Quantifying the Impact of Imaging through Body Walls on Shear Wave Elasticity Measurements

Abstract

In the context of ultrasonic hepatic shear wave elasticity imaging (SWEI), measurement success has been shown to increase when using elevated acoustic output pressures. Since SWEI sequences consist of two distinct operations (pushing and tracking), acquisition failures could be attributed to (1) insufficient acoustic radiation force (ARF) generation resulting in inadequate shear wave amplitude, and/or (2) distorted ultrasonic tissue motion tracking. Herein, an opposing window experimental setup that isolated body wall effects separately between the push and track SWEI operations was implemented. A commonly employed commercial track configuration was used, harmonic multiple track location SWEI (MTL-SWEI). The effects of imaging through body walls on the pushing and tracking operations of SWEI as a function of mechanical index (MI), spanning 5 different push beam MIs and 10 track beam MIs were independently assessed using porcine body walls. Shear wave speed yield was found to increase with both increasing push and track MI. Although not consistent across all samples, measurements in a subset of body walls were found to be signal limited during tracking and to increase yield by up to 35% when increasing electronic signal-to-noise ratio (SNR) by increasing harmonic track transmit pressure.

Introduction

Liver biopsy is the gold standard for staging liver fibrosis (Rockey et al., 2009; Goodman, 2007; Manning and Afdhal, 2008), but its invasive nature and lack of sensitivity (Piccinino et al., 1986; Bedossa et al., 2003; Regev et al., 2002) decrease its utility as a diagnostic tool for longitudinal monitoring. Shear wave elasticity imaging (SWEI) (Sarvazyan et al., 1998), a noninvasive diagnostic ultrasound technique that quantifies tissue stiffness, has found success in distinguishing advanced fibrosis (Wang et al., 2009; Schmeltzer and Talwalkar, 2011; Castéra et al., 2005; Kim et al., 2015).

SWEI uses acoustic radiation force impulse (ARFI) excitations from a diagnostic ultrasound transducer to induce localized tissue displacement (i.e., to push tissue) at depth (Sarvazyan et al., 1998). Equation 1 describes the acoustic radiation force (ARF), F, generated by an ARFI excitation (Nightingale et al., 2001). In the equation, α represents the tissue acoustic absorption textcolorbluecoefficient, I denotes the pulse averaged acoustic beam intensity, and c is the speed of sound in the tissue of interest.

$$\text{F}=\frac{2\alpha I}{c}$$ (1)

The tissue particle motion induced by the ARFI push and subsequent shear wave propagation is then tracked using the same transducer with standard ultrasonic tissue motion estimation methods to determine the shear wave speed (SWS) (Sarvazyan et al., 1998). Assuming the tissue is linear, elastic, isotropic, homogeneous, and incompressible, Equation 2 details the relationship between the estimated Youngs modulus, E, and the SWS, where ρ is the tissue mass density (Tanter et al., 2004).

$$E=3\rho SW{S}^2$$ (2)

During clinical application, studies have reported SWEI acquisition failures and unreliable SWS measurements in patients with elevated body mass index (BMI) and for some subjects with advanced hepatic fibrosis (Joshi et al., 2014; Virmani et al., 2013; Singal et al., 2013; Sporea et al., 2014; Klysik et al., 2014; Schuh et al., 2011). Because SWEI sequences consist of two distinct operations (pushing and tracking), acquisition failures have been attributed to (1) insufficient ARF generation resulting in inadequate tissue particle motion amplitude, and/or (2) distorted ultrasonic tissue motion tracking. Previous studies have demonstrated significantly decreased ARFI push amplitude in situ when imaging through some body walls as compared to phantoms (Deng et al., 2015; Amador Carrascal et al., 2016). Preliminary studies have also demonstrated that body wall distortion of the tracking beams decreases shear wave elasticity (SWE) measurement success (Deng et al., 2018). The use of elevated transmit pressures and therefore elevated mechanical index (MI) (Nightingale et al., 2015) (1.9 < MI < 3.5) for pushing and/or harmonic tracking beams has been demonstrated to improve SWE measurement success (Deng et al., 2015, 2018). However, it remains unclear whether one of the two SWEI sequence operations is more deleteriously impacted by the presence of body walls and if one of the operations might preferentially benefit from the use of elevated MI transmit pulse configurations.

