Contrast-enhanced ultrasound with optimized aperture patterns and bubble segmentation based on echo phase

Abstract

Amplitude modulation (AM) suppresses tissue signals and detects microbubble signals in contrast-enhanced ultrasound (CEUS) and is often implemented with checkerboard apertures. However, possible crosstalk between transmitting and non-transmitting array elements may compromise tissue suppression in AM. Using AM aperture patterns other than the conventional checkerboard approach (one on, one off) may reduce the degree of crosstalk and increase the contrast-to-tissue-ratio (CTR) compared to conventional AM. Furthermore, previous studies have shown that the phase difference between the echoes in AM pulsing sequence may be used to segment tissue and microbubbles and improve tissue signal suppression and the CTR of CEUS images. However, the CTR of the image produced by alternative AM aperture patterns and the effect of segmentation approach on these alternative apertures has not been investigated. We evaluated a number of AM aperture patterns in order to find an optimal AM aperture pattern that provides the highest CTR. We found that the aperture which uses alternating groups of 2 elements, AM2, had the highest CTR for the probe evaluated. In addition, a segmentation technique based on echo phase differences (between the full and the half pulses, ΔΦ_AM, between the complimentary half pulses, ΔΦ_half, and the maximum of the two ΔΦ_max) was also considered in the AM aperture optimization process. The segmentation approach increases the CTR by about 25 dB for all apertures. Finally, AM2 segmented with ΔΦ_max, had a ~7 dB higher CTR in a flow phantom and ~6 dB higher contrast in a perfused pig liver than conventional AM segmented with ΔΦ_AM and it is the optimal transmit aperture design.

Introduction

Contrast-enhanced ultrasound (CEUS) utilizes the nonlinear echoes from microbubbles (Leighton 2012) for real-time imaging of perfusion of organs and tumors (Averkiou, et al. 2020, Becher and Burns 2012, Wilson, et al. 2000). CEUS provides information of the macro- and microvasculature of organs and tumors (D’Onofrio, et al. 2015, Piscaglia, et al. 2010). Synthetic aperture imaging approaches have paved the way for faster frame rates (Jensen JA, et al. 2006). In recent years, ultrafast imaging techniques such as plane wave imaging (PWI) and diverging wave imaging (DWI) have emerged (Tanter and Fink 2014). PWI/DWI excites a large number of the array elements to cover the whole field of view in one transmit event while beamforming multiple receive lines and thus allowing a significant improvement in frame rate. A previous study has demonstrated the possibility of achieving frame rates close to 20,000 Hz with PWI, over 500 times greater than those achieved with conventional focused beamforming (Provost, et al. 2014). Combining ultrasound localization microscopy with CEUS has led to super resolution imaging. Some approaches used ultrafast imaging techniques like PWI/DWI (Couture, et al. 2012, Errico et al. 2015) and others used normal/conventional frame rate imaging (Siepmann, et al. 2011, Viessmann, et al. 2013, Christensen-Jeffries, et al. 2015, Kanulas et al. 2018). Super resolution imaging has enabled visualization of the microvasculature of the organs down to a few microns.

Nonlinear pulsing schemes are transmit imaging strategies that isolate the nonlinear components of the echo to form the image. In nonlinear pulsing schemes, the scattered echoes of a series of transmitted pulses are combined in a way that the linear signals (produced by tissue) are cancelled while the nonlinear signals (produced by microbubbles) are detected. The two most common nonlinear pulsing schemes today are pulse inversion (PI) (Simpson, et al. 1999) and amplitude modulation (AM) (Brock-Fisher, et al. 1996,). PI involves transmitting 2 pulses per line with the second pulse being the inverse of the first pulse (different in phase by π). By adding the produced echoes, the fundamental (and all odd harmonics) are removed prior to any RF-filtering operation while the second (and all even) harmonics are detected. The AM technique involves transmitting 2 pulses at different amplitudes. A commonly used version of AM is one where the amplitude of the second pulse is half that of the first pulse. The linear signals are removed by scaling the echoes of the second pulse by a factor of 2 and then subtracting them from the echoes of the first.

Tissue signal suppression is important in CEUS to achieve good image contrast and microbubble specificity. Incomplete tissue signal cancellation may mask real enhancement caused by microbubbles or be incorrectly interpreted as real enhancement (Averkiou, et al. 2020). However, transmitting two identical pulses with different polarities or with an exact amplitude ratio at different voltages is challenging for the hardware due to nonlinearities in the signal generation and power amplifier (Averkiou, et al. 2020). A slight imperfection of the pulse inversion or amplitude change of the transmitted pulses would inevitably result in incomplete cancellation of the linear tissue signal.

Previous studies have shown that implementing AM with complimentary apertures improves tissue signal cancellation and requires less complex hardware (Averkiou, et al. 2020, Brock-Fisher, et al. 1996, Couture, et al. 2012, Phillips and Gardner 2004, Tremblay-Darveau 2016). This technique involves transmitting 3 pulses with the following sequence: p_even, p_full, and p_odd.With this technique, the even elements are used for transmitting p_even, the odd elements are used for p_odd, while all the elements are used for p_full. Since only half the elements are used for p_even and p_odd (when considering AM with a factor of 2 between the two amplitudes), the pressure produced by either p_even or p_odd, is half of that produced by p_full. It is possible that the crosstalk between transmitting and non-transmitting array elements may compromise the degree of tissue signal suppression in AM (Averkiou, et al. 2020). A previous study has shown that the crosstalk, generated by the parasitic vibrations between the transmitting and non-transmitting array elements, changes the electrical impedance and directivity of the individual array element (Bybi, et al. 2020). Moreover, the degree of crosstalk may be more severe in ultrafast CEUS compared to that in conventional focused ultrasound since much larger apertures are used for ultrafast imaging. In our previous work, we observed that the linear signal cancellation of a L7–4 array in plane wave mode is ~7 dB lower than that in focused mode at low MIs (≤ 0.1) (Lai and Averkiou 2021). The spatial characteristics of the acoustic fields in the AM pulsing sequences may also impact the nonlinear response from microbubbles. Using an alternative combination (instead of p_even and p_odd) for AM that still uses half or all the elements may reduce the degree of crosstalk between on/off elements while generating more nonlinear microbubble response compared to that generated by conventional AM. However, to the best of our knowledge, no studies have investigated the possibility of performing AM with other combinations of aperture patterns.

