Three‐dimensional power Doppler: validity and reliability

In an Opinion in the September 2008 issue of this Journal, Alcazar stated that ‘more than 100 papers have been published analyzing the role of 3D power Doppler ultrasound in almost all areas of obstetrics and gynecology and, despite this abundance of literature on the application of 3D power Doppler ultrasound, it seems that so far few have stopped to ask what we are measuring’1. He was alluding to our 10-year experience of 3D power Doppler indices and commented on three articles published in that issue which, using in-vitro models, evaluated limitations of 3D power Doppler and its relationship with true flow2-4. Two of these three studies were conducted by the same group and used complex experimental models: a computer-controlled gear pump connected to different test tanks which were evaluated by 3D ultrasonography. Each test tank was designed for a specific purpose, but all contained tubes perfused with blood-mimicking fluid. Using this model, the authors evaluated the effect of flow rate, vessel number, attenuation, concentration of particles in the fluid3 and machine settings2 on the 3D power Doppler indices: vascularization index (VI), flow index (FI) and vascularization flow index (VFI). The other study evaluated two simpler models: a flow-free model using a tissue-equivalent phantom and a flow model, using a syringe infusion pump connected to a polyurethane tube that passed through a test tank, also evaluated by 3D ultrasonography. Using the models, the authors also evaluated the effect of certain machine settings (e.g. gain, pulse repetition frequency (PRF), wall motion filter) and fluid velocity on VI, FI and VFI. The three articles demonstrated that 3D power Doppler indices are highly dependent on both flow and machine settings as well as on the concentration of particles in the fluid and on attenuation. Their publication probably raised awareness and confidence in this technology: over 100 papers involving 3D power Doppler in obstetrics or gynecology have been published in just the last 2 years. After reading these three studies I tried to reproduce the model described by Schulten-Wijman et al.4. However, I stopped when I observed an intense attenuation caused by the tube wall, which clearly reduced the detection of flow by Doppler ultrasonography. This limitation of inorganic models has doubtless been observed by others and, aiming to better correlate 3D power Doppler indices with true perfusion, researchers went further. Jones et al.5 created an ingenious experimental model consisting of a human placental lobule exposed to dual perfusion (from both maternal and fetal sides) with a solution containing human erythrocytes. Using the model, the authors evaluated the influence of flow and erythrocyte concentration (0.5% and 5.0%) in a solution running within actual human vessels. The results of this work confirmed previous findings3 and demonstrated a strong association between flow and the 3D power Doppler indices (all R2 values exceeded 0.95, indicating that more than 95% of the variation observed in the indices could be attributed to variation in flow rate), a dependence of 3D power Doppler indices on erythrocyte concentration, and, by using agar-based tissue-mimicking material of different thicknesses (1–7 cm), that attenuation reduces all of these 3D power Doppler indices. In this issue of the Journal, we learn that researchers have gone still further in the quest to evaluate placental perfusion by 3D power Doppler and have developed an in-vivo model based on the pregnant sheep6. With the exception of the common uterine artery supplying the pregnant horn, which was dissected to apply occlusion and measure flow, all arterial supplies to the uterus were ligated and the 3D power Doppler indices within placentomes were evaluated. Different flow rates in the dissected uterine artery were applied in seven pregnant sheep. Ten datasets per sheep were obtained but 22 were excluded due to artifactual images, resulting in only 48 analyzed volumes. The authors observed a significant positive correlation between the 3D power Doppler indices and blood flow. The authors were very careful to consider the limitations of 3D power Doppler, especially the effect of attenuation, and they standardized their technique to evaluate only those placentomes situated 3–5 cm from the probe. The effect of attenuation can also be accounted for by normalizing the Doppler signal7 using ‘fractional moving blood volume’ (FMBV)8-10. Some believe that without this normalization 3D power Doppler indices should only be obtained using transvaginal gynecological ultrasonography, when the transducer is in close proximity to the organ under investigation11, 12. However, I would argue that standardizing power Doppler evaluation is more important than is the normalization performed during FMBV, because the latter is not able to correct for all the consequences of attenuation or machine settings. The measures performed during FMBV evaluation are dependent on both presence/absence