| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
|
by the American Institute of Ultrasound in Medicine J Ultrasound Med 23:1629-1639 • 0278-4297 Contrast-Enhanced Sonography Helps in Discrimination of Benign From Malignant Adnexal MassesDepartment of Gynecology, Obstetrics, Fetal Medicine, and Human Reproduction (H.M., S.S., G.B.) and Department of Medical Imaging and Ultrasound (F.T.), Bretonneau University Hospital, Tours, France; Institut National de la Santé et de la Recherche Médicale, Centre de Recherche Clinique 202, Faculty of Medicine, Tours, France (B.G.); Departments of Obstetrics and Gynecology and Biomedical Engineering, Division of Gynecologic Oncology, University of Arizona, Tucson, Arizona USA (M.B.); and School of Public Health, Arizona Cancer Center, Tucson, Arizona USA (J.R.-M.). Address correspondence and reprint requests to Henri Marret, MD, Department of Gynecology and Obstetrics, Bretonneau University Hospital, 2 Boulevard Tonnellé, 37044 Tours Cedex, France. E-mail: marret{at}med.univ-tours.fr.
Objective. To investigate the potential efficacy of real-time contrast-enhanced power Doppler sonography in the differentiation of benign and malignant adnexal masses in a pilot study. Methods. Before surgical treatment, adnexal masses were prospectively evaluated with power Doppler sonography before and after injection of a contrast agent. Real-time postinjection sequences were computerized with time-intensity analysis software to determine an enhancement curve and contrast parameters. The intraobserver and interobserver reproducibilities of these criteria were assessed on a subsample. These contrast parameters were compared between benign and malignant tumors using logistic regression. Sensitivity and specificity were used to compare contrast parameters with sonographic and Doppler variables. Results. Ninety-nine women were included, for a total of 101 adnexal masses. There were 23 cases of ovarian malignancies and 78 benign adnexal lesions. Our procedure had excellent intraobserver and interobserver reproducibility, with an average intraclass correlation coefficient of 0.92. The time before enhancement and intensity ratio did not reliably differentiate between the benign and malignant masses. Washout times and areas under the curves were significantly greater in ovarian malignancies than in other benign tumors (P < .001), leading to sensitivity estimates between 96% and 100% and specificity estimates between 83 and 98%. Contrast parameters had slightly higher sensitivity and slightly lower specificity when compared with transvaginal sonographic variables of the resistive index and serum cancer antigen 125 levels. Conclusions. Contrast-enhanced power Doppler imaging may easily and precisely discriminate benign from malignant adnexal lesions. Larger studies are needed to determine the appropriate use and benefits of this new procedure.
Key Words: adnexal mass • contrast agents • contrast enhancement • ovarian cancer • power Doppler sonography Abbreviations: AUC, area under the curve • CA 125, cancer antigen 125 • CI, confidence interval • OR, odds ratio • RI, resistive index • ROI, region of interest
Color Doppler sonography has been assessed as one of the sonographic techniques that can be used to describe specific characteristics of ovarian vascularization.1–6 Power Doppler sonography is useful to map ovarian vessels, including those associated with malignancy, whereas pulsed Doppler sonography is used to measure blood flow velocity.7–9 Doppler imaging and the subjective evaluation of the gray scale image improve our ability to make a correct diagnosis before surgery,10,11 which will enhance patient outcome if a malignancy is present.12–15 Recently, quantification of the vascularization of a tumor with power Doppler sonography has been done by analysis of images using special software,16–18 which allows us to calculate the number of color pixels on a digitized image. Until now, subjective evaluation of gray scale and Doppler sonography in experienced hands has been considered the most powerful method for discriminating between benign and malignant adnexal masses.11 The vascular index of a tumor can now be calculated and compared with other discriminating parameters. Our previous study showed that the power Doppler index (color pixel number inside the tumor/total pixel of the tumor) is a simple and accurate parameter for discriminating an ovarian malignancy from a benign ovarian mass and from others (H. Marret, S. Sauget, B. Giraudeau, G. Body, and F. Tranquart, unpublished data, 2004). Similarly, the use of intravenous contrast agents in sonography is being investigated in a multitude of potential clinical applications (liver, kidney, breast, and prostate) to identify aberrant vascular changes associated with malignancy. Intravascular sonographic contrast agents enhance depiction of tumor vessels by providing a stronger Doppler signal.19,20 Contrast agents such as Levovist (SH U 508A; Schering AG, Berlin, Germany) may improve the diagnostic ability of sonography to identify early microvascular changes that are known to be associated with early stage ovarian cancer, but only a few small studies21–25 have been published using this contrast agent for gynecologic purposes. We conducted a prospective pilot study for the following reasons: (1) to evaluate the contribution of contrast-enhanced color power Doppler sonography and its own objective parameters for the diagnosis of malignancy in ovarian tumors and (2) to compare these parameters with other variables that have been evaluated to differentiate malignant adnexal masses from benign ones.7
Design This study was planned as a prospective pilot study between February 2002 and March 2003 and was approved by the Ethics Committee at Bretonneau University Hospital. Informed consent was obtained from all patients. Any woman who was electively admitted to our gynecologic department for a known adnexal mass was recruited into the study. Tumors were evaluated by transvaginal gray scale and power Doppler sonography. All the women were treated by cystectomy or oophorectomy using surgical laparoscopy or laparotomy. All ovarian masses were examined by a single pathologist.