The image quality of CEUS is also limited by the tissue harmonics generated from the nonlinear propagation of sound waves in tissue. Optimal CEUS imaging is performed at low mechanical index (MI ≤0.1) to avoid microbubble destruction and the generation of tissue harmonics typically generated at MI>0.3, which may interfere with or mask the microbubble signals in CEUS images. However, even at low MIs the tissue still generates some nonlinear response, which degrades the image contrast of CEUS (Li, et al. 2006, Phillips 2001). Efforts have been made to segment tissue echoes from microbubble echoes. Recently, it has been shown that microbubble contrast may also be generated by a difference in the phase between full and half amplitude pulses in AM (Tremblay-Darveau, et al. 2018). This phase difference is amplitude dependent. Since AM sequences in tissue are not expected to result in the same degree of phase difference as that in microbubbles (Keller, et al. 2022), utilizing phase difference in AM pulses as a segmentation technique can therefore contribute towards better tissue signal suppression and higher image contrast (Tremblay-Darveau, et al. 2018, Keller, et al. 2022)]. For this tissue segmentation technique to work, the phase difference in microbubble echoes should be significantly greater than that in tissue. Previous studies have only utilized the phase difference produced by different pressure amplitude between full and half amplitude pulses for image segmentation. No studies have investigated the possibility of utilizing the phase difference between the complementary half amplitude pulses for phase segmentation. The phase difference between the complementary half amplitude pulses results from the difference between the spatial distribution of the complementary fields, which may induce additional microbubble phase response and would be useful for improving segmentation of microbubble signals.

In the present study, we first evaluate the contrast-to-tissue ratio (CTR) of ultrafast CEUS when imaging with different transmit aperture patterns and identify an optimal aperture for AM that would improve tissue signal suppression and increase CTR for ultrafast CEUS. Second, we implement a segmentation approach based on the phase difference between echoes from both full and half apertures and from the complementary half aperture pulses on the new transmit aperture patterns considered and evaluate the CTR benefits.

Materials & Methods

A. Flow phantom experiment setup

The experimental setup used for evaluating different apertures and different phase segmentation approaches is shown in Figure 1. A closed loop flow phantom with a single flow channel was used. The main components of this setup are: (1) a peristaltic pump that consists of Masterflex L/S Variable Drive, 600 rpm (Model 07528–10; Cole Parmer, Vernon Hills, IL) with an Easy-Load II pump head (Model 77200–62; Cole Parmer), and (2) a peripheral vascular Doppler flow phantom (Model 523A-modified; ATS Laboratories, Bridgeport, CT), of which we used the 8.0 mm conduit. Tygon Silicone Tubing (Cole Parmer) of 1/4 inch (=6.4 mm) inner diameter was used to connect the various components. A short stretch of Masterflex L/S Platinum-Cured Silicone Tubing (model 96410–24; Cole Parmer) was used for the peristaltic pump. Quick connect and disconnect fittings (Colder Products Company, St Paul, MN) were used for connections. The flow rate was controlled at 140 mL/min with an average flow velocity of 7.2 cm/s. The Philips L7–4 array (Philips Healthcare, Bothell, WA, USA) was coupled to the tissue phantom with ultrasound gel and was held by a passive arm. The alignment was such that the center of the imaged conduit was at 3.5 cm image depth. The total image depth was 6 cm. Optison microbubbles (GE Healthcare, Chicago, IL, USA) were prepared as described by the manufacturer. They were then drawn from the vial with a 21G needle and diluted to a 1/10000 concentration in deionized water, before the solution was flowed within the flow-system. The concentration of the microbubble is ~4.3E4 MB/mL.

Figure 1.

Experimental setup for the flow phantom system.

B. Perfused pig liver experiment setup

The perfused pig liver model provides an ideal environment for pre-clinical imaging investigations since it is very similar to an in vivo environment. The setup for perfused pig liver experiment is a slight modification of the setup described previously (Keravnou, et al. 2016). Briefly, healthy porcine livers were isolated and surgically removed from a deceased pig within 30 minutes of the animal’s death. This is to limit ischemic tissue death and vascular stenosis by thrombosis. The portal vein and hepatic artery of the liver were then flushed with 8 liters of lactated ringer’s solution (6 liters at room temperature and 2 liters ice cold) followed by 3 liters of ice-cold histidine-tryptophan-ketoglutarate (HTK) solution (prepared in-house). Subsequently, the liver was placed in 2 liters of HTK solution in static cold storage and transported back to the laboratory, where it was connected to a normothermic machine-perfusion system. The liver was perfused with 8 liters of blood mimicking fluid that consist of Powdered Williams Medium E, sodium bicarbonate, heparin, insulin, and dexamethasone. The liver was also supplied with 95% oxygen/5% carbon dioxide via an oxygenator in the machine-perfusion system. The alignment was such that the center of the liver was at 2 cm, which is also the center of the total image depth (4 cm). Optison microbubbles (GE Healthcare, Chicago, IL, USA) were prepared and injected in the liver in the same manner as done in the flow phantom.

C. Evaluation of aperture patterns for AM

AM consists of transmitting 3 pulses with the following sequence: p_even, p_full, and p_odd. With this technique, the transmit voltage and transmit beamforming delays are not changed between p_even, p_full, and p_odd. As a result, the effective beam of the sum of p_even and p_odd is identical to that of the p_full. We can further generalize AM by transmitting 3 pulses with the following sequence: p_h, p_full, and p_h∗ (not just even-full-odd). Half of the elements are used for transmitting p_h, while the other half are used for p_h∗, and all the elements are used for p_full.