Additionally, previous studies that reported improved SWE measurement success when using elevated MI for harmonic tracking only explored using single track location-SWEI (STL-SWEI) (Deng et al., 2018). STL-SWEI repeatedly uses a single tightly focused harmonic transmit tracking beam configuration combined with a series of ARFI pushes at increasing lateral distance from the track beam location to acquire a single SWE measurement (Elegbe and McAleavey, 2013; Hollender et al., 2017). From the tight transmit/receive focus of the tracking beam, STL-SWEI benefits from optimal focusing and reduced speckle bias (Elegbe and McAleavey, 2013; Hollender et al., 2014), but is slower to acquire data due to the use of multiple ARFI pushes for a single acquisition. In lieu of STL-SWEI, parallel receive multiple track location-SWEI (MTL-SWEI) (Shiina et al., 2015) is often used commercially. MTL-SWEI uses either plane-wave or weakly focused transmit track beams to image the tissue response to a single ARFI push. Although the defocused MTL tracking beam configurations are more susceptible to speckle bias as compared to STL (McAleavey et al., 2003) and typically are associated with larger displacement estimation jitter arising from lower SNR, MTL-SWEI sequences are much faster than their STL counterparts because they require only a single push per acquisition.

Herein, we employed an opposing window experimental setup that separately isolated body wall effects between the push and track SWEI operations. Additionally, instead of using the STL-SWEI tracking configuration of the previous study (Deng et al., 2018), we examined MTL-SWEI track sequences. We independently assessed the effects of body walls on the pushing and tracking operations of SWEI as a function of MI, spanning 5 different push beam MIs and 10 track beam MIs.

Materials and Methods

As shown in Figure 1, two opposing, synchronized transducers were used in combination for SWEI acquisitions: transducer I (an abdominal imaging DAX transducer, 1.6 MHz center frequency) imaging through body walls to best represent clinical abdominal imaging, and transducer II (an L7–4 transducer, 5.2 MHz center frequency) bypassing body walls to provide control reference measurements. Three imaging combinations were used as shown in Table 1: 1) push and track through body wall with transducer I, 2) push through body wall with transducer I and then track with transducer II to bypass body wall(ideal tracking), and 3) bypassing body wall on push with transducer II (ideal pushing) and then track through body wall with transducer I.

Figure 1:

Opposing window experimental setup. The DAX transducer was on top imaging the homogeneous liver phantom through body wall samples while submerged in attenuating fluid. The L7–4 transducer was placed below and aligned in the same imaging plane using custom transducer holders attached to a rigid aluminum rectangular frame. The L7–4 imaged into the liver speckle phantom directly using gel coupling and provided the control reference measurements.

Table 1:

3 SWE imaging combinations used for push and track sequences in the opposing window experimental setup. Since body wall samples are always secured to the DAX transducer, the DAX always pushes/tracks through body walls. Since the L7-4 is directly imaging into the liver phantom, the L7-4 always pushes/tracks without a body wall as a control reference for ideal conditions.

SWE Imaging Configuration	Push Transducer	Track Transducer
1	DAX	DAX
2	DAX	L7-4
3	L7-4	DAX

Transducer and System Configuration:

From Figure 1, the DAX transducer (Sequoia, Siemens Medical Solutions, Issaquah, WA, USA), operated using the Siemens Sequoia scanner, was secured to a rigid aluminum frame using a custom transducer holder in a downward imaging position. Opposing the DAX transducer and in the same imaging plane, the L7–4 transducer (Verasonics, Kirkland, WA, USA), operated using the Vantage 256 scanner, was also secured to the frame using a custom transducer holder in an upward orientation. The vertical distance between the two transducer faces was 85 mm. A custom stiff (5.13 m/s, 79 KPa) cirrhotic liver mimicking elastic phantom was used to model difficult imaging scenarios (Computerized Imaging Reference Systems, Norfolk, VA, USA). The L7–4 transducer imaged from the bottom through a thin layer of ultrasonic scanning gel (Clear Image Singles, NEXT Medical Products Company, Branchburg, NJ, USA). The phantom was removed from its original protective case to allow for imaging from both sides. To minimize precompression from the weight of the phantom on the L7–4 transducer face, a ring stand was used to support and hold the phantom in place. On top of the liver phantom, a tray with a Tegaderm membrane (Model 1626W, 3M, Saint Paul, MN, USA) cut-out window was used to hold a solution of evaporated milk and saline tuned to a sound speed of 1540 m/s and acoustic attenuation coefficient of 0.5 dB/MHz/cm as coupling media (Chivers and Hill, 1975; Duck, 1990; Bamber and Hill, 1979). A thin layer of water (< 1 mm) coupled the Tegaderm membrane to the liver phantom. Submerged in the evaporated milk coupling media, the DAX transducer imaged through fresh pork belly sections (including skin), sourced from the local grocery store, at room temperature after degassing to mimic human abdominal walls (Duck, 1990; Bamber and Hill, 1979). Herein, the pork belly sections will be referred to as body walls. To simulate compression of the transducer into the human abdominal wall that occurs during clinical imaging, 3-D printed clamps, which did not obstruct the imaging window, were used to push the body walls against the DAX transducer face.