of Doppler signals and intensity (similar to VFI in 3D power Doppler7) but only the information related to Doppler signal intensity is normalized. However, previous studies clearly demonstrated that attenuation and machine settings interfere not only with signal intensity (represented in 3D power Doppler by FI), but also with the presence/absence of Doppler signals (represented in 3D by VI)2, 3, 5. Therefore, even though FMBV has potential, at the moment this evaluation seems limited, especially if we consider that until now it has been restricted to use with 2D ultrasound, which reflects tissue/organ perfusion as a whole far less well than does 3D ultrasound. When describing machine settings, Morel et al. were, however, not so careful. They report 15 different settings: ‘CRI (compound resolution imaging), 1; persistence, 3; SRI (speckle reduction imaging), 4; dynamic contrast, 6; harmonic, low; quality, normal; wall motion filter, mid 2; PRF, 1.3 kHz; frequency, low; line density, 8; balance, 225; filter, 3; flow resolution set, mid 2; filter line, 3; enhance, 4’. However, some of these settings are related to B-mode while others to Doppler and this mix is potentially misleading: one might wonder, for example, what the relationship is between CRI, SRI or harmonics and 3D power Doppler indices, when actually there is none14. Such confusion should be avoided, and adjustments should be described separately, otherwise there will be even more studies evaluating the effect of one B-mode adjustment (e.g. harmonics) on 3D power Doppler indices15. Morel et al.'s choice to set the PRF at 1.3 kHz instead of at a lower value of, say, 0.6 kHz, may also be questioned: using a higher PRF may make it impossible to detect low-velocity flow4, leading to the observation of more artifacts13. Indeed, artifactual images led to the high rate (22/70, 31.4%) of discarded 3D datasets. Moreover, they did not describe one of the most important settings: color gain. By using high color gain, positive values for 3D power Doppler indices may be observed even in the complete absence of flow13. The main limitation of this work is, however, the authors' erroneous assertion that: ‘Because all other uterine arterial supplies were ligated, the blood flow measurements obtained from the common uterine artery of the pregnant horn represented the only blood supply at the time of the experiment16. As a consequence, the quantification of the Doppler signal within the placentomes can be considered as representative of the true blood flow within the VOI.’ This would be true if they had evaluated the uterus of non-pregnant sheep, as occurred in the study they cited16. However, in pregnant sheep both maternal and fetal circulations contribute to placentome blood flow. Therefore, the blood flow within the placentome would only be truly representative at the time of the experiment if the fetus was dead. One must also bear in mind that perfusion of a placentome in one sheep could be different from that in another sheep, even if uterine artery blood flow is the same, because of differences in both maternal and fetal vascularization. Additionally, as the authors state6, even in the same sheep, perfusion could differ from one placentome to another. What is interesting is that, even considering these sources of possible bias, the authors observed a strong correlation (Pearson r > 0.75) between blood flow in the uterine artery and two of the three 3D power Doppler indices (VI and VFI). Morel et al.6 also evaluated the intra- and interobserver reliability of VI, FI and VFI by calculating intraclass correlation coefficients (ICCs). They observed very high intraobserver ICCs (> 0.93) when assessing the placentome using 15° or 9° of rotation with VOCAL. The interobserver reliability was also high (0.80–0.92), but it was assessed using only 30° rotation, which resulted in lower intraobserver reliability (0.72–0.91). The reproducibility of 3D power Doppler indices from placental samples has been the focus of much research17-20. In one of these studies17, high reliability was observed for both intraobserver (ICCs of 0.92, 0.99 and 0.90 for VI, FI and VFI, respectively) and interobserver (ICCs of 0.75, 0.98 and 0.86, respectively) evaluation. These results are comparable to those of Morel et al.6 even though different techniques were applied: automatic sphere volume when evaluating a sample of human placenta containing one large vessel17 and manual mode when delimiting a sheep placentome6. However, an important limitation of both reliability studies6, 17 is that the different analyses were performed using a single 3D dataset per subject, which is not truly reflective of clinical reality. Evaluating a single dataset has the potential to produce perfect agreement, as only information on stable voxels is analyzed. Serial acquisition of 3D data cannot produce identical datasets, as each will be influenced by natural and unavoidable biological variation as well as by physical movements. Lai et al.18, in their reliability study, considered volume acquisition as well as