Scanning Procedures The entire mass was evaluated with power Doppler sonography. Power Doppler signals were detected in all masses. Identical power Doppler settings (ie, sensitivity, depth, gain, and filter) were used for all women, with a pulse repetition frequency of 500 Hz, which was selected for maximal sensitivity with minimal background noise. With pulsed Doppler sonography, a flow velocity waveform was obtained, and the resistive index (RI) was calculated for detected vessels. If more than 1 artery was detected within a given mass, the lowest RI value was used to characterize the tumor. On the basis of our previous report,11 an RI of 0.53 or lower was considered suggestive of malignancy.
Contrast Agent and Plan Selection The area to be imaged was identified by the sonographer before the injection according to 3 rules: (1) the whole tumor should be seen on the image, if possible; (2) the sectional plane should contain the solid part (wall, septa, and papillae) of the tumor; and (3) the most vascularized area should be selected by power Doppler sonography. Then a dose of 2.5 g of Levovist, at a concentration of 300 mg/mL, was injected as a rapid intravenous bolus through an antebrachial vein, followed by a 5-mL flush of saline. The same sonographic orientation was kept during the entire measurement. The sonography machine was programmed to store 5-minute clips for a total of 180 images.
Analysis of the Contrast Enhancement
The time-intensity curves were then derived in each patient for each mass. The software calculates the total power Doppler intensity, which is determined by the number and intensity of color pixels inside the ROI for each image. After the curve was drawn (Figure 3  ), baseline and peak intensities were noted, and the ratio of intensity enhancement (expressed as a percentage of enhancement) was calculated with the following formula: peak intensity – baseline intensity/baseline intensity; then the curve was smoothed and normalized. Values were compared between patients. Among all parameters, the time-intensity curves were analyzed for the following indices, defined in the legend to Figure 3: the uptake time (in seconds), the washout time, the half-intensity washout time, and the area under the curve (AUC), calculated by the means described for Figure 4.
To measure the reproducibility of the ROI, intraobserver and interobserver reproducibility were assessed in a random subsample of 15 images by 2 different operators (S.S. and H.M.). For the interobserver reproducibility study, the entire 5-minute clip was viewed by 2 different examiners. They selected the most representative area of solid tissue containing the largest number of vessels with the fewest artifacts, and they drew the ROI. Then the software constructed the curve, allowing the calculation of the parameters. Both operators evaluated 1 measure per observer and the mean of 3 measurements per observer. For intraobserver reproducibility, each analyst independently evaluated the same clip on 3 occasions, redrawing the ROI each time.
Tumor Marker Examination
Statistical Analysis
Patients and Tumors Ninety-nine patients were included, for a total of 101 tumors. None of the women had a benign tumor on one side and a malignant tumor on the other. The ages of the patients ranged from 19 to 72 years (mean, 46.2 years). Forty-one patients were postmenopausal. The interval between scanning and surgery was usually 24 hours and never exceeded 1 week. All the tumors were clearly enhanced after Levovist injection. Twenty-three adnexal masses (23%) proved to be primary epithelial malignant ovarian tumors, including 2 borderline, 1 stage I, 3 stage II, 12 stage III, and 5 stage IV, whereas 78 (77%) were benign. Only 6 (10%) of 60 tumors in premenopausal women were malignant, but 17 (41%) of 41 tumors were malignant in the postmenopausal group (P < .001). The median largest tumor diameter was 58 mm (range, 28–170 mm), corresponding to a median tumor volume of 124 cm3 (range, 8–1511 cm3).
