carotid artery and the jugular vein in humans in a single acquisition.

Methods

System Infrastructure

A customized, programmable, 1024-channel ultrasound system was designed to drive a 32-by-32 matrix array centered at 3 MHz with a 50-% bandwidth at -3 dB and a 0.3-mm pitch (0.3-mm element size, PZT piezocomposite technology, Vermon, Tours, France) as described previously [13]. The system was composed of four 256-transmit/128-receive-channels units initially designed for 2D imaging on an Aixplorer systems (Supersonic Imagine, Aix-en-Provence, France), assembled into a new device and synchronized. The total channel count was thus 1024 in transmit and 512 in receive. Since the receive channels were multiplexed to 1 of 2 transducer elements, each emission was repeated twice to synthetize a total of 1024 receive channels. Specifically, the data from each pair of acquisitions were concatenated to form one synthetic RF dataset corresponding approximately to an acquisition in which all transducers had been receiving simultaneously.

The sequences that were used consisted in emitting tilted plane waves defined by a pair of angles. A larger number of angle pairs results in a rapid improvement of the image quality at the cost of a lower volume rate. 3-MHz, 2-cycles, tri-state (-1,0,1) waveforms were emitted per plane wave to maintain a large bandwidth.

Beamforming and Doppler processing

The beamforming and Doppler processing steps are described in Figure 1. A real-time software bi-plane beamformer was used for positioning and could provide images at more than 30 bi-plane frames per second, with the processing frame rate depending on the depth and resolution required. All the results shown herein were reconstructed at an isotropic resolution of 300 um, which corresponds to approximately one half-wavelength at 3 MHz. Once the positioning of the probe at the desired location was performed using the real-time bi-plane B-mode image, a 3D ultrafast sequence was launched, which consisted in the sequential emission of tilted plane waves. Each tilted plane wave was emitted twice to allow for synthetic receive aperture, i.e., the radio-frequency signals were recorded with a different set of 512 elements at each of these repeated emissions. All the acquisitions performed in this study used symmetrically-distributed tilted plane waves and axially-oriented receive sub-apertures, although it would have been technically possible to use a non-zero Doppler angle both in transmit and receive. Volume beamforming was performed using a standard delay-and-sum approach. Specifically, for each voxel, the RF signals delayed according to the tilt angles and - the position of an element were summed as previously described. Hence, each voxel of a given image was the result of the summation over all plane waves and all 1024 piezoelectric elements and is a calculation that was repeated for each volume.

Figure 1.

(a) 3-D volumes were formed by coherently compounding multiple tilted plane waves with varying propagation angles. (b) For each emitted plane wave, pre-beamformed radiofrequency data was recorded. Beamforming was then performed and consisted in summing, for each voxel, the delayed contribution of each tilted plane wave and each piezoelectric element. (c) Hundreds of imaging volumes were formed in that manner. Clutter in the slow-time dimension was filtered, an the voxel-wise energy was obtained by integrating in the slow-time dimension.

Full-view, 3-D Ultrafast Doppler Imaging volumes were obtained by first removing clutter from the beamformed baseband data using multidimensional spatiotemporal filtering, as described in [32], applied in four dimensions (i.e., 3D space + time). Briefly, this technique consists in eliminating clutter signal by removing the low-frequency principal components of the temporal realizations of each voxel. While this approach provided qualitatively better images, a standard high-pass filter could have also been used. Power Doppler volumes were then obtained by integrating the energy of the clutter-free baseband signal over time (we define the temporal dimension as the temporal succession of imaging volumes). Voxel-wise Pulsed Doppler was obtained by calculating the short-time Fourier transform of the clutter-free baseband signals in the temporal direction using overlapping windows of 50 ms and Color Doppler maps were obtained by calculating the first moment of the spectrum for each window [13].