Two possible combinations of aperture patterns used for p_h, p_h∗, and p_full are shown in Figure 2, where the elements in blue are transmitting elements and the elements in white are non-transmitting elements. To evaluate the different aperture patterns, we consider the following two parameters: (1) the number of on/off interfaces in the selected aperture pattern, and (2) the degree of overlap between two complementary apertures, which is calculated as the number of transmitting elements in the overlapped area between complementary apertures. The area of an aperture is defined as the total space starting from the first transmitting element to the last transmitting element. As shown in Figure 2, the complementary apertures in example 1 are highly overlapped and there are 12 transmitting elements in the overlapped area (within the red dashed lines). In addition, each aperture pattern has many on/off interfaces between elements. Conversely, the complementary apertures in example 2 (which admittedly is an extreme case and shown here for demonstration purposes) are not overlapped at all and have only 1 on/off interface between elements. We expected the acoustic fields of p_h and p_h∗ to be more uniform across the whole aperture when the complimentary apertures are more overlapped with each other. In addition, we expected the crosstalk would increase when more on/off interfaces are in an aperture pattern and hence reduce linear signal suppression. We evaluated the acoustic field and the contrast-to-tissue ratio (CTR) of the CEUS images produced by 26 different aperture patterns, spanning a wide range of possible combinations. Each combination has different degree of overlap between the complimentary apertures and the number of on/off interfaces as it is shown in Figure 3. AM1, AM2, and AM3 shown in Figure 3 are aperture patterns that use alternating groups of 1, 2 or 3 elements, respectively. For example, AM3 means 3 elements on and 3 elements off.

Figure 2.

Two examples of the 26 aperture patterns considered. The blue elements are transmitting elements and the white elements are non-transmitting elements. Example 1 has high overlap of its two complementary aperture patterns (indicated by red dashed lines) while example 2 has no overlap. Example 1 has more on/off interfaces than example 2 which has only 1.

Figure 3.

The degree of overlap between complementary apertures and the number of on/off interfaces for the 26 apertures considered. Each blue circle represents a unique AM aperture pattern. AM1, AM2, and AM3 are aperture patterns that use alternating groups of 1, 2 or 3 elements, respectively.

We also considered using spatially more dissimilar p_h and p_h∗ fields generated by random selection of the elements, referring to this scheme as random AM (RAM). We evaluated the acoustic field produced by 3 combinations of random aperture patterns. RAM is considered here to further increase the spatial difference between the p_h and p_h∗ fields. For RAM, half of the array elements were randomly selected for transmitting p_h while the other half for transmitting p_h∗. We evaluated 3 RAM examples with different number of on/off interfaces (47, 65, and 92) to produce various degrees of spatial differences between p_h and p_h∗ fields. The apertures for RAM were carefully chosen such that they have comparable degree of overlap between complementary apertures and minimal grouping of elements in the aperture pattern to ensure the acoustic fields that RAM produced are as uniform as possible despite having different level of spatial variations. Combinations of aperture patters were implemented on the Verasonics with a Philips L7–4 array that has 128 elements.

1). Simulations of the acoustic field produced by different apertures

Since linear propagation would typically apply in low MI (MI<0.1) contrast applications, the acoustic field produced by a L7–4 array using each of the 26 apertures was modeled with Field II (Jensen 1996, Jensen and Svendsen 1992). Field II is a transducer field simulation software based on the Rayleigh integral and is capable of modeling the acoustic field of non-axisymmetric sources with arbitrary focusing, apodization, and attenuation (Jensen 1996, Jensen and Svendsen 1992). The physical dimension of the transducer stack of the L7–4 array is 38.4 mm × 7.5 mm (azimuthal × elevation). For the simulations in this study, each simulation was modeled with 200 MHz sampling frequency and 25 sub-division in element height and width. The transmitted pulse has 2.5 cycles with a center frequency at 3.5 MHz. The imaging depth for the simulation was 6 cm.

2). CTR when using different apertures

We evaluated the CTR of CEUS images produced by 26 different apertures in a flow phantom system. Ultrafast acquisitions were performed with a Verasonics scanner (Verasonics Inc., Kirkland, WA, USA) connected with a L7–4 array. The transmitted pulse has 2.5 cycles with a central frequency at 3.5 MHz. The pulse repetition frequency (PRF) was 4000 Hz. The derated mechanical index (MI) of the full pulse in the pulsing sequence was 0.07. These imaging parameters were used for all the experiments in this study.

CTR was calculated as the mean intensity difference between a region with microbubbles and a region with tissue only and was recorded as dB. Three trials that each has 100 CEUS images were recorded. The CTR calculation was performed on 100 CEUS images produced by the same combination of aperture pattern and the average of the trial was recorded. The average CTR from three trials was reported as the final CTR value for that aperture pattern for comparisons.

D. Segmentation of microbubble signals based on the phase of echoes

1). Phase difference between the echoes from the full and the half amplitude pulses in AM

AM was implemented on the Verasonics scanner according to the following equation (Averkiou, et al. 2020, Brock-Fisher, et al. 1996, Couture, et al. 2012, Phillips and Gardner 2004, Tremblay-Darveau 2016):

$$p{s}_{AM}(i,j)=\frac{1}{3}p{s}_{\mathit{\text{full}}}(i,j)-\frac{1}{3}p{s}_h(i,j)-\frac{1}{3}p{s}_{h\ast }(i,j),$$ (1).

where ps_h, ps_full, and ps_h∗ are the echoes from microbubbles when transmitting p_h, p_full, and p_h∗ at point (i,j) where i is the time index (depth) and j the line number. The phase difference (ΔΦ_AM) resulting from the pressure amplitude difference between ps_full and the sum of ps_h and ps_h∗ were calculated according to equation (2):

$${\Delta \Phi }_{AM}(i,j)=\angle \frac{p{s}_{\mathit{\text{full}}}(i,j)}{\left\{p{s}_h(i,j)+p{s}_{h\ast }(i,j)\right\}},$$ (2)

where ∠ is the “angle” operator on the complex signals. Equation (2) is equivalent to a two-pulse 1-D autocorrelator (Kasai, et al. 1985). Since the sum of p_h and p_h∗ is identical to that of p_full, the acoustic field phase produced by the sum of p_h and p_h∗ is identical to that produced by p_full. We can then simplify equation (2) to:

$${\Delta \Phi }_{AM}={\Phi }_{\mathit{\text{full}}}-{\Phi }_{\mathit{\text{half}}},$$ (3)

where Φ_full and Φ_half are the phase from the microbubble scattering of ps_full and the sum of ps_h and ps_h∗, respectively.