Sonographic Parameters
Contrast Parameters After Levovist injection, on visual evaluation, the intensity of the power Doppler signals and the number of vessels subjectively increased in all tumors. The contrast agent was well tolerated with no side effects. A time-intensity curve with a clear uptake time, rapid peak, and decrease with return to the baseline level was seen in all scanned ovaries.
Interobserver and intraobserver reproducibility results are presented in Tables 3
On univariate analysis, the time-intensity curves for the malignant and benign tumors were different (Figure 5 , A and B). The baseline, the maximal peak intensity levels, and the ratio of enhancement tended to be higher in malignant tumors but did not reach statistical significance (ratio, 5 ± 2.7 versus 4 ± 2.2; P = .064). The uptake time did not differ between diagnostic categories. The washout time was longer in the malignant than in the benign tumors. Malignant tumors had a different curve with a slow decrease of enhancement even after normalization of the curve. The form of the curve approaches a linear slope. Both the mean washout and half-washout times were shorter in the benign tumors, and the washout phase could be divided into 2 phases with a fast initial decrease followed by a second slower decrease to baseline. The AUC and washout time were the 2 best parameters with highest sensitivity: only 1 cancer was missed with the use of these 2 parameters with the cutoff points (derived from the ROC curve) fixed at 88 seconds–1 and 175 seconds, respectively. Specificity was also very high at 95% and 98%. (Table 5)
We attempted to construct a single model including all the contrast, Doppler, and gray scale variables in the same logistic regression model. However, parameters for most of the independent variables could not be estimated by maximal likelihood because of perfect prediction of most of the cancer and noncancer cases. Inspection of the data indicated that a combination of identical covariate patterns and multicolinearity in the independent variables was responsible for preventing complete model estimation. Instead, it was necessary to assess the models in parallel and to choose the model that performed best. Correlation coefficients showed that the contrast agent parameters were highly correlated; thus, multivariate regression including the gray scale, Doppler, and contrast parameters showed multicolinearity between variables.26 Unfortunately, the inability to estimate a full model is a characteristic of the data set and is thus not necessarily correctable through changes in design. As a result, the best approach seems to be a practical one, that is, identifying the model that results in the best classification performance. With the use of the least correlated variable, washout time, logistic regression showed that it was highly predictive of malignancy (OR, 1.074; 95% CI, 1.038–1.111; P < .0001). When comparing the predictability of the 2 models (CA 125 and RI versus washout time) by logistic regression, it appeared that washout time had slightly better sensitivity of 96% and slightly lower specificity of 98%, a positive predictive value of 96%, and a negative predictive value of 98%, with a correct classification of 98% of the cancers.
We present the ROC curves for these 2 models (Figures 6
The first borderline tumor showed a 4-cm cystic mass with a thick wall but no septum or papillary projection. The RI was 0.54; the CA 125 level was 18 IU/mL; and the contrast parameters in favor of malignancy were washout time of 193 seconds, half-washout time of 104 seconds, and AUC of 92 seconds–1. The second borderline tumor had an 8-cm cystic mass with a thin wall and thin septa without papillary projections. The RI was 0.52; the CA 125 level was 43 IU/mL; but the contrast parameters favored malignancy, with a washout time of 222 seconds, half-washout time of 113 seconds, and AUC of 113 seconds–1. In both cases, the contrast parameters were the criteria most suggestive of malignancy. Three of the missed cancers with RI and CA 125 data were borderline tumors, and 1 was stage I.