All beamforming and Doppler processing were performed on graphical processing units and were coded using CUDA C language within a Matlab (2014b, The Mathworks, Nattick, MA) interface. A typical acquisition including all data transfers and calculations such as volume beamforming and Doppler processing required a total of fewer than 5 minutes. Calculation times are dependent on imaging depth and sampling. 3-D rendering was performed using the ‘volren’ function of the Amira software (6.0 Beta, Visualization Sciences Group, Burlington, MA).

Experimental Setup

To validate that 3D coherent compounding could be used to map flow in small vessels, a phantom consisting of Tygon tubes of 600-um, 1.2-mm, and 2.1-mm inner diameters filled with whole milk put in motion using a peristaltic pump at 2 mL/s was used to characterize the resolution achieved by the system. More specifically, each tube was imaged using plane waves tilted at -2°, 0° , 2° in both lateral directions (which corresponds to a total of 9 plane waves repeated twice per volume for a total of 18 emissions) with a synthetic receive aperture at a depth of 3 cm, which corresponds to a pulse repetition frequency of 10304 Hz and a volume rate of 572 volumes per second. A total of 200 volumes were acquired. The thyroid of one healthy volunteer was then imaged using the same parameters. To show the feasibility of quantitative coherent compound Voxel-wise Pulsed Doppler Imaging and Color Doppler Imaging were performed in a field of view comprising the carotid artery and the jugular vein of one healthy volunteer using 4 plane waves tilted at ±2° in both lateral directions at a pulse repetition frequency of 10304 Hz, resulting into a volume rate of 1288. A smaller number of tilted plane waves was used to increase the volume rate in order to limit alisasing. A total of 2000 compounded volumes were acquired, corresponding to an acquisition time of 1.55 s.

Results

Figure 2 shows the results obtained when performing 3D Power Doppler Imaging of tubes of 0.6-mm, 1.2-mm, and 2.1-mm inner diameters. One can observe that in the case of the two latter, their contours are well defined in the entire field of view. For the smaller tube, while it remained visible, we can observe that we are approaching the imaging resolution. Indeed, the wavelength of a 3-MHz pulse is 0.513 mm, and, since, 2 cycles were emitted, the vessel is actually smaller than the emitted ultrasound pulse. A full 3-D depiction of this imaging volume is provided in supplementary materials (supplementary video 1).

Figure 2.

Three views of the three tubes imaged using Ultrafast Power Doppler in this study. The contrast originates from the flow of whole milk in tubes of varying diameter, namely, 0.6, 1.2 and 2.1 mm. One can observe that the two latter can be readily resolved, while we are close to the resolution limit in the case of the 0.6-mm tube. Corresponds to Supplementary Video 1. 572 volumes per second.

Figure 3 and supplementary video 2 show the thyroid of a human subject from three different points of view when using coherent compounding. Complex architectures can be resolved in full the three dimensional space and down to very small diameters, sometimes on the order of a few wavelengths.

Figure 3.

Three views of the vasculature of the human thyroid. Vessels irrigating the thyroid (approximately delineated). Corresponds to Supplementary Video 2. 572 volumes per second.

Finally, Figure 4 shows frames of supplementary video 3, which depicts the Ultrafast Color Doppler ciné-loop. One can observe the pulsatility of the flow in the carotid artery, and its absence in the jugular vein. Moreover, opposite flows are clearly depicted. Using the same dataset, Voxel-wise Pulsed Doppler was obtained in the entire volume in less than one minute of calculations. These time-frequency planes fully quantify the velocity of the blood in a given voxel over time. Indeed, one can observe the increase in velocity in the carotid artery at the beginning of the cardiac cycle, while a flat spectrum is obtained in the jugular vein.

Figure 4.

Quantitative Ultrafast Doppler Imaging of the carotid artery and the jugular vein of a healthy volunteer. For each voxel, Color Doppler maps and the Pulsed Doppler spectrum can be obtained using a single dataset. One can observe the presence and lack of pulsatility in the carotid artery and jugular vein, respectively. Corresponds to Supplementary Video 3. 1288 volumes per second.