2). Phase difference between the echoes of complementary half amplitude pulses in AM

In order to further differentiate tissue and microbubble echoes we will also use the phase difference between ps_h and ps_h∗. We hypothesized that the difference between the spatial distribution (diffraction pattern) of acoustic fields produced by p_h and p_h∗ may induce additional phase difference in microbubble response, which could be used to improve the segmentation of microbubble signals. The phase difference (ΔΦ_half) resulting from the difference between the spatial distribution of p_h and p_h∗ fields were calculated according to equation (4):

$${\Delta \Phi }_{\mathit{\text{half}}}(i,j)=\angle \frac{p{s}_h(i,j)}{p{s}_{h\ast }(i,j)}.$$ (4)

However, since p_h and p_h∗ fields are spatially different, the acoustic field phases produced by them are different and would not cancel out each other in equation (4). Since the phase of microbubble echoes are constantly changing from −π to π over time during volumetric oscillation of microbubbles, the average value of the phase of microbubble echoes over time would be close to 0. As a result, the average of ΔΦ_half over time would be equivalent to the phase difference between the acoustic field phase of ps_h and ps_h∗. The phase difference (ΔΦ_half), after subtracting the average of ΔΦ_half over time is:

$${\Delta }_2{\Phi }_{\mathit{\text{half}}}={\Delta \Phi }_{\mathit{\text{half}}}-Avg\left[{\Delta \Phi }_{\mathit{\text{half}}}\right],$$ (5)

where Δ₂Φ_half is the phase difference from the microbubble scattering of ps_h and ps_h∗. According to equation (5), Δ₂Φ_half increases as the phase difference from the microbubble echoes of ps_h and ps_h∗ increases, which we hypothesized that can be achieved by using spatially more dissimilar p_h and p_h∗ fields.

3). Combined phase segmentation approach

Finally, we proposed an approach that uses the maximum phase difference (ΔΦ_max) between ΔΦ_AM and Δ₂Φ_half for phase segmentation. ΔΦ_max was calculated by using the maximum absolute phase difference at every point (i,j) between ΔΦ_AM and Δ₂Φ_half such that:

$${\Delta \Phi }_{max}(i,j)=\underset{i,j}{\text{max}}\left\{\left|{\Delta \Phi }_{AM}(i,j)\right|,\left|{\Delta }_2{\Phi }_{\mathit{\text{half}}}(i,j)\right|\right\}.$$ (6)

We expect that by utilizing the phase difference resulting from the pressure amplitude and the spatial difference of the fields at the same time, it would improve the segmentation of microbubble signals from tissue signals.

4). Phase difference and image quality when using different phase segmentation approaches

The phase difference between complex signals was calculated at the fundamental frequency. Complex signals for calculating the phase difference, including ps_h, ps_full, and ps_h∗, were filtered with a passband FIR filter of 2–4 MHz bandwidth to preserve the signal only at the fundamental frequency. We investigated the change in the phase difference as spatial differences between p_h and p_h∗ increased. Five combinations of aperture patterns, including AM1, AM2, and 3 RAM examples, were used for comparisons. We calculated the ΔΦ_AM, Δ₂Φ_half, and ΔΦ_max for all points in the image for AM1, AM2, and the 3 RAM examples.

To perform image segmentation, the phase difference information was used as a mask to remove the noise and the nonlinear tissue clutter. More specifically, the phase segmented signal ps^′_AM at point (i, j) was calculated by multiplying the original nonlinear signals ps_AM with the Heaviside function H(x) such that:

$$p{s}_{AM}^{\prime }(i,j)=p{s}_{AM}(i,j)H\left(\left|\Delta \Phi (i,j)\right|-{\Delta \Phi }_{th}\right),$$ (7)

where ΔΦ_th is a threshold value in rad. Pixels in ps_AM(i, j) that have smaller ΔΦ than ΔΦ_th were set to the average of the lowest value in each image of ps_AM(i, j). We then selected a threshold value ΔΦ_th that maximizes the CTR of the image. The segmented images may still present tissue or noise signals when the ΔΦ_th value is very low; however, most the microbubble signals may be eliminated when ΔΦ_th is too high. We used different ΔΦ_th values that maximized the CTR of the images produced by each combination of aperture pattern and phase segmentation approach. We compared the CTR of the CEUS images of a flow phantom with microbubbles produced by AM1, AM2, and 3 RAM examples, each segmented with ΔΦ_AM, Δ₂Φ_half, or ΔΦ_max.

Finally, we compared AM2 with AM1 in a perfused pig liver model (Izamis, et al. 2014). A 32s CEUS image loop of the pig liver vasculature was acquired by each of AM1 and AM2. We also compared the time-intensity curves (TIC) calculated with images produced by AM to that produced by AM2. TIC provides information of the wash in and washout of the microbubbles in a region of interest (ROI) (Averkiou, et al. 2020,) and it may be used (Aoki, et al. 2011, Pei, et al. 2013) for assessing tumor response after treatment (Lassau, et al. 2014). We formed TICs from a ROI in the liver microcirculation. The CEUS images produced by AM1 and AM2 for calculating the TICs were segmented with ΔΦ_AM and ΔΦ_max , respectively, to compare the performance of the proposed phase segmentation approach to that of a previous phase segmentation approach. ΔΦ_th values that maximized the difference in the average intensity of the ROI between 2 time points (before bolus injection and when microbubbles were fully perfused) were used for image segmentation.