Synchronization between the two scanners followed the diagram in Figure 2. In addition to the native line trigger output from the Sequoia scanner, a slower gating command via Ethernet cable using a TCP interface was used to isolate the trigger corresponding to the beginning of the SWEI sequence. The custom-built intermediary trigger box and computer controller combined the two Sequoia control signals and generated a SWEI acquisition trigger signal for the Verasonics scanner to synchronize data acquisition between the two scanners.

Figure 2:

Control diagram for one-way synchronization from Sequoia system to Vantage system. For a given transmit condition, two signal streams were output from the Sequoia scanner to an intermediary computer with triggering control. One output was a line trigger via BNC cable while the other was a synchronization gating command via an ethernet cable. The line trigger signals and gating signals were combined by the intermediary control computer to output a single acquisition trigger corresponding to the beginning of each SWEI acquisition as an input signal for the Vantage system via BNC cable.

SWEI Data Acquisition:

DAX Push and Track (Table 1: Imaging Combination 1):

When examining the effects of changing MI for ARFI push and SWEI track while imaging through body walls, only the DAX transducer was used, and the L7–4 transducer was turned off such that both the push and track signals propagated through the body walls. The DAX push parameters shown in Table 2 were selected to match a previous study (Deng et al., 2018). When varying the push MI, the push duration was held constant, thus the total energy deposited increased with increasing MI. The following 5 push MI values were used for the DAX transducer: 1.1, 1.5, 1.7, 2.1, and 2.3. Track transmit parameters used by the DAX are shown in Table 3. Fully sampled pulse inversion (PI) was used to recover harmonic 2.6 MHz center frequency data, which is typical for a commercial SWEI system. For the tracking sequence described, 10 track MI values were investigated ranging from 0.2 to 1.5. The maximum track transmit MI was limited by the scanner software and focal depth. When varying push MI, the maximum DAX tracking MI of 1.5 was used. When varying track MI, the maximum DAX push MI of 2.3 was used. Each acquisition was repeated 5 times with the same speckle realization to characterize scanner repeatability. A total of 28 body walls ranging between 1.8 cm and 4.1 cm thick were imaged, corresponding to body wall thicknesses seen clinically (Chang et al., 2021; Cho et al., 2019).

Table 2:

Push pulse parameters for the DAX and L7-4 transducers which were selected to generate matching displacement amplitudes at the same position in the phantom from each transducer when imaging without a body wall.

Push Sequence Parameters	DAX	L7-4
Tx Frequency [MHz]	2.2	4
Focal Depth [mm]	64	25
Lateral F/#	1.6	2
Push Cycles	400	790
Push Displacement [μm]	2.3	2.3
Default MI	2.3	2.5

Table 3:

SWEI tracking parameters for the DAX and L7-4 transducers.

Track Sequence Parameters	DAX	L7-4
Tx Frequency [MHz]	1.3	5.2
Tx Focal Configuration	Deep Focal Depth	Plane Wave
Rx Frequency Configuration	Harmonic	Fundamental
# Parallel Receive Beams	32	136
Lateral FOV [mm]	9.56	19.96
PRF [KHz]	5	10

DAX Push with L7–4 Track (Table 1: Imaging Combination 2):

When isolating the effects of body wall on only the push, DAX tracking was turned off and the control reference L7–4 track sequence bypassing the body wall was turned on. The L7–4 track transmit parameters used are displayed in Table 3. The L7–4 tracking did not use harmonic, angular compounding, or transmit focusing because it was imaging directly into the liver mimicking phantom which is a straightforward imaging situation without aberration or clutter generating structures. All 5 DAX push MIs used in imaging combination 1 shown in Table 2 were collected for imaging combination 2 for the same set of 28 porcine body walls. For each push transmit configuration and each body wall, 5 repeat acquisitions were collected.

L7–4 Push with DAX Track (Table 1: Imaging Combination 3):

When isolating the effects of body wall on only the track, the DAX tracking configuration was used with the L7–4 push configuration to provide reference control pushing. The L7–4 push parameters are shown in Table 2. This matched the peak displacement generated by the DAX at a magnitude of 2.3 μm at the same imaging depth plane in the phantom. Since the liver mimicking phantom was elastic instead of viscoelastic, the slight differences in push beamwidths and corresponding shear wave frequency bandwidths between the control and commercial scanners were not expected to exhibit bias from wave speed dispersion (Hall et al., 2013). The 10 DAX track MIs used in imaging combination 1 shown in Table 3 were collected for all 28 porcine body walls for imaging combination 3. For each track transmit configuration on each body wall, 5 repeat acquisitions were collected.

