Characterization of Adnexal Mass Lesions on MR Imaging

Transcription

1 S.. slam Sohaib 1,2 hju Sahdev 1 Philippe Van Trappen 3 Ian J. Jacobs 3 Rodney H. Reznek 1 Received June 7, 2002; accepted after revision October 9, Presented in part at the annual meeting of the merican Roentgen Ray Society, tlanta, pril May S... Sohaib was partially supported by the Joint Research oard of St. artholomew s Hospital. 1 Department of Diagnostic Imaging, St. artholomew s Hospital, West Smithfield, London, EC1 7E, England, United Kingdom. 2 cademic Department of Diagnostic Radiology, Royal Marsden Hospital, Downs Rd., Sutton, Surrey SM2 5PT, United Kingdom. ddress correspondence to S... Sohaib. 3 Department of Gynaecological Oncology, St. artholomew s Hospital, London, EC1 7E, England, United Kingdom. JR 2003;180: X/03/ merican Roentgen Ray Society Characterization of dnexal Mass Lesions on MR Imaging OJECTIVE. The aim of our study was to evaluate the accuracy of MR imaging in the detection and characterization of adnexal mass lesions and to determine which imaging features are predictive of malignancy. SUJECTS ND METHODS. We prospectively performed MR imaging in 104 patients (age range, years; mean age, 50 years) with clinically or sonographically detected complex adnexal masses. We used a 1.5-T unit to perform T1-, T2-, and fat-suppressed T1-weighted sequences before and after IV injection of gadolinium. The adnexal lesions were examined for several features including size, shape, character (solid cystic), vegetation, signal intensity, and enhancement. Secondary signs such as ascites, peritoneal disease, and lymphadenopathy were noted. We compared the imaging features with the surgical and pathologic findings. Multiple logistic regression analysis was performed on all MR imaging features. RESULTS. total of 163 lesions 94 benign and 69 malignant lesions were examined. On MR imaging, 95% (155/163) of the lesions were detected. The overall accuracy for the diagnosis of malignancy was 91%. On univariate analysis, the imaging features associated with malignancy were a solid cystic lesion, irregularity, and vegetation on the wall and septum in a cystic lesion, the large size of the lesion, an early enhancement on dynamic contrast-enhanced MR images, and the presence of ascites, peritoneal disease, or adenopathy. On multiple logistic regression analysis, ascites and vegetation in a cystic lesion were the factors most significantly indicative of malignancy. CONCLUSION. MR imaging is highly accurate in the characterization of adnexal mass lesions, and the best predictors of malignancy are vegetation in a cystic lesion and ascites. D etermining whether a clinically diagnosed adnexal mass is benign or malignant is frequently not possible until surgical exploration and histologic examination are performed. Consequently, it may not be possible to decide preoperatively whether conservative or radical surgery is appropriate. reliable method with which to differentiate a benign from a malignant adnexal mass would provide a basis for optimal preoperative planning and may also reduce the number of unnecessary laparotomies patients undergo for benign disease. Studies have shown that MR imaging has a potential role in the characterization of adnexal masses [1 3]. Researchers in these studies evaluated various MR techniques of imaging pelvic masses and found that gadolinium-enhanced MR imaging provides the best assessment of complex adnexal masses [1, 3]. Imaging criteria used to distinguish benign from malignant lesions have been based on surgical and pathologic findings [3]. However, only limited information is available as to which MR imaging features are best to use in distinguishing benign from malignant adnexal lesions [4]. The aim of our study was to evaluate the accuracy of MR imaging in the detection and characterization of adnexal mass lesions and to determine which morphologic features are most predictive of malignancy. Subjects and Methods Patients prospective study was conducted of all women referred to our institution for evaluation of adnexal mass lesions between pril 1998 and November The local research ethics committee approved the study, and consent was obtained from all patients before they were enrolled in our study. total of 119 patients with clinically or JR:180, May