RESULTS

A. The acoustic field produced by different apertures

Example Field II simulations of the acoustic field produced by 6 (of the 26) aperture patterns evaluated in this study are shown in Figure 4. In Figure 4, we chose these 6 combinations of aperture pattern to show how the acoustic fields are affected when complementary apertures are more overlapped (a)–(c) and how the acoustic fields are affected when number of on/off interface increases (d)–(f).We only show the p_h fields in the figure since p_h and p_h∗ are complementary and p_full is simply the whole aperture. The numbers in parenthesis in each figure represent the number of on/off interfaces and overlap in terms of elements of the complementary apertures, respectively. The acoustic field produced by p_h of AM1 (conventional AM with 1 on and 1 off elements) is shown in Figure 4 (f). As expected, the acoustic field is more uniform across the whole aperture when the complementary apertures are more overlapped with each other. In addition, the acoustic field is more uniform across the whole aperture when the aperture pattern has more on/off interfaces [Fig. 4 (c)–(f)]. The peak pressure of the p_h fields is larger when the complementary apertures are less overlapped with each other. For example, the peak pressure in Figure 4 (a) is about twice the peak pressure in Figure 4 (f). Field II simulations of the p_h and p_h∗ fields produced by RAM₄₁, RAM₆₅, and RAM₉₂ (subscript indicates the number of on/off interfaces) are shown in Figure 5. As seen in the figure, p_h and p_h∗ fields are spatially more dissimilar when RAM has a smaller number of on/off interfaces.

Figure 4.

Field II simulations of the acoustic field produced by the 6 (out of the 26) apertures considered. Only the p_h field is shown in the figure since p_h and p_h∗ are complementary. The numbers in parenthesis in the figure represents the number of on/off interfaces and overlap in terms of elements of the complementary apertures, respectively.

Figure 5.

Field II simulations of the acoustic field produced by the 3 random aperture patterns (RAM₄₁, RAM₆₅, and RAM₉₂). The subscript indicates the number of on/off interfaces in the aperture pattern.

B. The effect of different aperture patterns

CEUS images of the flow phantom produced by the apertures in Figure 4 (a), (e), and (f) are shown in Figure 6. In Figure 6, we chose these 3 combinations of aperture pattern to show the image produced by (a) the combination that has no overlap between the complementary apertures, (b) AM1 (conventional AM), and AM2 (the one with highest CTR). The intensity of the contrast signal is higher [Fig. 6 (a) vs (c)] as p_h and p_h∗ fields are more overlapped with each other [Fig. 4 (a) vs (f)]. We also observed that the intensity of the tissue signals and noise is lower [Fig. 6 (b) vs (c)] when complementary apertures have similar degree of overlap but lower number of on/off interfaces [Fig. 4 (e) vs (f)].

Figure 6.

CEUS images of a flow phantom produced by the 3 (out of the 26) apertures evaluated. The numbers in parenthesis in the figure represents the number of on/off interfaces and overlap in terms of elements of the complementary apertures, respectively.

Next, we evaluated the CTR of the images acquired with the 26 apertures considered. The ROIs for calculating the CTR of the image are shown in Figure 7 (a), where the ROI for contrast signals is shown in red and the ROIs for tissue signals are shown in yellow (we took the average between the two yellow ROIs). The ROI for calculation of the CTR covers the whole width of the field of view as it is shown in Figure 7. As a result, aperture patterns that produce spatially inhomogeneous fields would have low CTR results and would not be considered a good aperture pattern for CEUS imaging. To better illustrate how CTR changes between different aperture patterns, the CTRs of the images produced by 5 (out of the 26) apertures evaluated are shown in Figure 7 (b). We quantified these 5 configurations in Figure 7 to not only show the CTR of the 3 aperture patterns that has the highest CTR (AM1, AM2 and AM3), but also how changing the overlap between complimentary apertures affects the CTR [(41,20–60)].When the number of on/off interfaces remains the same, the CTR increased with more overlapped complementary apertures since more overlapped complementary apertures have higher sensitivity in detecting contrast signals, which can be clearly seen in Figure 7 (c). AM2 had the highest CTR. Although AM1 and AM2 have similar degree of overlap of complementary apertures (63 and 62, respectively), AM2 has half the number of on/off interfaces, which indicates lower crosstalk and thus better tissue signal suppression, compared to AM1 as shown in Figure 7 (d). The standard deviation of CTR, contrast signals, and tissue signals is shown in Figure 7(b), (c), (d), was 0.45–0.7 dB, 0.61–0.71 dB, and 0.01–0.17 dB, respectively.

Figure 7.

(a) The CEUS image of the flow phantom and the ROIs for calculating CTR. ROIs for contrast signal are red and tissue signal are yellow. (b) CTR of the image, (c) contrast signals, and (d) tissue signals produced by the 5 (out of the 26) apertures evaluated. Error bars represent standard deviation. The numbers in parenthesis in the figure represents the number of on/off interfaces and overlap in terms of elements of the complementary apertures, respectively.

C. The effect of phase segmentation

Images of ΔΦ_AM are shown in Figure 8 (a)–(c), where (a) shows Φ_full, (b) Φ_half , and (c) ΔΦ_AM produced by AM2. The phase values are shown in absolute scale for better visualization. As it is shown in Figure 8 (c), tissue has lower phase difference than microbubbles. Phase images for calculating Δ₂Φ_half are shown in Figure 8 (d)–(e), where (d) shows ΔΦ_half, (e) Avg[ΔΦ_half] from 100 CEUS images, and (f) the Δ₂Φ_half produced by RAM₄₁. The phase values are shown in absolute scale. As it is shown in Figure 8 (e), the average value of the phase produced by microbubbles over time was close to 0, as expected. Unlike an amplitude segmentation technique that removes both tissue and contrast signals below an amplitude threshold, a phase segmentation technique (Figure 8) is able to remove tissue signals without heavily impacting the contrast signals.

Figure 8.

Absolute phase of (a) Φ_full , (b) Φ_half , and (c) ΔΦ_AM produced by AM2. Absolute phase of (d) ΔΦ_half, (e) Avg[ΔΦ_half], and (f) Δ₂Φ_half produced by RAM_41.