Spatial Coherence Data Acquisition:

Harmonic lag one coherence (LOC) is a B-mode image quality metric that has previously been shown to increase with increasing transmit pressure until a plateau with sufficient electronic SNR (Long et al., 2018). Harmonic short-lag spatial coherence (HSC) is another image quality metric sensitive to noise (Dahl et al., 2012). HSC has previously been shown to strongly correlate with B-mode in situ peak pressures (Zhang et al., 2021). In this study, these two B-mode image quality metrics were translated to search for correlation with SWEI quality.HSC and LOC were computed using pulse inversion B-mode channel data collected using the DAX transducer imaging through the body walls and the Sequoia scanner. For these measurements, the DAX B-mode channel data were collected using a transmit beam focal configuration matched to the push focal depth. From the center 11 B-mode lines, harmonic spatial coherence was computed from which HSC and LOC were extracted.

SWS Estimation:

Standard SWEI data processing methods were implemented (Deng et al., 2015, 2018). Briefly, tissue particle velocity was determined using a progressive-reference phase shift estimator, Kasai algorithm (Kasai et al., 1985), on beamformed in-phase and quadrature (IQ) data. The same processing code was used for data from each system. An axial averaging kernel size of 6 track beam wavelengths was applied to suppress noise. Additionally, a low pass filter with a cutoff frequency at 1 kHz was applied to the tissue velocity data to simulate clinical data processing (Deng et al., 2018). To improve shear wave signal SNR, tissue velocity data within the push beams 9 mm axial depth of field was averaged axially prior to SWS estimation. The mean SWS across the lateral field of view was then assessed using a Radon sum algorithm (Rouze et al., 2010). Each SWS measurement corresponding to each repeated acquisition was considered successful if the estimated SWS was within 10% of the ground truth SWS (5.13 m/s), which was measured in the phantom without a body wall. In addition to the Radon sum method, a deep-learning based SWS estimation technique, SweiNet (Jin et al., 2022), was also applied to the entire dataset. SweiNet has been shown to give consistent SWS results to Radon sum, and in addition it provides a quantitative uncertainty metric for each estimate with noisy data generating larger uncertainty.

Peak Displacement, temporal correlation, and Jitter Estimation:

Since it was not possible to measure peak in situ pressures directly for each body wall, peak in situ on-axis focal displacement amplitude from an ARFI push pulse was used as a proxy measurement for peak in situ pressure through the body walls. In situ peak displacement is a function of the time average intensity of the in situ acoustic beam (Equation 1) and scales with the square of in situ pressure. In situ peak displacement amplitude was quantified with the Verasonics system (no body wall). From the L7–4 tracked data collected after pushing with the DAX, point displacements were estimated using the same Kasai algorithm (Kasai et al., 1985) along with an axial averaging kernel of 6 track beam wavelengths. To compute the on-axis peak displacement, point displacements spanning 2 mm laterally centered around the push focus and 9.24 mm axially (the axial depth of field DOF) were averaged.

Temporal correlation is a measure of electronic SNR. In this case, temporal correlation was computed between the reference track frames before the push to determine the electronic SNR of the tracking beam for each body wall. This measurement was determined by computing the single sample cross correlation for each track beam line in sequentially acquired tracking frames. The correlation value was averaged axially throughout the push beam DOF and then across the 32 lateral track lines.

To compute the displacement estimation jitter magnitude, point displacements were computed for the reference track frames before the push, where there was no motion. These displacements were averaged over the DOF of the push beam. Jitter magnitude was then determined by taking the standard deviation from 19 consecutive frames (Walker and Trahey, 1995).

MI Measurement:

After completing 3D scans using a computer-controlled three-axis translation stage (MM3000, Newport Corporation, Irvine, CA, USA) in a deionized (DI) water tank, peak intensity pressure signals for each transmit configuration were recorded using an Acertara 804 polyvinylidene fluoride (PVDF) membrane hydrophone with a 0.6 mm active spot size (Acertara Acoustic Laboratories, Longmont, CO, USA) and a LeCroy Wavesurfer MXs-B oscilloscope (Teledyne LeCroy, Chestnut Ridge, NY, USA). The output voltage waveforms from the membrane hydrophone were then converted to pressures using magnitude deconvolution (Wear et al., 2014), where the frequency-dependent magnitude sensitivity of the hydrophone was accounted for using a 1–20 MHz calibration. The MI was then computed from the derated peak rarefactional pressure per AIUM/FDA guidelines (US FDA, 2019; AIUM, 1992; IEC, 2010, 2007). It should be noted that recent publications have detailed underestimation of MI due to spatial averaging across the hydrophone sensitive element (Wear, 2021; Wear et al., 2021). Using data from Wear et al. (2021), MI reported in this study are likely about 10% low for DAX push beams and 20% low for L7–4 push beams.