2 Sohaib et al. sonographically detected adnexal masses were recruited for pelvic MR imaging. Of the 119 patients, eight patients did not keep their appointment for the MR imaging examination or were unable to complete the MR imaging examination because of claustrophobia. Seven more patients were either lost to follow-up or did not undergo surgery at our institution and so had no corresponding pathologic data available. The remaining 104 women (age range, years; mean age, 50 years) were the subjects of this study. MR Imaging Techniques MR imaging was performed on a 1.5-T MR imaging unit (Signa Horizon; General Electric Medical Systems, Milwaukee, WI). pelvic phased array coil was used in all patients. The following sequences were obtained: axial T1-weighted spin-echo MR imaging from the renal hilum to the symphysis pubis or beyond if necessary to cover the larger adnexal masses (TR range/te range, /10 14; slice thickness, 5 8 mm; gap, 1 2 mm; field of view, cm; excitations, 1 2; matrix, ; and respiratory compensation); axial T2-weighted fast spin-echo MR imaging of the pelvis or beyond if necessary to cover the larger adnexal masses (TR range / effective TE range, /90 110; echo-train length, 8; slice thickness, 5 7 mm; gap, 1 2 mm; field of view, cm; excitations, 2; and matrix, ); sagittal T2-weighted fast spin-echo imaging from one femoral head to the other (TR range/te range, /90 110; echo-train length, 8; slice thickness, 5 7 mm; gap, 1 2 mm; field of view, cm; excitations, 2; and matrix, ). Unenhanced and enhanced fat-suppressed spoiled gradient-echo T1-weighted imaging was performed in the best plane for visualizing the particular lesion (TR/TE, 150/4.2; flip angle, 90 ; slice thickness, 5 8 mm; gap, 1 2 mm; field of view, cm; excitations, 2 4; and matrix, ; and spectral fat suppression). Contrast-enhanced images were obtained after IV injection of 0.1 mmol/ kg of gadopentetate dimeglumine (Omniscan; Nycomed mersham, Little Chalfont, United Kingdom). In the latter half of the study, dynamic contrast-enhanced imaging was performed instead of the unenhanced and contrast-enhanced sequences. The dynamic contrast-enhanced fat-suppressed spoiled gradient-echo T1-weighted MR imaging was performed through the lesion in the optimal plane with imaging parameters of 150/4.2; flip angle, 90 ; slice thickness, 5 8 mm; gap, 1 2 mm; field of view, 30 cm; excitation, 1; and matrix, The total acquisition time for this sequence was sec. This sequence was performed before and immediately after a rapid hand IV injection of 0.1 mmol/kg of gadopentetate dimeglumine and then repeated at 30, 60, 90, and 120 sec into the examination. MR Image nalysis The MR images were evaluated by two radiologists in consensus without knowledge of the surgical or pathologic findings. The MR imaging features were then correlated with the surgical and pathologic findings. The imaging features documented include the number of adnexal masses per patient, origin of lesion (ovarian or extraovarian), lesion shape, lesion size, and content of lesion (solid only, mainly solid, solid cystic, mainly cystic, and cystic only). If a wall could be identified, its thickness, character, and enhancement were noted. If septa were present in the lesion, the number, thickness, character, and enhancement of the septa were recorded. ny vegetation appearing on the wall or the septum of the lesion was measured and noted. In addition, we documented the presence of a hemorrhage or fat. We determined that a hemorrhage was present if signal intensity was high on T1-weighted spin-echo and fat-suppressed T1- weighted MR sequences. We determined that fat was present if the lesion showed high signal on T1-weighted MR images that lost signal on the fat-suppressed T1-weighted MR images. Tissue with low signal intensity on T2-weighted MR images (i.e., signal intensity of skeletal muscle) was also noted. Such low-signal-intensity tissue is indicative of fibrous tissue, which is found in benign ovarian tumors [5]. TLE 1 Histopathologic Diagnosis for 163 dnexal Masses Note. ecause of rounding, percentages do not total 100%. On the dynamic contrast-enhanced images, the signal intensity of the solid components was measured before and then 60 and 120 sec after injection of IV gadolinium. We calculated the percentage of increase in signal intensity at 60 sec (early) and