Figure 9 shows |ΔΦ_AM| produced by AM1 and AM2. We observed that |ΔΦ_AM| of AM1 at most of the tissue regions is higher than 0.1 rad. In contrast, |ΔΦ_AM| of AM2 at most of the tissue regions has a lower value. As a result, it may be easier to segment contrast and tissue signals with a phase threshold for AM2 compared to AM1. We then compared the mean phase difference in the contrast region produced by different apertures. The ROI for calculating the average value of the phase differences in the contrast region is shown in red in Figure 7 (a). The CTR and |ΔΦ_AM| of a selection of aperture options with the highest performance (including the random apertures) that helps summarize the trends is shown in Figure 10. In (a), AM1 produced the highest |ΔΦ_AM| (blue line) and RAM₄₁ produced the lowest from the aperture patterns considered. This suggests that uniform p_h and p_h∗ fields would produce higher |ΔΦ_AM|. On the other hand, RAM₄₁ produced the highest |Δ₂Φ_half| (black line) and AM produced the lowest compared to other apertures. This confirms what we expected from equation (5) that more spatially different p_h and p_h∗ fields (Fig. 4) would produce higher |Δ₂Φ_half|. Finally, we observed that |ΔΦ_max| (pink line) is larger than |ΔΦ_AM| and |Δ₂Φ_half| for all apertures. Images segmented with ΔΦ_max will have higher CTR. In Figure 10 (b) we show the CTR of the images segmented with different phase segmentation approaches. The ROIs for calculating CTR of the image are shown in Figure 7 (a). The CTR of the image produced by different apertures without phase segmentation are shown as brown dashed line in Figure 10 (b). When the images were segmented with ΔΦ_AM (blue line), the images produced by AM2 have the highest CTR (42.9 dB), 7.4 dB higher than that produced by AM1 (35.5 dB). Images produced by the RAM approach have 2.6–4.6 dB higher CTR than that produced by AM1. Although AM1 produced the highest |ΔΦ_AM| in the contrast region [Fig. 10 (a)], it has the lowest CTR due to lower tissue signal suppression that compromised the CTR. Next, when the images were segmented with Δ₂Φ_half (black line), those produced by RAM₄₁ have the highest CTR (42.3 dB), 11.9 dB higher than that produced by AM1 (30.4 dB). The images produced by RAM₆₅ have the second highest CTR (41.3 dB) while the images produced by AM1 have the lowest CTR. This trend is consistent with what we observed in Figure 10 (a), where RAM produced larger Δ₂Φ_half than AM1 and AM2. Therefore, it would be optimal to segment contrast signals with Δ₂Φ_half in images produced by RAM. Finally, when images were segmented with ΔΦ_max (pink line), the ones produced by AM2 have the highest CTR (45.5 dB), 5.9 dB higher than that produced by AM1 (39.6 dB). We observed a 2.6–4.7 dB improvement in CTR (depending on combination of aperture pattern) when the images were segmented with ΔΦ_max compared to when segmented with ΔΦ_AM. The images produced with AM2 and segmented with ΔΦ_max have the highest CTR compared to that produced by any other aperture pattern and segmentation approach.

Figure 9.

The phase difference (ΔΦ_AM) between ps_full and ps_h + ps_h∗ in images produced with AM1 (a) and AM2 (b).

Figure 10.

(a) 3 types of phase differences (ΔΦ_AM, Δ₂Φ_half , and ΔΦ_max) produced by 5 different apertures. (b) The CTR of the image produced by the 5 apertures before and after phase segmentation with ΔΦ_AM, Δ₂Φ_half , or ΔΦ_max.

CEUS images of the flow phantom produced by AM1 and AM2, each segmented with ΔΦ_AM or ΔΦ_max are shown in Figure 11. The ΔΦ_th value used in each case is shown in the figure. We observed that the segmented images [(b), (c), (e), and (f)] have better tissue signal suppression compared to the original images [(a) and (d)], yet some contrast signals were inevitably lost. Comparing images segmented with ΔΦ_max [(c) and (f)] with those segmented with ΔΦ_AM [(b) and (e)], we observed that the bubbles were brighter and the tissue dimmer than the images segmented with ΔΦ_AM. We also observed that the bubbles were brighter in the segmented AM2 images [(e) and (f)] compared to those with AM1 [(b) and (c)]. When using ΔΦ_AM for segmentation, AM2 has higher CTR than AM1 since it has better tissue signal suppression and it requires a smaller ΔΦ_th value to remove tissue signals. Using a smaller ΔΦ_th value would preserve more contrast signals in the image. When using ΔΦ_max for segmentation, AM2 has more contrast signals than AM1 since AM2 produced a much higher |ΔΦ_max| than AM1 as it is shown in Figure 10 (a).

Figure 11.

CEUS images of a flow phantom produced by AM1 (a) and AM2 (b), each segmented with ΔΦ_AM (b, e) and ΔΦ_max (c, f).

CEUS images from the perfused liver produced by AM2 before and after phase segmentation with ΔΦ_max at different time points during a bolus injection (0, 9, and 13 sec) are shown in Figure 12 (a)–(f). The phase masks used to segment microbubble signals are shown in Figure 12 (g)–(h). Avg[ΔΦ_half] from 100 images before any microbubbles arrived was used for calculating Δ₂Φ_half [equation (5)]. We observed that before microbubbles arrived (at t = 0 s), the image segmented with ΔΦ_max (b) has better tissue signal suppression than without phase segmentation (a). The strong reflections from the cloth membrane holding the liver were also reduced in (b). The signals at the bottom left corner of the image were true contrast signals produced by microbubbles flowing in the solution. When the microbubbles first arrived in the field of view through the artery (at t = 9 s), we observed in (d) that performing phase segmentation with ΔΦ_max was still able to remove most of the tissue signals and reflections from the plastic plate in (c) without losing much of the contrast signals. Finally, when more microbubbles were in the liver at 13 s, the difference between (e) and (f) was small since most of the signals were produced by microbubbles. However, reflections from the plastic plate were not removed in (f) due to the nonlinear propagation artifact generated by contrast agents (Thapar, et al. 2012). The TICs in Figure 12 (g) show that AM2 (red line) has slightly better tissue signal suppression and thus better image contrast compared to AM1 (blue line) before segmentation. We next compared TICs calculated from images after segmentation in (h). We observed that the TIC from images produced by AM2, segmented with ΔΦ_max (pink line), shows better tissue signal suppression while preserving more contrast signals compared to that produced by AM1, segmented with ΔΦ_AM (black line).