Results

Figure 3 plots the percentage of successful SWS measurements as a function of DAX transducer push MI. For both the DAX tracked and L7–4 tracked data, the SWS yield increases monotonically with increasing DAX push MI. When tracking with the L7–4 bypassing body walls, SWS yield increased between 26% and 42% compared to when tracking with the DAX through body walls. Still, tracking without body walls using the L7–4 at the maximum DAX push MI of 2.3 yielded 16% unsuccessful measurements highlighting the presence of cases where not enough energy was deposited through the body walls to generate a measurable shear wave even with ideal tracking.

Figure 3:

Percentage yield of successful SWS measurements as a function of ARFI push beam MI from acquisitions for all 28 body walls. (Blue) Results from imaging combination 1 where the DAX transducer was used for both push and track pulses through body walls. (Red) Results from imaging combination 2 where the DAX pushed and the L74 tracked.

Figures 4a–d illustrate sample velocity-based shear wave trajectories from a single body wall. When pushing and tracking with the DAX through the body wall (imaging combination 1) with a push MI of 1.1, no shear wave propagation is observed (Figure 4a); however, the control system was able to image and estimate the shear wave as shown in Figure 4c (5.67 m/s), albeit with a relatively low shear wave amplitude. Increasing the push MI to 2.3 through this body wall resulted in successful SWS measurement through the body wall (Figure 4b, 5.16 m/s), as well as with the control tracking system (Figure 4d, 5.31 m/s).

Figure 4:

Sample shear wave trajectories of acquisitions from varying push MI in one body wall. The left column corresponds to a push MI of 1.1. The right column corresponds to push MI of 2.3. The top row contains velocity data tracked through the body wall using the DAX (imaging combination 1). The bottom row displays the corresponding control L7–4 tracked velocity data (imaging combination 2). Note the same velocity scales are used for all 4 subplots.

Figure 5 shows trends of decreasing SWS estimation uncertainty with increasing push MI through body walls for both imaging combinations (tracking with (Figure 5a) or without (Figure 5b) body walls). As expected, tracking without body walls also decreased the SWS uncertainty at every MI measured compared to tracking through the body walls.

Figure 5:

SweiNet SWS estimation uncertainty as a function of push MI for (a) imaging combination 1 (pushing and tracking through the body walls), and (b) imaging combination 2 (pushing through body walls and tracking without body walls).

Figure 6 plots the percentage of successful SWS measurements as a function of DAX transducer track MI. SWS yield increased monotonically with DAX track MI when the DAX was used to both push and track through the body walls (imaging combination 1). Pushing with the L7–4, bypassing the body walls for the push while tracking through the body wall with the DAX (imaging combination 3), resulted in increasing SWS yield with DAX track MI up to a plateau (DAX track MI 1.0). As expected, L7–4 pushing resulted in higher SWS yield than DAX pushing through the body wall at every DAX track MI level examined. The difference in SWS yield between L7–4 pushes (no body walls) and DAX transducer pushes (through body walls) was larger for intermediate DAX track MIs: 21% for a DAX track MI = 0.8, decreasing to only 4% as DAX track MI increased to 1.5.

Figure 6:

Percentage yield of successful SWS measurements as the DAX harmonic SWE track pulse MI increased from 0.2 to 1.5 across acquisitions for all 28 body walls. (blue) Results from imaging combination 1 where the DAX transducer imaged through the body walls for both push and track pulses. (red) Results from imaging combination 3 where the DAX transducer was only used for track pulses through the body walls and the L7–4 transducer generated push beams without the body walls.

Figure 7 illustrates sample velocity shear wave trajectories from a single body wall obtained when tracking using two different track MIs. When tracking through the body wall with an MI of 0.2, no shear wave propagation was observed (Figures 7a and 7c) even when the control system was used to push (Figure 7c). Increasing the track MI to 1.5 resulted in a successful SWS measurements both when pushing through the body wall (Figure 7b), and when pushing with the control system (Figure 7d).

Figure 7:

Sample shear wave trajectories from varying DAX track MI in one body wall. The left column corresponds to a DAX track MI of 0.2. The right column corresponds to a DAX track MI of 1.5. The top row portrays shear wave velocity data pushed and tracked through the body wall using the DAX transducer (imaging combination 1). The bottom row displays the corresponding data from the same body wall while tracking through the body wall with the DAX but pushing with the control transducer bypassing the body wall (imaging combination 3).

Figure 8 shows trends of decreasing SWS uncertainty with increasing track MI through the body walls when applying SweiNet to estimate SWS for both imaging combinations 1 and 3 regardless of body wall presence during the push transmit. For track MIs > 0.5, pushing without body walls (Figure 8b) decreased the SWS uncertainty compared to pushing through body walls (Figure 8a).