at 120 sec (late) of enhancement. Other MR imaging features included in the study were the presence of ascites or peritoneal disease, lymph node size and site, and involvement of adjacent organs and pelvic sidewall. The radiologists subjective impression of the probability that a lesion was malignant was scored on a scale of 1 5 (1, benign; 2, probably benign; 3, possibly malignant; 4, probably malignant; and 5, malignant.) Statistical nalysis Each MR imaging feature was assessed individually with regard to its relationship with the final diagnosis (benign vs malignant) with the Student s t test for normally distributed continuous variables, the Mann-Whitney U test for abnormally distributed continuous variables, and the chi-square test (with the Yates correction, as appropriate) for categorical variables with more than two values. Descriptive statistical values such as accuracy, sensitivity, and specificity were also determined for each MR imaging feature. Diagnosis No. % of ll Masses Studied enign 94 Endometrioma Nonneoplastic ovarian cyst Cystadenoma Teratoma Fibroma or thecoma Leiomyoma Pelvic inflammatory disease Ovarian infarction Hydrosalphinx Malignant 69 Serous cystadenocarcinoma orderline tumor Undifferentiated carcinoma Mucinous cystadenocarcinoma Endometroid cancer Metastasis Carcinoid tumor Transitional cell cancer Clear cell cancer Sarcoma Cervical cancer Lymphoma Small cell cancer JR:180, May 2003

3 MR Characterization of dnexal Mass Lesions Stepwise logistic regression analysis was used to identify which group of features allowed the best prediction of benignity versus malignancy. The imaging features that were found to be statistically significant using univariate analysis were entered into a multivariate model to gauge the independent predictive value and determine which combination of findings would be most predictive of malignancy. The results were expressed as an odds ratio of malignancy in a lesion in which a specific MR imaging feature was present [6]. In differentiating between benign and malignant lesions, we also performed receiver operating characteristic curve analysis. Features such as the confidence of the observers in differentiating benign and malignant masses, the size of the lesions, and the percentages of increase in enhancement were compared with the actual nature of the lesions using the receiver operating characteristic analysis. receiver operating characteristic curve for the regression model was constructed using the estimated probability of malignancy from the model as thresholds to generate sensitivities and specificities [6]. The mean area under the curve and the standard error were also determined. ll statistical analyses were performed using a statistical software package (SPSS version 9; SPSS, Chicago, IL), with significance taken as p less than were a cystadenoma, a paratubal cyst, and a simple cyst. ll these lesions were smaller than 2 cm. One of the five malignant lesions not detected was a small deposit of clear cell cancer from a large contralateral ovarian cancer. In four patients, what was thought to be a single large mass on MR imaging was found to be bilateral ovarian cystadenocarcinoma at surgery. In terms of characterizing the detected lesions as malignant, MR imaging had a sensitivity of 95% (61/64), specificity of 88% (80/91), a positive predictive value of 86% (61/72), a negative predictive value of 96% (80/83), and an overall accuracy of 91% (141/155). The three false-negative findings were thought to be benign cystadenomas on MR imaging, but histologic examination revealed all three to be borderline ovarian tumors (Fig. 1). Eleven falsepositive findings were thought to be malignant on MR imaging but were benign on histology: three cystadenomas, three hemorrhagic cysts or endometriomas (Figs. 2 and 3), two adenofibromas, one infarcted ovary, one case of granulomatous salpingo-oophoritis (Fig. 4), and one leiomyoma. ll the false-positive cases had some enhancing soft-tissue component visualized on MR imaging. Table 2 summarizes the characteristics of the benign and malignant lesions. Features that were shown not to be significantly different between benign and malignant masses included the number of different solid or cystic components seen on MR imaging and the shape of the lesion. On the univariate analysis, the