analysis. They observed very low ICCs and concluded that ‘poor reproducibility is present at the most fundamental level’. However, they produced some surprising results. Even when only reproducibility of analysis was evaluated, using a single dataset for both intra- and interobserver evaluation, they observed very low reliability, especially for the intraobserver evaluation: the intraobserver ICCs were 0.24–0.57 for VI, 0.33–0.78 for FI and 0.12–0.48 for VFI, while the interobserver ICCs were 0.38–0.92 for VI, 0.40–0.85 for FI and 0.10–0.92 for VFI18. When asked about this unexpected result20, the authors revealed a methodological flaw: they apparently ‘store a single dataset from a section of placenta containing an isolated large vessel, and then take two sample volumes from within this dataset, one containing the large vessel, and the other not’. In fact, the observers should have aimed to measure exactly the same thing while blinded to their previous or other measurements. What they did in this study was practically the same as evaluating the reproducibility of assessing a person's height using a measuring tape, once with the person standing and then with them sitting, to end up concluding that the measuring tape is not reliable. Regardless of the potential sources of error in their experimental set-up, the take-home message from Morel et al.'s study is that 3D power Doppler indices from placental samples are related to uterine artery flow6. This has implications for the screening of pregnancy-related diseases such as pre-eclampsia and intrauterine growth restriction. At least theoretically, placental perfusion is associated with such problems, probably more strongly than is uterine artery impedance, which is currently the most frequently used ultrasonographic parameter to predict these adverse outcomes. Of great interest will be the evaluation of 3D perfusion indices in the late first trimester of pregnancy (between 11 and 14 weeks), when the entire placenta can be assessed for both volume and vascularity. Although uterine artery pulsatility index sensitivity in the diagnosis of pre-eclampsia has a better performance between 22–24 weeks21, 22 than it has in the late first trimester23-27, the latter seems to be more relevant, since prevention strategies should be started as soon as possible; acetylsalicylic acid, for example, has recently been found to be beneficial in reducing pre-eclampsia only when initiated before the 16th week of gestation28. Fortunately, recently published studies have shown 3D power Doppler indices to be reproducible in the late first trimester29, 30 and, even better, have associated volume and 3D power Doppler indices of the placenta with pre-eclampsia and intrauterine growth restriction31, 32. One of these articles demonstrated that VI obtained from the placenta or adjacent myometrium at 11–14 weeks performed better at predicting the risk of pre-eclampsia than did mean uterine artery pulsatility index or PAPP-A at the same time32; a new perspective in predicting and preventing pre-eclampsia may be emerging. However, to achieve successfully such research goals, it is imperative that 3D power Doppler indices are standardized and I would like to reiterate Alcazar's appeal to the International Society of Ultrasound in Obstetrics and Gynecology (ISUOG) that it should formulate a proposal for standardized settings to be used worldwide1. This step is essential not only for settings but also, considering the limitations demonstrated by previous studies2-4, 13, for the evaluation process itself (e.g. distance of the probe from the region of interest, avoidance of assessment of cystic areas). However, I feel that, unfortunately, describing a standard will not be enough. There are so many available settings that it would be very difficult (and unlikely) for someone to properly adjust them all. For example, in the Voluson 730 Expert (GE Medical Systems, Zipf, Austria) there are 13 settings which could possibly impact on the 3D power Doppler indices: power, gain control, quality, wall motion filter, PRF, frequency, flow resolution, balance, smoothing, ensemble, line density, artifact suppression and line filter14, resulting in an infinite number of possible combinations. Moreover, there are two different manufactures (GE Medical Systems and Medison Co., Seoul, South Korea) that produce several different ultrasound machines able to evaluate 3D power Doppler indices, and adjustments between them vary. For proper standardization, ‘ISUOG settings’ would have to be developed in collaboration with the manufacturers and incorporated into machines so that they are available to the operator at the click of a button. Only then will researchers be able to create for the 3D power Doppler indices nomograms of different organs and tissues that can be used universally to test if diseases or conditions are related to impaired 3D power Doppler indices. Without this, it may be another 10 years before the advances in 3D power Doppler that are made in the laboratory today can be introduced into the clinical field.