Figure 8:

SweiNet SWS estimation uncertainty as a function of track MI when using (a) imaging combination 1 (push and track through body walls), and (b) imaging combination 3 (push without body walls and track with body walls).

Figure 9 plots the distributions of the peak displacement measured with the control system (bypassing the body wall) resulting from DAX pushes through the 28 body walls corresponding to imaging combination 2. The distribution of peak displacement amplitudes from all 5 repetitions, all 5 DAX push MIs, and across all 28 body walls were grouped based on the success or failure of the corresponding SWS estimate in Figure 9a. A Wilcoxon rank sum test performed on this binary grouping showed a significant difference of median peak displacement between successful and failed SWS estimates (p < 0.0001). When the same data was grouped by push MI in Figure 9b, a Kruskal-Wallis test showed that there were significant differences in median peak displacements between some of the different push MIs (p < 0.05). Specifically, Mann-Whitney U-tests demonstrated differences between push MIs of 1.1/1.5, 1.7/2.1, and 2.3 with p < 0.05. Push MIs of 1.1 and 1.5 were grouped together because no significant median peak displacement differences were observed between the two. The same was observed and down for Push MIs of 1.7 and 2.1.

Figure 9:

Distributions of peak ARFI displacement amplitude (tracked using the control system bypassing the body wall) when pushing through all 28 body walls with 5 push MIs (a) grouped based on SWS measurement success/failure and (b) as a function of DAX push MI through the body wall. Note that red bars indicate significant.

Figure 10 investigates the effects of the body wall on the mean jitter magnitude for the DAX transducer as a function of track MI. For a given track MI, inter-body wall median jitter was consistently 4.0 to 4.7 times larger than the no body wall case (obtained using the DAX transducer without an intervening body wall as a reference).

Figure 10:

Mean displacement jitter magnitude across the 32 parallel receive track lines for the DAX transducer imaging with (blue) and without (red) body walls using track MIs from 0.2 to 1.5. Error bars depict median and interquartile range across 28 body walls.

Figure 11 illustrates spatial coherence derived image quality metrics obtained when imaging through the body walls in the experiment. Second harmonic LOC, a B-mode image quality metric (Long et al., 2018), was plotted as a function of increasing track transmit MI for the body walls and the no body wall control (Figure 11a). The MI at which the LOC plateaus represents the minimum MI needed to achieve optimal B-mode image quality (Long et al., 2018). Harmonic LOC plateaued at an MI of 1.2 without a body wall (Figure 11 a, black curve) while most body walls were associated with plateaus at higher MIs. From Figure 11a, the body wall cases could be categorized into 3 groups based on LOC curve types: 1) linearly increasing then reaching a plateau (Plateau), 2) linearly increasing without reaching a plateau (Linear +), or 3) low flat LOC over the entire range of MIs tested (Low Flat). Figures 12–14 demonstrate the differences between the three groups on track and pushing during SWEI acqusitions.The entire spatial coherence curves for a transmit MI = 2.3 are plotted for 3 examples in Figure 11b. The no body wall case (LOC plateau of 0.99) exhibited the highest spatial coherence across all lags (black), while a body wall sample with relatively high LOC (0.95) that exhibited relatively high spatial coherence across all lags is portrayed in blue; the spatial coherence across all lags for the body wall with the lowest LOC (0.17) is portrayed in red. Finally, in Figure 11c the Harmonic Spatial Coherence (HSC), which is the sum of the coherences across multiple lags (Dahl et al., 2012), is plotted against peak ARFI displacement amplitude obtained when pushing with an MI of 2.3 through each body wall. The HSC is a metric previously shown to strongly correlate with in situ pressures (Zhang et al., 2021). While in situ pressure was not directly measured in these experiments, displacement amplitude is directly proportional to the applied ARF, and thus also to the in situ intensity (Eqn. 1). Therefore, the square root of the ARFI displacement amplitude measured with the control system bypassing the body walls is reported as a proxy for peak in situ pressure transmitted through the body walls. A linear regression between these metrics demonstrated moderate correlation (r² = 0.48).