most significant fea- Results t surgery and pathology, 163 masses were found in the 104 patients; 94 (58%) were benign and 69 (42%) were malignant. The histopathologic diagnoses of the 163 lesions are shown in Table 1. On MR imaging, 155 (95%) of the 163 masses were identified, with 91 (97%) of the 94 benign lesions and 64 (93%) of the 69 malignant lesions being detected. No statistically significant difference in the rate of detection of benign and malignant lesions was found. The three benign lesions not detected Fig year-old woman with bilateral borderline mucinous ovarian tumor. xial T2-weighted fast spin-echo MR image shows large left cystic lesion (black arrows) with small septum and small right cystic lesion (white arrow). ecause no features on MR imaging suggested malignancy, lesions were thought to be benign. Fig year-old woman with large endometriotic cyst (arrow) in left ovary. xial T1-weighted fast spinecho MR image shows large thick-walled cystic lesion containing high signal intensity and thick nodular septa that enhanced after IV administration of gadolinium (not shown). On MR imaging, this lesion was incorrectly interpreted as malignancy. Fig year-old woman with large endometriotic cyst. and, xial T1-weighted spin-echo MR images obtained with frequencyselective fat saturation before () and after () IV administration of gadolinium shows cystic lesion (arrows) with high signal intensity and thick wall with enhancing nodule (arrowhead, ) seen posteriorly. Lesion was incorrectly interpreted as malignancy on MR imaging. JR:180, May

4 Sohaib et al. TLE 2 Note. NS = not statistically significant. a Mann-Whitney U test. b Chi-square test. c Student s t test. Characteristics of 155 enign and Malignant dnexal Masses Detected on MR Imaging Variable enign (n = 91) Malignant (n = 64) Mean or No. Fig year-old woman with granulomatous salpingooophoritis. xial T2-weighted fast spin-echo MR image shows cystic lesion (arrows). Note septum with nodule (arrowhead). On MR imaging, lesion was incorrectly interpreted as malignancy. Range,%, or SD Mean or No. Range,%, or SD Maximal diameter (cm) a Volume (cm 3 ) , <0.001 a ilateral masses 52/ / b Content Solid only 10/ / NS b Solid cystic 66/ / b Cystic only 15/ / b Wall characteristics Thickness > 3 mm 26/ / NS b Irregularity 28/ / b Nodule present 5/ / <0.001 b Septum characteristics Number of septa No septum 48/ / NS b One septum 12/ / NS b Few (2 5) septa 21/ / NS b Multiple (>5) septa 10/ / b Thickness > 3 mm 3/ / b Irregularity 10/ / b Nodule present 5/ / b Hemorrhage in cystic lesion 36/ / b Necrosis in solid lesion 3/ / NS b Low signal on T2-weighted imaging 27/ / b % of early enhancement (n = 40 masses) c % of late enhancement (n = 40 masses) c scites 4/ / <0.001 b Peritoneal disease 0/ / <0.001 b Lymphadenopathy 3/ / b p tures indicative of malignancy were vegetation (Fig. 5) on the wall or septum in a cystic lesion, large size of the lesion, and the presence of ascites and peritoneal disease. On a stepwise multivariate logistic regression analysis, imaging features predictive of malignancy were a maximal diameter greater than 6 cm, vegetation on the wall of a cystic lesion, and the presence of ascites. The results of the multivariate regression analysis are shown in Table 3. The larger odds ratio reflects a greater association of the imaging feature with malignant masses. Figure 6 shows the receiver operating characteristic curves for this model, as well as those for lesion volume, the percentage of early enhancement, and the radiologists subjective impressions of malignancy. Discussion ccurate evaluation of adnexal masses is important for optimal treatment planning. Sonography remains the primary imaging modality for the evaluation of adnexal masses; MR imaging is considered a problem-solving technique. Contrast-enhanced MR imaging appears to be more accurate than sonography for the assessment of adnexal masses [7]. Our study confirms the high detection rate and accurate characterization of adnexal lesions that are possible using MR imaging. Our overall detection rate of 95% is comparable to rates reported in previous studies [1, 4]. Potential problems in MR imaging detection of adnexal lesions include difficulty in revealing small lesions and occasional difficulty in determining whether a large adnexal mass is