To demonstrate the influence of gain setting on the calculated Virtual Organ Computer-aided AnaLysis (VOCAL™) three-dimensional (3D) indices and define a point, the sub-noise gain (SNG), at which maximum information is available without noise artifact.
Pregnant women were recruited at the time of their pregnancy-dating scan. Five identical static 3D power Doppler volumes of the placenta were acquired using identical machine settings apart from altering the power Doppler gain setting. The gain settings included the individualized SNG setting (determined by increasing gain until noise artifact was visible, then reducing it until the artifact just disappeared). The data were analyzed using VOCAL II. Vascularization index (VI), flow index (FI) and vascularization flow index (VFI) were calculated for the same sample at five different power Doppler gain levels. The relationship between the values calculated for the VOCAL indices and the gain value was explored using linear regression analysis.
Results from 50 women were analyzed. The percentage difference in VI and VFI from that observed at the SNG level in each woman was significantly linearly related to the gain setting relative to that at the SNG point (VI: r2 = 0.68, P < 0.0001; VFI: r2 = 0.72, P < 0.0001), with the values produced for VI and VFI decreasing as the gain was turned down. There was a distinct ‘turning point’ at the SNG level with linear relationships above and below, but with significantly different gradients (P ⩽ 0.001). This relationship was not demonstrated for FI.
For the last two decades researchers have been exploring the prospect of using non-invasive means for the evaluation of tissue perfusion1, 2. Power Doppler ultrasound has promised to be such a method for examination of the placenta, with its ability to summate the amplitude of multidirectional scatterers, virtual angle independence, sensitivity to low flow velocities and increased dynamic range for display3–5. As power Doppler evaluation has developed, further refinements have been introduced to standardize two-dimensional (2D) imaging and attempt to quantify flow, including the fractional moving blood volume (FMBV)6–9. Despite power Doppler ultrasound displaying amplitude of signal scattering and not velocity of particle movement, a minimum velocity of movement above a threshold is required in order for the signal to be displayed. The gain level chosen is thus able to influence the amount of power Doppler displayed on the ultrasound screen. Early investigators in power Doppler ultrasonography appreciated a need to individualize the gain to maximize the meaningful signal recorded. This was attempted by increasing the gain to the level at which obvious artifactual noise became present, then slowly reducing it to a level just below this threshold (i.e. to where the noise artifact just disappears)10, 11.
The advent of three-dimensional (3D) ultrasound machines has introduced ease and automation to the process of power Doppler imaging and has given clinicians a tool to generate 3D indices. Commonly used indices are often quoted as approximating to ‘flow’ or ‘perfusion’ in the volume sampled12. While they are very unlikely to be representing flow, their exact meaning in relationship to the underlying tissue vasculature remains debatable. This has generated a number of recent research papers, though as yet there is no clinical application for the indices9, 13. However, although most authors acknowledge that gain may have an influence on calculated 3D indices14, the exact nature of this remains unknown. There is therefore a need to evaluate this relationship in vivo to ensure that the 3D Virtual Organ Computer-aided AnaLysis (VOCAL™) indices generate the best representation of the underlying vasculature. This should help to maximize the usefulness of VOCAL for future studies.
The aim of this study was to investigate the influence of power Doppler gain on VOCAL 3D indices using each study participant as their own control to prevent the introduction of other between-patient differences (such as distance from probe to target) that are known to affect the results13.
The data for this study were collected as part of another, larger, study. The study had National Health Service Research Ethics Committee approval (REC ref: 08/H0604/163). Women over the age of 16, with a singleton pregnancy and body mass index (BMI) of < 35 kg/m2, were prospectively recruited at the time of their pregnancy-dating scan for a study involving additional antenatal ultrasound scans, after they had been provided with an information leaflet and informed consent obtained. Sociodemographic and obstetric data were collected including age, parity, family history and past medical and obstetric histories. Gestational age was calculated from crown–rump length measurement at the first scan, performed between 11 + 0 and 13 + 6 weeks' gestation15.