Figure 11:

(a): Pulse inversion derived second harmonic Lag One Coherence (LOC) measured through 28 body walls (squares) and 1 control sample without a body wall (black circles) as a function of increasing transmit MI from 0.4 to 2.3. The body wall cases could be categorized into 3 groups based on LOC curve types: 1) (blue)linearly increasing then reaching a plateau (Plateau), 2) (purple) linearly increasing without reaching a plateau (Linear +), or 3) (red) low flat LOC over the entire range of MIs tested (Low Flat). For these experiments, the DAX track transmit focal depth was matched to the push beam focal depth (6 cm). (b): Sample second harmonic spatial coherence curves as function of lag for the no body wall control (black), a body wall with high spatial coherence (blue), and a body wall with low spatial coherence (red). The error bars represent the median and interquartile range of 11 overlapping transmit beams. The three colors represent the same samples between plots (a) and (b). (c): Harmonic Short Lag Spatial Coherence (HSC) (measured through the body walls) versus square root of ARFI peak displacement (measured with the control system bypassing the body walls as a proxy for in situ pressure) for the 28 body walls obtained when pushing through the body walls with an MI of 2.3 (imaging combination 2). The x-axis error bars represent the median and interquartile HSC range from 11 transmit beams. The y-axis error bars represent the median and interquartile ARFI displacement range measured from 5 repeated acquisitions.

Figure 12:

Distributions of temporal correlation values when tracking through the 28 body walls with the DAX using a track MI = 1.5. Body walls were separated into 3 LOC curve types (plateau, linearly increasing, flat). Note that red bars denote statistically significant differences.

Figure 14:

SWS yield obtained when tracking through the body walls with an MI of 1.5 plotted as a function of body wall LOC curve profile type. (blue) Pushing through body walls. (red) Pushing with the control system bypassing the body walls.

Temporal correlation between sequentially acquired beamlines is a commonly used metric to quantify additive electronic SNR (Deng et al., 2018). Figure 12 plots the distributions of temporal correlation values computed for the tracking beams of the DAX transducer when tracking through the 28 body walls with 5 repeated measures each at a track MI of 1.5. The temporal correlation values are reported as a function of body wall LOC type: Plateau (13 body walls), Linear Increase (11 body walls), and Low Flat (4 body walls). A Kruskal-Wallis test and Mann-Whitney U-tests demonstrated differences in temporal correlation values between all groups with statistical significance (p < 0.0001).

Figure 13 plots the distributions of peak displacement amplitudes generated when pushing through the body walls with an MI = 2.3 using the DAX grouped as function of body wall LOC type. Data from all 28 body walls with 5 repeated measures each are shown. A Kruskal-Wallis test and Mann-Whitney U-tests demonstrated no differences in median displacement between the Plateau and Linear Increase group, and a significant difference between these two groups and the low flat LOC group.

Figure 13:

Distributions of ARFI displacement amplitude when pushing through body walls as a function of body wall LOC curve type. 5 repeated measures for each of the 28 body walls obtained using the DAX transducer push MI of 2.3 are depicted. Red bars denote statistically significant difference.

Figure 14 plots the percentage of successful SWS measurements as a function of LOC curve profile type when tracking with the DAX through body walls with a track transmit MI of 1.5. Five repeated measures in each of the 28 body walls are depicted. The red bars were obtained with matched shear wave amplitude generated by the control transducer bypassing the body walls. For these data, differences in SWS yield measured when tracking through the body walls can be attributed directly to differences in tracking beam quality and the corresponding displacement amplitude jitter levels. The blue bars represent data obtained when both pushing and tracking through the body walls, such that the shear wave amplitude is decreased in addition to the tracking beam degradation shown with the red bars. For the group of body walls for which the LOC curve profile has reached a plateau, no difference is observed between SWS yields for the two pushing configurations where optimal tracking quality has been achieved. Likewise, for the group of body walls with flat and uniformly low LOCs, the tracking is so poor that the shear wave speed measurement yield is zero for both pushing configurations. However, for the linearly increasing LOC group, where optimal tracking quality has not yet been achieved at this tracking MI through the body wall, the increased shear wave amplitude generated when bypassing the body wall during the push is associated with an increased SWS yield of 35%.

Discussion

SWS yield increased with increasing push MI through body walls which, as expected, resulted in higher shear wave displacement amplitudes. In Figure 9b, median displacement amplitude was shown to increase with increasing push MI through body walls. Figure 9a also confirmed that median displacement amplitudes for successful SWS estimates were significantly higher than median displacement amplitudes for failed SWS estimates. Thus, in Figure 3, both SWSs tracked with and without body wall presence demonstrate monotonically increasing yield as a function of increasing push MI. When pushing through body walls, higher track beam MIs are needed to achieve the same level of SWS yield by decreasing tracking jitter magnitude in order to compensate for the lower amplitude shear waves.

Displacement estimation jitter magnitude also contributes to SWS yield, which is demonstrated when comparing tracking through body wall and without body wall. Jitter levels increased 4–4.7 times when tracking through body walls as seen in Figure 10. Furthermore, comparing Figures 4a and 4c demonstrates the larger jitter levels associated with tracking through a body wall that can obscure a low amplitude shear wave (Figure 4a) as compared to bypassing the body wall during tracking (Figure 4c). Both tracking without body wall and tracking with increasing MI are associated with decreased displacement estimation jitter levels which leads to improved SWS measurement yield.