unilateral or bilateral [4]. The excellent characterization of adnexal masses on contrast-enhanced MR imaging is due to depiction of the internal architecture of complex adnexal masses. Furthermore, the multiplanar imaging capability allows accurate identification of the origin of adnexal mass lesions. The results of our study show that the overall diagnostic accuracy of 91% for distinguishing benign from malignant adnexal lesions is comparable to the accuracy of previous reports [2 4, 8]. Lesions that were incorrectly classified were borderline ovarian tumors (Fig. 1) and other benign lesions with some solid enhancing elements (Figs. 2 4). orderline ovarian tumors are often difficult to characterize because their morphologic features are similar to those of benign ovarian lesions and are therefore frequently misclassified, whether MR imaging or another technique is used JR:180, May 2003

5 MR Characterization of dnexal Mass Lesions We evaluated multiple imaging features to determine the best predictors of malignancy (Table 2). Solid cystic lesions are more likely to be malignant. whereas purely solid or purely cystic lesions are more likely to be benign. For cystic and solid cystic lesions, the imaging characteristics of the wall of cystic lesions are important to evaluate. Wall irregularity and vegetation on the wall of these lesions are both indicative of malignancy. Unlike previous researchers [4, 9], we did not find a thick wall to be indicative of malignancy. Our analysis included all masses, and many benign lesions such as endometriomas (Figs. 2 and 3) and teratomas (Fig. 7) that were excluded in the previous studies can have a thick wall. Unlike the case of a lesion with one or few septa, a lesion with multiple (> five) septa is suspicious for malignancy. s with the lesion wall, irregularity and vegetation on the septum are also strongly indicative of malignancy. However, unlike a thick lesion wall, a thick septum is suggestive of malignancy (Figs. 5 and 8). The presence of a hemorrhage in a cystic lesion detected on MR imaging is more common in benign lesions than in malignant lesions. However, this feature alone does not help one to differentiate benign from malignant lesions because a substantial number (25%) of malignant lesions may contain a Fig. 6. reas under receiver operating characteristic curves ( z ) compare performance of four diagnostic criteria: radiologists interpretations of masses as malignant ( ; z = 0.96 ± 0.02), regression model based on MR imaging features ( ; z = 0.86 ± 0.03), percentage of enhancement seen within 60 sec of rapid bolus injection of IV gadolinium ( ; z = 0.74 ± 0.1), and volume of lesions ( ; z = 0.68 ± 0.04). Radiologists interpretations had largest area under curve and thus was best discriminator between benign and malignant adnexal masses. Sensitivity Fig year-old woman with poorly differentiated serous cystadenocarcinoma. xial T2-weighted fast spin-echo MR image shows cystic lesion (white arrow). Note vegetation on lesion wall (black arrow) and thick irregular septa (arrowhead) within lesion. TLE 3 hemorrhage (Fig. 8). Other features indicative of malignancy are present in such cases. For predominantly solid lesions, we did not find the presence of necrosis to be a feature of malignancy [4] (Figs. 9 and 10). This 0.1 Variable Multiple Logistic Regression nalysis of MR Imaging Features Found in 133 dnexal Masses Regression Coefficient b Specificity Standard Error (b) finding may be due to the fact that benign solid lesions, such as degenerating fibroids, also showed necrosis, and some fibromas and thecomas had cystic changes that mimicked necrosis in a solid lesion. The low-sig p 0.8 Odds Ratio Wall nodule < scites Maximal diameter > 6 cm Constant < JR:180, May