Our results suggest that kinetic parameters derived from power Doppler sonography enhanced with Levovist provide new valid criteria, which appear accurate in the discrimination of benign pelvic masses from malignant primary epithelial ovarian tumors and could improve the preoperative diagnosis of ovarian cancer. Total washout time of the contrast agent was the parameter that performed the best, with sensitivity of 96% and specificity of 98%. One cancer was missed, and 1 benign mass was diagnosed as cancer. Using the traditional criteria of RI and CA 125 level, there was sensitivity of 86%, with 4 missed cancers and no false-positive results (specificity, 100%). The risk of missing a cancer must therefore be balanced with the cost of false-positive cases: given the high mortality associated with ovarian cancer, the numbers of missed cancers will be more costly than the false-positive cases. There have been many publications concluding that guidelines are needed for the referral of women with a high likelihood of having ovarian cancer to physicians trained in gynecologic oncology because a woman with ovarian cancer has improved survival if she undergoes primary surgery by a gynecologic oncologist.12–15 In 2002, the American College of Obstetricians and Gynecologists released guidelines for referral of women with a pelvic mass to a gynecologic oncologist, which included women with a pelvic mass suggestive of malignancy. The most common imaging modality for evaluating the ovaries is transvaginal sonography, and abnormal sonographic findings are the basis for most referrals for presumed ovarian cancer. However, even in experienced hands and particularly for early stage ovarian cancer, gray scale sonography has good but insufficient ability to discriminate malignant from benign neoplasms, which has stimulated the development of Doppler imaging6,7,11 and contrast agents to enhance visualization of the vasculature in the ovary. Increased vascularization is one of the hallmarks of neoplasia6,9,11 and should be visualized better with an agent that targets the vasculature. The 2 main types of false-positive results on 2-dimensional gray scale sonography have been dermoid cysts and cystadenofibromas.10,27 The differentiation of solid benign tumors from malignant tumors is problematic, although the prevalence of cystadenofibromas and other solid ovarian tumors (fibromas and thecomas) is low (<7% in one of the largest studies published).27 These benign tumors, which have a large number of vessels that are both peripheral and central and have moderate velocities and resistance (0.5–0.6), can still be misclassified on the basis of both gray scale and Doppler sonography. In a previous study,11 we showed that papillary projections and tumor density (solid or cystic) are the 2 best morphologic parameters. However, this study suggests that RI and CA 125 associations can perform better (no false-positive results) than sonographic morphologic parameters and the presence of papillations. Logistic regression models are limited when significant multicolinearity exists, as in this study. However, our goal was to test new parameters, and the use of contrast agents has been hypothesized to reduce the false-positive and -negative rates of sonography in the differentiation of benign from malignant neoplasms. There is limited literature evaluating contrast enhancement in adnexal masses. To our knowledge, only 4 studies have been published.21–25 The use of contrast agents was first described in 1994 by Suren et al.21 A second study24 using 3-dimensional sonography in 45 women reported an improvement in the diagnosis of malignancy by increasing the visualization of the vascular structure, which was more pronounced in malignant tumors. A third study22 (n = 58) showed an increase in intensity before and after the contrast agent injection when 2 digital images were compared. A fourth study23 (n = 32) analyzed the kinetics of Levovist in 10 images and showed differences in the washout periods of benign and malignant tumors. The most recent study (n = 70), published in 2003 by Orden et al,25 showed the usefulness of Levovist contrast-enhanced power Doppler sonography in real time. It described several parameters derived from a time-intensity curve in a 5-minute video clip. That study provides a comparison for our results. The authors showed that baseline and maximal Doppler intensities were significantly higher in malignant tumors when compared with benign tumors. The time to initial imaging of the contrast agent was shorter (17.5 versus 22.5 seconds; P = .005), and the washout period was longer (190.4 versus 103.6 seconds; P < .001) in malignant tumors than in benign tumors. Different areas under the time-intensity curves were also significantly greater in malignant tumors. In this study, results obtained with washout time and AUC were similar in terms of sensitivity and specificity, but intraobserver and interobserver reproducibility parameters defined the AUC as the most reproducible parameter and washout time as the less correlated variable, which is in full agreement with the findings of Orden et al.25 In contrast, we found that arrival time, as reported by Szymanski et al,23 did not help differentiate benign from malignant tumors. Levovist achieves a maximal intensity peak quickly and has a biphasic washout phase with a decrease in intensity down to the baseline level. This type of time-intensity curve was shown in dog cerebral vessels16 and in human ovaries by Orden et al.25 On visual evaluation, we did not find the same kinetics. In our present study, a biphasic washout was present in benign cysts but not in malignant tumors, where the washout phase was more monophasic and constant. In the reproducibility study by Orden et al,25 the coefficient of variation varied from 14 to 51 for the analysis of these 2 areas because the precise definition of the 2 different parts of the washout phase was difficult to assess; total AUC and half-intensity washout time are easier to measure and thus were selected in our study. Moreover, we can suggest the use of the mean of 3 measures that improve reproducibility. Use of a contrast agent allows a more accurate map of the vascular anatomy by enhancement of the signal strength from power Doppler sonography, which increases the number of large vessels and allows recruitment of small vessels. In solid tumors, this enhancement has shown a larger number of vessels and color pixel densities in