Figure 12.

(a)-(f) CEUS images produced by AM2 before and after segmentation with ΔΦ_max at different time points (0, 9, and 13 sec) in the image loop. The time intensity curves acquired with AM1 and AM2 (g) before and (h) after segmentation. The ROI for tracking the time intensity curve is shown in yellow in (a).

DISCUSSION

A. Implications of aperture patterns for contrast imaging

We first evaluated the acoustic fields produced by different apertures and its relation to CEUS performance. Although it is impossible to evaluate every possible aperture pattern, the 26 apertures evaluated in this study provided a good variety to examine the overlap between complementary apertures and the number of on/off interfaces (Fig. 3), for optimal CTR and cover every possible case.

We observed in Figure 4 that the peak pressure of p_h field is less uniform and has areas where it is higher when the two AM complementary apertures are less overlapped with each other. Since the nonlinear signals in AM arise because the microbubbles respond nonlinearly to different amplitude pulses (Averkiou, et al. 2020, Eckersley 2008), the sensitivity of detecting contrast signals with AM is reduced when the peak pressure of p_h or p_h∗ field is closer to the peak pressure of p_full field. A clear example of this phenomenon was demonstrated when one aperture used the left half elements and the other the right half. This is obviously one combination to be avoided because it would produce images where the intensity would not be uniform [see for example Figure 6 (a)]. When the complementary apertures are highly overlapped as it is the case of the “classic” AM (checkerboard apertures with every other element off) the sensitivity of detecting contrast signals is greatly improved.

An approach to minimize the number of on/off interfaces for better tissue signal suppression while maintaining large overlap between p_h and p_h∗ fields is to use alternating groups of multiple elements for p_h and p_h∗ (e.g., 2 on, 2 off, etc.) as it is shown in Figure 4 (e) where groups of 5 elements are turned on and off. However, we observed that the produced p_h or p_h∗ fields became nonuniform as the number of grouped elements increased. Although grouping more elements for transmit may further reduce the effect of crosstalk between on/off elements, the sensitivity of detecting contrast signals is compromised since the peak pressure of the produced p_h and p_h∗ fields would be closer to that of p_full field.

Moreover, we observed increased generation of grating lobes in p_h and p_h∗ fields when using groups of 3 or more elements for p_h and p_h∗ [Fig. 4 (c) and (d)]. A grating lobe is an unwanted feature of the ultrasound beam that produces image artifacts, and it is generated as a result of undersampling of the spatial frequencies in the transmitted ultrasound beams (Ilovitsh, et al. 2019). Although the grating lobe artifacts are suppressed in the receive path with a fully sampled receive aperture, we still observed image artifacts produced by grating lobes in the nearfield (pointed by a red arrow) in Figure 13 (b) compared to that in Figure 13 (c) due to the gap between transmitting elements is larger (compare for example Fig. 4 (d) and (f)). In addition, the angle of the grating lobes from the center axis is smaller when the spacing between individual elements is larger (Paul, et al. 1997, Szabo 2004). Further increasing the gap between transmitting elements has a similar effect to the angle of the grating lobes. We observed that as the gap between transmitting elements increased from that in Figure 4 (d) to that in Figure 4 (c), the angle of the grating lobes is reduced, producing image artifacts at deeper regions of the image in Figure 13 (a). In Figure 13 (a), we observed increased image artifacts in deeper regions (15–30 mm) compared to that in (b) and (c). Additionally, since the main lobe would have lower pressure amplitude when grating lobes occur compared to that without grating lobes, grating lobes may also compromise the signal-to-noise ratio in the far field and therefore results in a reduced penetration depth.

Figure 13.

CEUS images of the flow phantom produced by the 3 (out of the 26) apertures evaluated in this study. The numbers in parenthesis in the figure represents the number of on/off interfaces and the overlap in terms of elements of the complementary apertures, respectively.

We found that the aperture pattern that uses alternating groups of 2 elements (2 on, 2 off) for this specific probe/frequency had the highest CTR. This is due to AM2 having comparable contrast sensitivity to AM1, yet better tissue signal suppression. When comparing AM2 to the aperture patterns that use alternating groups of 3 or more elements, AM2 has better contrast sensitivity and less image artifacts produced by grating lobes.

B. Phase segmentation of the microbubble signals

We have shown in our previous work that the phase difference in tissue echoes from full and half amplitude pulses should be close to zero (Keller, et al. 2022) when imaging at low MIs (MI ≤0.1) and generation of tissue harmonics is negligible. However, even at low MIs, the crosstalk between on/off elements may affect the precision of the pulse and therefore compromise the degree of tissue signal cancellation, resulting in false phase differences in the tissue [Fig. 9(a)]. Since AM2 has better tissue signal suppression compared to AM1, tissue has lower |ΔΦ_AM| in the image produced with AM2 than that with AM1 (Fig. 9). Consequently, it would be easier to segment contrast signals from tissue signals when using AM2 than when using AM1. As it is shown in Figure 10 (b) and Figure 11, images of AM2 have greater CTR improvement after segmentation with ΔΦ_AM compared to that of AM (28 dB vs 22 dB, respectively). The ΔΦ_th values that produced the highest CTR were consistent with the |ΔΦ_AM| values (~0.1 to 0.2 rad) observed in both the bubble simulations and the RF data of microbubbles in our previous work (Keller, et al. 2022).

In addition, we demonstrated that it is feasible to utilize the phase difference (Δ₂Φ_half) between complementary half amplitude pulses for segmentation of contrast signals. We found that Δ₂Φ_half increased as p_h and p_h∗ fields are more spatially dissimilar from each other. RAM produced larger |Δ₂Φ_half| compared to AM1 and AM2. Although AM2 also have similar p_h and p_h∗ fields, |Δ₂Φ_half| produced by AM2 is 0.3 rad larger than that produced by AM1 and only 0.05–0.3 rad lower than that produced by RAMs.