6 Sohaib et al. nal-intensity pattern seen in solid lesions on T2-weighted MR imaging was helpful in distinguishing benign from malignant lesions. Many benign ovarian tumors, especially fi- brotic tumors, characteristically have low signal intensity on T2-weighted MR images [5, 10] (Figs. 9 and 11). The low signal intensity of the fibrotic component of ovarian fibroma (Fig. 9) contrasts with the intermediate signal intensity of the carcinoma (Fig. 10). The vascularity of ovarian tumors has been extensively studied using Doppler Fig year-old woman with cystic teratoma in left ovary., xial T1-weighted spin-echo MR image shows well-defined thick-walled adnexal mass (arrows) with extremely high signal intensity., xial T1-weighted spin-echo MR image obtained with frequency-selective fat saturation shows central portion of mass (arrows) with low signal intensity confirming presence of fat. ppearance is typical of teratoma, which was confirmed at histopathology. C Fig year-old woman with clear cell cancer of ovary., Sagittal T2-weighted fast-spin echo MR image shows solid cystic mass (arrow). Note thick irregular septum (arrowhead). and C, Unenhanced () and contrast-enhanced (C) sagittal T1-weighted spin-echo MR images obtained with frequency-selective fat saturation show enhancing soft tissue (arrowhead, C). High signal intensity in cystic component of mass (arrow) was due to blood related products. Fig year-old woman who had benign ovarian fibroma with central necrosis., Sagittal T2-weighted fast spin-echo MR image shows large heterogeneous mass (arrows) with areas of low signal intensity., Contrast-enhanced T1-weighted spin-echo MR image obtained with frequency-selective fat saturation shows areas of nonenhancement (arrowheads) in keeping with necrosis within mass (arrows). t histopathology, nonenhancing areas corresponded to areas of myxoid degeneration JR:180, May 2003

7 MR Characterization of dnexal Mass Lesions Fig year-old woman with poorly differentiated adenocarcinoma of ovary., xial T2-weighted fast spin-echo MR image shows large heterogeneous mass (arrow)., Contrast-enhanced T1-weighted spin-echo MR image obtained with frequency-selective fat saturation shows areas of necrosis (arrowhead) within mass (arrow). sonography. However, data on the dynamic contrast enhancement pattern on MR imaging in adnexal masses are scarce. small series showed early enhancement in borderline ovarian tumors to be a better predictor of malignant tumors than C-125 levels and sonographic findings [11]. Our study found that malignant lesions show greater enhancement than benign lesions during the early phase of enhancement rather than the late phase of enhancement. Using a threshold of a 100% increase in signal intensity gives a specificity of 100% but a sensitivity of 30% in identifying malignant lesions. This pattern of strong early enhancement is similar to the observation made in studying tumors in other parts of the body that malignant tumors show rapid early enhancement and washout, whereas benign tumors show a slower sustained enhancement Fig year-old woman with right ovarian fibroma (arrow). xial T2-weighted fast spin-echo MR image shows well-defined solid lesion with low signal intensity. [12]. This feature is also in keeping with previously reported sonographic data that have shown that malignant ovarian lesions have a low resistance to blood flow, explaining the rapid enhancement and washout [13, 14]. Secondary features such as ascites, peritoneal disease, or lymphadenopathy were all strongly indicative of malignancy. These features are highly specific but lack sensitivity for characterizing malignant lesions, in particular for early-stage disease. Multivariate logistic regression analysis indicated that of the many features indicative of malignancy, the only features significantly and independently associated with malignancy were vegetation in a cystic lesion, the presence of ascites, and a maximal diameter greater than 6 cm. To our knowledge, only two studies on the multivariate analysis of the MR imaging features exist [4, 9]. The authors of these studies reported that on multivariate analysis, ascites, vegetation in a cystic lesion, necrosis in a solid lesion, bilateral lesions, and solid or irregular wall structures were features indicative of malignancy. Similar features have been reported on logistic regression analyses of the sonographic appearance of adnexal mass lesions [15 17]. Of all the methods for differentiating benign from malignant lesions, the radiologists subjective impression of malignancy was the best discriminator (Fig. 6). Subjective assessment has also been found to be superior to mathematic models in studies of sonographic data [18]. Our model included only three imaging variables. dding any further variable did not statistically improve the performance of the model, perhaps because of the relatively small population size