The scans were undertaken by a single operator (S.C.) with the participant in a semi-recumbent position using an RAB4-8-D 3D/4D curved array abdominal transducer (4–8.5-MHz) on a Voluson E8™ (GE Healthcare, Milwaukee, WI, USA). The data included in this study were taken from scans performed between 12 and 17 weeks' gestation. After confirmation of viability and identification of placental position, the optimal position for 3D acquisition of the whole placenta was identified. This was usually a cross-sectional plane close to the center of the placenta. A static 3D power Doppler volume of the placenta was acquired using a previously saved machine configuration (thus ensuring that identical machine settings were used for each patient: wall motion filter low 1 and pulse repetition frequency 0.9 kHz) with the power Doppler gain set at − 3 dB. Without moving either the participant or the probe, the power Doppler gain was quickly changed to − 5 dB and a second static 3D image captured. This technique was repeated for gain settings of − 7 dB, − 9 dB and the individualized ‘sub-noise’ gain (SNG) setting for that participant. The value of the SNG setting was set by increasing the gain until artifactual noise was present (seen as a power Doppler signal scattered throughout the whole image), then slowly reducing it to a level just below this threshold (i.e. to where the noise artifact just disappeared). The volume was captured and the SNG value was recorded. Apart from power Doppler gain, all the machine settings were kept constant for all scans. Throughout the data acquisition the participant was asked to remain as still as possible. To ensure that all five volumes captured contained exactly the same anatomy if either the probe or the participant moved, the sequence of five scans was started again. The images were then checked for ‘flash artifact’ (secondary to fetal or maternal movements). If any such artifact was detected all five scans were rejected and the full sequence of scans repeated. This was done to ensure that the same anatomical volume was captured five times, at the same time point, in the same participant, the only change being in the power Doppler gain setting. The data were then stored using Sonoview™ (GE Healthcare) and analyzed off-line on a personal computer using VOCAL II on 4D View™ (GE Healthcare).
Pregnancy data were collected by reviewing the individual patient notes postdelivery. The customized birth-weight centile was calculated for each neonate using the Grow™ software package, version 7.5.1 (West Midlands Perinatal Institute, Birmingham, UK). To avoid the introduction of any potential confounding due to pathology, only data from pregnancies resulting in a normal outcome were included. A normal pregnancy outcome was defined as: uncomplicated live birth after 37 completed weeks' gestation; customized birth weight ≥ 10th centile; no admission to the neonatal intensive care unit; and no evidence of maternal hypertensive disorders, gestational diabetes or other significant pregnancy-related maternal morbidity.
Each static volume was opened in 4D View and analyzed without any initial image manipulation. The maximum sized sphere allowed by VOCAL was selected, using 6° rotational steps centered on the reference image (Figure 1). The sphere included the whole captured placenta. The sample volume was checked to make sure that it was in the same anatomical position and identical in size for each of the five volumes. This ensured that the same vasculature was included in each sample, thereby minimizing any potential error that could be introduced by sampling different sized vessels. This meant that for each participant the only difference in the five samples analyzed was the setting of the power Doppler gain. The VOCAL histogram was then viewed and the values of the vascularization index (VI), flow index (FI) and vascularization flow index (VFI) were recorded.
The absolute values of the VOCAL indices were different for each participant. This is not only because they actually had different numbers of scatterers according to the vascularity of the target but also because of other attenuating influences such as the depth of the target away from the probe (due to differences in BMI and position of the placenta). Therefore, the values of VI, FI and VFI captured at the individualized SNG were used as each individual's internal standard. The percentage difference from this absolute value was calculated for all the other gain settings and then plotted against the absolute difference in gain value (e.g. if SNG setting is − 2, VI at the SNG is 100 and VI at a gain of − 5 is 75, the percentage difference is − 25% and the absolute difference in gain value is − 3).
The relationship between the percentage value and the absolute difference in the gain value from the SNG value was explored using linear regression analysis. To test the difference in the gradients between the lines above and below the SNG we proposed a null hypothesis that the gradient of the line above was equal to the gradient of the line below. It can be shown that this hypothesis takes a t-distribution and therefore it was appropriate to analyze it with Student's t-test. The Gauss–Markov theorem states that linear regression is the best unbiased estimator among the class of linear estimators and this is therefore an appropriate test regardless of whether the underlying distribution is normal or not. Statistical analysis was performed using SPSS version 16.0 (SPSS Inc., Chicago, IL, USA) & Excel 2004 (Microsoft Corp., Washington, USA).
Sixty-six women with a singleton pregnancy were recruited at between 11 + 0 and 13 + 6 weeks' gestation. Fourteen of these were not included; one was lost to follow up, seven neonates had a birth weight < 10th centile (including one case of pre-eclampsia), four had pregnancy-induced hypertension or pre-eclampsia and two women delivered before 37 weeks. Fifty-two women were considered to have a normal pregnancy outcome, but two were found to have at least one suboptimal image in their set of five and were therefore excluded from the study. Thus data from 50 women were included in the final analysis (Table 1). The capture of all five volumes at the first attempt was successful in the majority of women, with four requiring a second attempt.
Table 1. Patient demographics of the 50 women in the study
Linear regression analysis showed that the percentage difference in VI, and consequently VFI, was significantly linearly related to the difference in gain setting from the SNG value (Table 2, Figure 2a and b). While the relationships above and below the SNG point were both linear, their gradients were significantly different (P < 0.001). Above the SNG point the slope was steeper as the noise artifact appeared. No such relationship was demonstrated for FI when the gain was changed (Table 2, Figure 2c).
Table 2. Results of simple linear regression analysis comparing difference in vascularization index, flow index and vascularization flow index with difference in gain setting below and above the sub-noise gain (SNG) value
The values for the SNG setting were weakly related to the participant's BMI at booking (r2 = 0.21, P = 0.001; Figure 3).