We observed in Figure 10 (b) that images segmented with ΔΦ_max have higher CTR than that segmented with either ΔΦ_AM or Δ₂Φ_half. Since microbubbles have different phase response in ΔΦ_AM and Δ₂Φ_half, a microbubble that has small phase difference in ΔΦ_AM may have high phase difference in Δ₂Φ_half and vice versa. Utilizing both ΔΦ_AM and Δ₂Φ_half at the same time for segmentation may therefore increases the CTR. Since tissue is expected to have low phase difference, it is reasonable to combine ΔΦ_AM and Δ₂Φ_half by using the maximum absolute phase difference (ΔΦ_max) at every point (i,j) between |ΔΦ_AM| and |Δ₂Φ_half|. Higher contrast sensitivity is very important in the segmentation process to remove tissue signals without losing much contrast signals. As it is shown in Figure 11, since ΔΦ_max has higher contrast sensitivity than ΔΦ_AM , a higher threshold value ΔΦ_th can be used for images segmented with ΔΦ_max to further improve tissue signal suppression without losing much of the contrast signals compared to that segmented with ΔΦ_AM only. This trend was also demonstrated in the perfused pig liver [Fig. 12 (h)], where the TIC in images segmented with ΔΦ_max shows better tissue signal suppression (lower intensity before microbubble appeared) and higher contrast sensitivity (higher intensity when microbubbles were well perfused) compared to that in the images segmented with ΔΦ_AM. The enhanced tissue signal suppression and contrast sensitivity offered by the proposed segmentation approach may be beneficial to imaging techniques that require high bubble specificity such as super-resolution imaging.

However, we observed in Figure 12 that although the phase segmentation technique had good tissue signal suppression on the reflections from the cloth membrane when no bubbles were present, it offered less improvements when bubbles were present due to the nonlinear propagation artifact. This is due to the ultrasound pulse being distorted after passage through dense microbubbles (Thapar, et al. 2012).

Although phase segmentation may not affect some perfusion parameters derived from TIC analysis such as time to peak and rise time, it could potentially influence as the calculation of peak intensity, and area under the curve (Averkiou, et al. 2020). Further studies are needed to investigate if using the perfusion parameters derived from the TIC in images after phase segmentation improves or reduces the accuracy of using those parameters for diagnosis.

Phase variations as a result of microbubble destruction at high pressure amplitudes (4.7 MPa) have been previously reported (Chomas, et al. 2001) and utilized as an imaging technique (Siepmann, et al. 2013). However, the generation of tissue harmonics at high MIs could also cause phase differences in tissue. In addition, certain microbubble contrast agents, such as Sonazoid (GE Healthcare, Amersham, UK), are used at higher MIs (Jang, et al. 2013), and could potentially benefit from the proposed segmentation techniques here to improve tissue signal suppression.

Tissue and probe motions may add an offset on the phases, impacting the efficacy of segmentation. Although motion artifacts are minimized in ultrafast imaging due to the ultrafast frame rate (>4000 Hz), the phase change between two successive full pulses could be used to further decouple the effect of the motion. In addition, the effect of motion is also compensated by the three-pulse contrast sequence used in this study (Brock-Fisher, et al. 2003, Whittingham 2005). To further reduce motion artifacts caused by probe motions, a surgical-type articulated arm can be used to fix the probe to the patients (Averkiou, et al. 2010).

C. Limitations

A selection of 26 combinations of aperture patterns were investigated in this study and there may be possibly others not considered. However, we have covered a broad range of aperture patterns including those that produce spatially homogeneous acoustic fields such as AM1-AM5 and those that produce spatially inhomogeneous acoustic fields. From the selected aperture patterns, we demonstrated the compromise between the number of on/off interfaces and the overlap of the effective apertures in improving CEUS images and specifically producing higher CTR.

In the present study we implemented all aperture combinations and evaluated their effect on CTR on a L7–4 array. Since the degree of crosstalk and the generation of the grating lobes is highly related to the specification of the imaging array such as the width and kerf of the individual element and the frequency, it is possible that apertures which use alternating groups of 3 or 4 elements (i.e., AM3 or AM4) have better CTR than AM2 when they are implemented on other configurations. However, the scientific analysis of our work still provides the basis for understanding the compromise between the sensitivity of detecting contrast signals and the tissue signal suppression when using different combinations of aperture patterns, which can also be utilized with other probes. The degree of crosstalk when using various transducer materials, e.g., single crystal transducer or capacitive micromachined ultrasonic transducer, must be further evaluated but we expect the physical principles explained here to hold.

The simple per-pixel threshold used in the presented phase segmentation technique may create a pixelated image due to removing signals in the image. Although our proposed phase segmentation technique, ΔΦ_max, has mitigated this issue compared to the original phase segmentation technique (ΔΦ_AM), some black holes can still be seen in the image. However, it is possible that other more advanced spatial smoothing algorithms could further improve the implementation of this technique.

CONCLUSION

We have evaluated plane wave AM aperture approaches including random apertures (reported here for first time) for CEUS in terms of CTR in a flow phantom and a perfused pig liver. We investigated how the number of on/off interfaces and the overlap of the effective apertures used in AM affects the CTR in contrast imaging. We found that AM2 which uses alternating groups of 2 elements (2 on, 2 off) had the highest CTR, 0.7 dB higher than that of conventional AM or any other aperture considered. We also developed a bubble-tissue segmentation approach based on differences in the phase of the echoes from the individual ultrasound firings in AM in order to further increase the CTR. We demonstrated that by using phase differences between the full pulse and the half pulse as well as between the two complementary half pulses we can increase the CTR of CEUS images by about 25 dB for all apertures due to further suppressing tissue signals and noise. We found that AM2 combined with our echo phase segmentation technique produced images of a single channel phantom and a machine-perfused liver with 6–7 dB greater CTR than any other aperture combination considered. The segmentation technique presented can be implemented any time AM is used and our aperture optimization methodology may be used for other probe geometries and frequencies.