Discussion

In this study, we demonstrated the feasibility of performing non-invasive Ultrafast Ultrasound Imaging in three dimensions using a 3-MHz 2-D phased array probe. More specifically, we have shown that by coherently compounding plane waves, highly sensitive 3-D Power Doppler imaging can be performed. Objects on the order of the ultrasound wavelength were resolved, indicating that very small vessels could be imaged in vivo. More specifically, we have shown in phantoms that a resolution of at least 0.6 mm could be achieved using a 3-MHz probe, which corresponds to fewer than 2 wavelengths, and that the method was sufficiently robust to detect vessels of similar sizes in the human thyroid in vivo, despite potential hand motion, heartbeat, and breathing artifacts. The technique also maintained its quantitative aspects even when using coherent compounding; indeed, Color Doppler ciné-loop and voxel-wise Pulsed Doppler time-frequency planes were obtained in a single acquisition using 4 tilted plane waves in the human carotid artery and jugular vein in vivo.

Mapping the vasculature is widely used in many medical specialties. Indeed, coronary angiography for the diagnosis of cardiac infarcts, peripheral angiography of deep vein thrombosis, and cerebral angiography of stroke patients are commonly used in practice. While techniques based on magnetic resonance imaging and computed tomography are well established, they can often be slow and costly, and require catheterization, contrast-agents injection, and ionizing exposure in most cases. While the technique presented herein has not been compared to a gold standard and can probably not image with a similar resolution -a ballpark value for angiography resolution would be a few hundreds of microns [33]- when used as is, smaller vessels albeit at shallower depths are expected to become visible simply by using a higher frequency probe. 3-D Ultrafast Ultrasound Imaging thus paves the way for the fully non-invasive imaging of the vasculature in humans and can be performed at the bedside of patients that may be too frail to undergo a magnetic resonance imaging or computed tomography exam. Additionally, although this study focused on the use of power Doppler processing for the segmentation of blood vessels, functional information emanating from Color and Pulsed Doppler processing can also be obtained using the same acquisition dataset, and can therefore be used to map blood velocity and vessel resistivity [18] in vivo as well.

Limitations to this study include the low frequency used, which is not optimal for mapping small blood vessels. The acquisition system would also benefit from a larger number of receive channels. Indeed, having to repeat each emission twice increases the number of plane waves that need to be compounded and increases the level of decorrelation induced by blood and tissue motion in in-vivo applications [24]. As our system contained only 1024 channels, the aperture associated with a fully sampled matrix array in which elements are separated by a half-wavelength is small. Indeed, even at the low frequency of 3 MHz, the probe used in this study had a limited aperture (approximately 1cmX1cm) and thus limited resolution and field of view. The same number of element used with a higher frequency probe would have resulted in a very small field of view. To palliate this issue, one approach would consist in building higher-channel-count programmable scanners and is the object of on-going work in our group. Another consists in using diverging waves to increase the field of view, but in turn may reduce the signal-to-noise ratio and contrast when compared to plane waves [13]. Detectability in Ultrafast Power Doppler Imaging depends not only on the ultrasound resolution, but also on the blood flow and blood velocity. The phantom study was performed for the purpose of a proof of principle, but consists of a crude approximation of blood and blood vessels. Indeed, the material of the tubes created reflection artifacts and whole milk is not a potent blood phantom. Phantom studies that include a more realistic blood phantom, along with the thorough study of the influence of blood volume, blood flow, and blood velocity on 3D Ultrafast Ultrasound Doppler Imaging is the object of on-going studies.

In conclusion, by extending the use of 3-D coherent compounding of tilted plane waves for Ultrafast Power Doppler, we have demonstrated the feasibility of ultrasound-based 3-D mapping of blood vessels in humans using Ultrafast Power Doppler Imaging. As this technology is fully non-invasive, non-ionizing, low-cost, and can be applied at high volume rates, a number of applications are foreseeable in which computed tomography or magnetic resonance imaging are currently required, and opens new possibilities for the diagnosis of cardiovascular diseases.