cancers than in benign lesions.28–31 In the analysis of these new contrast-associated dynamic parameters as well as those described previously,11 all the parameters were highly predictive of malignancy on univariate analysis, although their multicolinearity and quasicomplete separation problem limited our regression analysis. However, the addition of washout time slightly improved the sensitivity of sonography, thus reducing the false-negative rate and improving the diagnosis of cancer. Most physicians are afraid of missing a cancer and would classify a tumor as malignant if there were the slightest suspicion of malignancy. Thus, improving specificity without sacrificing sensitivity is one of the main goals of this technology. Our 98% specificity was very high, with only 1 tumor misclassified as cancer. A limitation of our study could be the calculation of the intensity of color pixels. Our software calculates the total intensity, which is determined by the number and intensity of color pixels for each image of the power Doppler signals inside the ROI. Calculating the intensity of color pixels is subject to many variables, including resolution, frequency, penetration, depth, maternal body habitus, and preprocessing settings. Even if some of them can be standardized, maternal body habitus and the depth of an adnexal mass cannot. Therefore, the amount of pixels displayed is dependent on the resolution achieved in the particular patient and at the particular depth at which the adnexal mass is located. In other words, the intensity of the color pixels will be higher in thin patients with a mass close to the near field of the transducer (good resolution) and lower in heavy patients or in masses in the far field of the transducer. With improvements in transducers and post processing, these limitations will be reduced, with a concomitant increase in sensitivity. There is an advantage to the current study, however: we studied and measured the kinetic parameters rather than just the total value of the color pixels. The kinetic variables are not largely dependent on improved technology. Moreover, our cutoff values for washout time and AUC were not tested prospectively and will need further validation. The contrast enhancement was evaluated in 2-dimensional images in this study. Sonographic volume vasculature quantification techniques have recently been described using 3-dimensional power Doppler flow indices that are reproducible, thus allowing them to be used in clinical practice or research.32 Further studies are necessary to validate these parameters for ovarian tumor discrimination and will be the focus of these subsequent studies. Ovarian angiogenesis and neovascularization occur in patients with normal ovarian physiologic characteristics in the premenopausal period,33 so normal vascular changes must be differentiated from those seen with malignancies in sonographic evaluations. Thus, studies in women with normal cycles and no physiologic abnormalities are necessary. In addition, correlation of contrast parameters with other angiogenic factors such as microvessel density, vascular endothelial growth factor, and up-regulation of angiogenesis genes may be helpful. We can speculate that changes in the vascularization of benign-appearing lesions with abnormal vascularization may be predictive factors for ovarian cancer. Removal of these lesions would therefore be a preventive approach, but this speculation must be confirmed. Our population had advanced stage cancers that were diagnosed well on the basis of morphologic and power Doppler parameters. Our population was representative of the usual distribution of primary epithelial cancers, with most cancers being diagnosed in stages III and IV. Contrast parameters in this study missed only 1 stage II cancer and facilitated the diagnosis of the 3 low-stage tumors. Therefore, contrast enhancement may be very useful for discriminating low-stage cancers from benign pelvic masses. This could be a great advantage because stage I ovarian cancers are difficult to diagnose before surgery. Larger studies are necessary to confirm these pilot results. The diagnosis of malignancy in the 2 borderline tumors in this study, made only on the basis of contrast-enhanced imaging, supports this hypothesis. A recent case report of a normal-sized ovarian cancer, which was detected on power Doppler imaging with Levovist after failure of all the other imaging modalities, also supports the use of contrast agents in ovarian sonography.34 At present, contrast agents in gynecologic sonography are time-consuming and thus not appropriate in every case. We suggest using contrast agents in the following women: (1) those with adnexal masses and at least 1 suggestive gray scale or Doppler parameter, (2) those with no sonographically suggestive parameter but an elevated CA 125 level, and (3) women at potentially high risk (because of family history or BRCA1 and BRCA2) if this technology proves its usefulness in such a population. In conclusion, we have shown that contrast parameters were not independent from other Doppler and morphologic criteria. However, contrast agents, in particular during the washout period, conferred higher sensitivity to correctly differentiate ovarian cancer from benign masses with only a slight decrease of specificity, and use of these agents warrants further study. Our findings suggest a new way to assess the vascularization of an ovarian mass. Moreover, newer contrast agents and probes with harmonic capabilities are being adapted to accurately depict microcirculation, which should help correlate this minimally invasive technique with other biological and histologic parameters. In addition, screening for familial ovarian cancer with sonography, which is currently not successful, may be improved with the use of a contrast agent as a potential secondary test.35,36
This article has been cited by other articles:
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
Top
Abstract
Introduction
2 tests (or Fisher exact tests when appropriate). Multivariate logistic regression was done with all statistically significant variables in the univariate study. Correlation coefficients were computed to determine which parameters were least correlated, and these were used in developing the logistic regression analysis to determine the most predictive model. Data analyses were done with SAS (SAS Institute Inc, Cary, NC) and Stata (Stata Corp, College Station, Texas) software. 