We also conclude from the data presented that while the presence of a body wall affected both the ability to push and track, the effects on tracking were more substantial. When using imaging combination 2 instead of 1 (i.e., to track without body walls compared to through body walls), SWS yield increased an average of 37% (Figure 3). When using imaging combination 3 instead of 1 to push without body walls, however, SWS yield only increased an average of 9% (Figure 6). Specifically, when matching the push and track MI across the SWEI tracking and ARFI pushing data (right most bars in Figures 3 and 6), SWS yield decreased from 85% to 46% with the addition of body walls on track (imaging combination 2 to 1) and SWS yield decreased from 53% to 47% with the addition of body walls on push (imaging combination 3 to 1). That was despite the presence of the body wall decreasing the median ARFI displacement from 2.33 μm to 0.43 μm in Figure 9b, which was a factor of 5.4 and similar to decreases shown in previous studies [20, 21]. This change in shear wave amplitude when pushing through body wall was on the same order of magnitude as the jitter amplitude increase of 4.7 times when tracking through body wall (Figure 10). This begs the question as to why the SWS yield increases were so different between Figures 3 and 6 despite the displacement to jitter ratio having not drastically changed between only pushing through body wall and only tracking through body wall. Reverberation clutter is a potential source for the discrepancy in SWS yield between imaging combination 2 (push through and track without body wall) and 3 (push without but track through body walls). Reverberation clutter has been shown to be generated in the body wall (Pinton et al., 2009). In addition to decreasing pressure amplitude and thus signal at the focus, reverberation clutter can write into the tracking data as multiplicative noise. This leads to underestimation or even masking of particle motion.

There were three categories of body walls based on their LOC curve behaviors as function of MI. One group increased in LOC to a plateau as function of increasing MI. A second group also linearly increased LOC with MI, but did not reach a plateau for the range of MIs tested. This suggests that further increases in MI would continue to improve tracking quality for this group. A third group had low LOC throughout the entire range of MIs tested. Given that the first two groups did not have significantly different peak push displacement (Figure 13) and the fact that the second group saw increased SWS yield by 35% when pushing without a body wall while the first group did not see an increased yield (Figure 14), we can infer that the increased SWS yield in the second group was due to increased displacement to jitter amplitude ratio. Thus, we conclude that body walls with LOC curves consistent with the second group will benefit from higher track transmit MIs.

Since group one (LOC plateau) did not increase SWS yield when comparing pushing through body wall and pushing without body wall (Figure 14), we conclude that cases of group one failures should be attributed to acoustic clutter during the tracking of shear waves.

Group three had a 0% SWS yield for pushing with and without body walls (Figure 14), lower peak displacement (Figure 13), and significantly lower electronic SNR (Figure 12) than the other two groups. From these observations, we conclude that the LOC of group three is dominated by acoustic clutter. As such, this group would likely not benefit from increasing acoustic output in either the push or tracking beams.

Previously, strong positive linear correlation between HSC and in situ peak pressure had been reported when imaging through various body walls (Zhang et al., 2021). Herein, Figure 11c shows the moderate linear correlation between HSC and the square root of peak displacement through body walls. We posit that the moderate nature of the correlation observed here may be due to the fact that HSC was computed using a B-mode pulse (single cycle) while displacement was generated using a long, narrowband pulse (> 100 cycles). As such, phase aberrations of more than one wavelength could still sum coherently to generate energy for displacement for a push pulse while the same phase shifts would not contribute to the spatial coherence value we computed for our single-cycle coherence measurements. Given this observation, future work should ensure matched pulse durations when investigating the potential of HSC measurements to serve as a proxy for in situ pressure.

Conclusions

This study evaluated and quantified the impact of imaging through body walls on both ARFI push and MTL-SWEI track beams and demonstrated that the effects on the two were not equivalent. Body walls cause MTL-SWEI tracking failures even when the corresponding push beams successfully generate shear waves. For MTL-SWEI tracking beams, the presence of body walls decreased SWS yield by an average of 37% across the 5 push MIs tested (1.1–2.3). For SWEI pushing beams, the presence of body walls decreased SWS yield by only an average of 9% across the 10 track MIs tested (0.2–1.5). Additionally, in some of the body walls SWS measurement yields were found to benefit from increased track MI, suggesting the selective use of elevated acoustic output. The body walls where this occurred exhibited linearly increasing LOC trends with MI that did not reach a plateau. These findings inform the focus of future sequence optimization in SWE imaging.