The exact derivation of the VOCAL indices from the raw signal is proprietary information and not available to users of the software. Therefore the relationship between these calculated indices and the behavior of underlying scatterers, and hence tissue perfusion, is unfortunately unknown. We have confirmed, in vivo, previous experimental findings that the power Doppler gain setting directly affects the VOCAL indices16. However the phantom experiment did not demonstrate a ‘turning point’ where the noise artifact appears. This artifact is probably a phenomenon specific to biological systems where there are many vessels at different depths, angles and velocities. However, movement will also produce a similar artifact and care must be taken not to confuse this with the static noise from a high power Doppler gain setting. Given this uncertainty and the differences in machine type, probe and settings between the single-vessel16 and flow-free phantom14 studies and our in-vivo work, it is difficult to draw strong associations between results. In particular, the positive FI trend shown by Raine-Fenning et al.16 appears to occur over a much higher range of gain level (24.8 to − 44.0 dB) compared to those PD gains demonstrated to be appropriate for the participants in our study (−7.8 to − 0.8 dB). From histogram analysis of volumes it appears that as the power Doppler gain is increased there is a uniform lifting of the scalar color values. Histograms also showed a larger number of power Doppler values discarded at the top end of the scalar range in comparison to the other values in the range. This may be explained by high power Doppler values being truncated at this point. If this hypothesis is correct, it would explain the effect on the VI, which is calculated from the absolute color values17. As the FI is a ratio of the weighted values and the absolute color values17 it would be relatively unaffected by the uniform increase in the lower values. This explanation could be verified if access to the raw radio-frequency signal was provided.
The linear relationship of the power Doppler gain to the VI value indicates that what is essentially the ‘volume control’ (gain in decibels) directly affects the calculated value of the VI. As the gain setting is dropped below the SNG level, the value calculated for VI decreases proportionately as Doppler information is lost, but the displayed image on the screen does not visibly change, leaving the operator unaware of this. Above the SNG point a significantly different linear relationship is seen, with a steeper slope, as the image rapidly becomes disrupted by noise artifact.
Our study suggests that use of the SNG might be a step towards improving the information obtained for each subject, thereby improving comparisons between patients, as it ensures that the maximum information available for one subject is being compared with the maximum information available for another. While selecting the gain setting just below the noise point is inherently subjective we believe it is a simple, ‘user-friendly’ method. However, further studies will be required to confirm its reproducibility.
The previously documented suggestion that gain should be set ‘somewhere in the middle of the range’14 and kept constant for all patients may actually result in information being lost for some patients, while artifact is introduced in others. Our protocol was developed to collect 3D volumes with gain settings between − 3 and − 9 dB. These values were arbitrary, and selected because we expected that this was the most likely range for the population considered in our clinical research. In fact, the SNG value ranged between − 0.8 and − 7.8 dB (mean − 3.9 (SD 1.9) dB) in our 50 participants. Had − 6 dB been selected as the mid-point gain we would have selected the SNG and calculated the optimal information for only two women. The indices for 10 women would have been recorded above the SNG, thereby overestimating the value of VI and VFI, and in the remaining 38 participants the values would have been underestimated. This means that 96% of these VOCAL vascularization measurements would either have been sub-optimal or have contained noise artifact. Rather than suggesting abandoning the use of VOCAL indices, we recommend optimizing the information provided by using the SNG technique.
This study has several limitations. It was relatively small, but all statistical tests demonstrated a very high level of significance (P < 0.001), indicating that 50 was a sufficient number of participants. It may have been better to evaluate more gains (possibly including 0 and + 3). However, taking a sequence of more than five volumes in the same place on the abdomen without the operator or participant moving would be difficult. Therefore, pragmatically, we decided on a series of five.
Several studies have demonstrated that power Doppler is subject to attenuation, rendering it dependent on factors including the depth of intervening tissue18, 19. We showed a weak relationship between the SNG level and the participant's BMI at booking, presumably reflecting the need for gain increase to compensate for signal attenuation, as previously demonstrated in phantom models19. BMI is known to be an imperfect measure of degree and distribution of adiposity and hence intervening attenuating tissue20, explaining the weak relationship between BMI and SNG. Other factors such as scar tissue may also be responsible for signal attenuation.
In summary, we have presented a new acquisition step aimed towards optimizing 3D power Doppler ultrasound for clinical use in placental imaging, using a simple technique that underpinned early 2D power Doppler imaging21. The ultimate goal of power Doppler vascular imaging is the development of truly internally standardized 3D FMBV indices. If the reliability of this technique can be proven, adoption of the SNG to maximize power Doppler imaging potential may prove a useful step towards this goal.
We thank Dr D. Reynard for his statistical advice and the staff of the ultrasound department at the Women's Centre, John Radcliffe Hospital, especially Mrs Katie Blissett, for all their help in recruiting the study participants. Sally Collins and Gordon Stevenson are both funded by NIHR Biomedical Research Centre program.