of our study. The radiologists are probably using more of the imaging information than can be modeled in a multivariate regression analysis. potential criticism of our study is that the patient selection was biased. Only patients with adnexal masses were recruited from a gynecologic oncology clinic. However, our subjects were patients referred to a regional cancer center and reflect the group of patients referred in clinical practice for MR imaging evaluation of complex adnexal masses. In summary, MR imaging is an excellent technique for the detection and characterization of adnexal mass lesions. The multiplanar capability of MR imaging allows the accurate determination of the origin of the tumor. Contrast-enhanced MR imaging provides a depiction of the internal architecture of lesions, particularly vegetation in a solid cystic lesion. Many features are indicative of malignancy, including large lesion size, bilateral masses, multiple septa, and irregularity and vegetation on wall and septa. However, of the multiple features evaluated, vegetation in a cystic lesion and ascites were the most important features for distinguishing benign from malignant adnexal lesions. cknowledgments We thank Roger Hern for statistical advice and Janet MacDonald for assistance with the illustrations. References 1. Thurnher S, Hodler J, aer S, Marincek, von Schulthess GK. Gadolinium-DOT enhanced MR imaging of adnexal tumors. J Comput ssist Tomogr 1990;14: Komatsu T, Konishi I, Mandai M, et al. dnexal masses: transvaginal US and gadolinium-enhanced MR imaging assessment of intratumoral structure. Radiology 1996;198: Stevens SK, Hricak H, Stern JL. Ovarian lesions: detection and characterization with gadoliniumenhanced MR imaging at 1.5 T. Radiology 1991;181: Hricak H, Chen M, Coakley FV, et al. Complex adnexal masses: detection and characterization with MR imaging multivariate analysis. Radiology 2000;214: Outwater EK, Siegelman ES, Talerman, Dunton C. Ovarian fibromas and cystadenofibromas: MRI features of the fibrous component. J Magn Reson Imaging 1997;7: ltman DG. Practical statistics for medical research. London: Chapman & Hill, 1991: Strigini F, Gadducci, Del ravo, Ferdeghini M, Genazzani R. Differential diagnosis of adnexal masses with transvaginal sonography, color flow imaging, and serum C-125 assay in pre- and post- JR:180, May

8 Sohaib et al. menopausal women. Gynecol Oncol 1996;61: Scoutt LM, McCarthy SM, Lange R, ourque, Schwartz PE. MR evaluation of clinically suspected adnexal masses. J Comput ssist Tomogr 1994;18: Yamashita Y, Hatanaka Y, Torashima M, Takahashi M, Miyazaki K, Okamura H. Characterization of sonographically indeterminate ovarian tumors with MR imaging: a logistic regression analysis. cta Radiol 1997;38: Outwater EK, Siegelman ES, Kim, Chiowanich P, lasbalg R, Kilger. Ovarian renner tumors: MR imaging characteristics. Magn Reson Imaging 1998;16: van Vierzen P, Massuger LF, Ruys SH, arentsz JO. orderline ovarian malignancy: ultrasound and fast dynamic MR findings. Eur J Radiol 1998;28: Padhani R, Husband JE. Dynamic contrast enhanced MRI studies using ECF agents. Clin Radiol 2001;56: Kurjak, Shalan H, Zalud I. Evaluation of adnexal masses with transvaginal color ultrasonography. (commentary) J Ultrasound Med 1992;11: lcazar JL, Errasti T, Zornoza, Minguez J, Galan MJ. Transvaginal color Doppler ultrasonography and C-125 in suspicious adnexal masses. Int J Gynaecol Obstet 1999;66: Mol W, oll D, De Kanter M, et al. Distinguishing the benign and malignant adnexal mass: an external validation of prognostic models. Gynecol Oncol 2001;80: slam N, anerjee S, Carr JV, Savvas M, Hooper R, Jurkovic D. Prospective evaluation of logistic regression models for the diagnosis of ovarian cancer. Obstet Gynecol 2000;96: Timmerman D, ourne TH, Tailor, et al. comparison of methods for preoperative discrimination between malignant and benign adnexal masses: the development of a new logistic regression model. m J Obstet Gynecol 1999;181: Valentin L, Hagen, Tingulstad S, Eik-Nes S. Comparison of pattern recognition and logistic regression models for discrimination between benign and malignant pelvic masses: a prospective cross validation. Ultrasound Obstet Gynecol 2001;18: The full text and images from the merican Journal of Roentgenology may also be viewed online at or JR:180, May 2003