Professional Documents
Culture Documents
CHARACTERIZATION OF SONOGRAPHICALLY
INDETERMINATE OVARIAN TUMORS WITH MR IMAGING
Y. YAMASHITA~,
Y. HATANAKA', M. TAKA HAS HI^, K. MIYAZAKI~
M. TORASHIMA', and H. OKAMURA'
Departments of 'Radiology and 'Obstetrics and Gynecology, Kumamoto University School of Medicine, Kumamoto, Japan.
Abstract
Purpose: The goal of this study was to maximize the discrimination between Key words: Ovarian neoplasm, MR im-
benign and malignant masses in patients with sonographically indeterminate ovarian aging; tissue characterization, computer-
lesions by means of unenhanced and contrast-enhanced MR imaging, and to develop a assisted diagnosis.
computer-assisted diagnosis system.
Material and Methods: Findings in precontrast and Gd-DTPA contrast-enhanced Correspondence: Yasuyuki Yamashita,
MR images of 104 patients with 115 sonographically indeterminate ovarian masses Department of Radiology,
were analyzed, and the results were correlated with histopathological findings. Of 115 Kumamoto University School of
lesions, 65 were benign (23 cystadenomas, 13 complex cysts, 11 teratomas, 6 fibro- Medicine, 1-1-1 Honjo,
thecomas, 12 others) and 50 were malignant (32 ovarian carcinomas, 7 metastatic Kumamoto 860, Japan.
tumors of the ovary, 4 carcinomas of the fallopian tubes, 7 others). A logistic regres- FAX +81 96 362 4330.
sion analysis was performed to discriminate between benign and malignant lesions,
and a model of a computer-assisted diagnosis was developed. This model was pro- Accepted for publication 2 December
spectively tested in 75 cases of ovarian tumors found at other institutions. 1996.
Results: From the univariate analysis, the following parameters were selected as sig-
nificant for predicting malignancy (p10.05):a solid or cystic mass with a large solid
component or wall thickness greater thgn 3 mm; complex internal architecture; ascites;
and bilaterality. Based on these parameters, a model of a computer-assisted diagnosis
system was developed with the logistic regression analysis. To distinguish benign from
malignant lesions, the maximum cut-off point was obtained between 0.47 and 0.51. In
a prospective application of this model, 87% of the lesions were accurately identified
as benign or malignant.
Conclusion: Benign and malignant ovarian lesions can be distinguished in most
sonographically indeterminate lesions by means of parameters obtained from contrast-
I enhanced MR imaging.
The detection and characterization of an ovarian characterize a mass mainly when a sonogram is sub-
mass is a continuing clinical and radiological chal- optimal or indeterminate (11, 14). MR imaging can
lenge. The imaging characteristics of ovarian reliably diagnose fat-containing cystic teratoma and
masses do not always permit the differentiation of endometrioma, which are often problematical in ul-
malignant lesions from benign lesions. Although ul- trasound. However, the characterization of some
trasonography remains the foremost imaging mo- epithelial or stromal ovarian lesions is occasionally
dality for screening patients with adnexal lesions, difficult. In order to differentiate these diagnosti-
MR imaging may be of great help in identifying ma- cally indeterminate lesions, various trials have been
lignancy before surgery (8-10, 16, 17). However, performed with morphological characteristics (6,
the cost of MR is high so it should be used to help 12) or flow analysis using color Doppler (4, 5, 18).
572
MR IMAGING OF OVARIAN TUMORS
573
Y. YAMASHITA ET AL.
574
MR IMAGING OF OVARIAN TUMORS
Fig. 2. Serous ovarian cancer in the left ovary of a woman aged 51 years, visualized with a phased-array coil. a) T1-weighted SE
(600/14), b) T2-weighted turbo SE (4500/120/15 ETL) and c) postcontrast T1-weighted (600/14) images show a mass composed of a
solid and a cystic component. Ascites is not present. In this case, x,=18 (size), x,=O (massive ascites absent), x,=O (unilateral), x,=l
(heterogeneous internal architecture), and x,=l (solid) is given. The estimated probability of malignancy is 53%.
Fig. 3. Struma ovarii in the right ovary of a woman aged-58years, visualized with a phased-array coil. a) T1-weighted SE (600/14), b)
T2-weighted turbo SE (4500/120/15 ETL) and c) postcontrast T1-weighted (600/14) images show a predominantly cystic mass. The
wall of the cyst is thick and irregular. In this case, xl=8 (size), x,=O (ascites absent), x,=O (bilateral), x,=O (no internal architecture),
x,=l (wall with nodules). The estimated probability of malignancy is 31%.
Fig. 4.Fibroma associated with a simple cyst in the left ovary of a woman aged 37 years, visualized with a body coil. a) Precontrast
(600/15), b) T2-weighted TSE (2000/80) and c) postcontrast T1-weighted SE (600/15) images show a solid and cystic mass in the left
adnexa. In the solid mass, the signals in the T1-weighted and T2-weighted images are both hypointense, indicating a fibrous tumor. In
this case, x1=15(size), x,=O (ascites absent), x,=O (bilateral), x,=O (no internal architecture), x,=l (wall with nodules). The estimated
probability of malignancy is 38%.
architecture; c) presence of tumor necrosis; d) mas- ality had little value. From the data obtained from
sive ascites; and e) bilaterality (Table 2). Thick sep- the logistic regression analysis, a model of compu-
tum and signal intensity in either TI- or T2-weighted ter-assisted diagnosis was developed. By giving 0 or
images had less significant predictive values. 1 in xl,x2, --- x5,depending on the presence or ab-
The results of the logistic regression analysis are sence of findings, the predicted likelihood of malig-
shown in Table 3. The most significant finding for nancy was calculated for the individual tumor. If y is
malignancy was massive ascites, followed by wall the probability of malignancy, the equation is logit
structure and internal architecture. Size and bilater- (y)= -2.7+0.04x1+1.8x2+1.4x3+0.5x4+1.6x5.For
575
Y. YAMASHITA ET AL.
Discussion
The cost of MR imaging precludes its use in initial
screening for ovarian cancer. It should be used spe-
0.7- cifically to help characterize a mass, particularly
when a sonogram is suboptimal or indeterminate (11,
0.6- 14). On the basis of advances made in MR imaging
0.5- over the past several years, evaluation of adnexal
masses has become more accurate and a number of
0.4-
studies have evaluated the sensitivity and specificity
0.3- of MR imaging in distinguishingbenign from malig-
0.2- nant ovarian masses. Contrast enhancement with
Gd-DTPA allows better depiction of the internal ar-
0.1 - chitecture and differentiation of cystic from solid le-
0' I sions, and is found to be useful in differentiation be-
5 10 15 20 25 tween malignant and benign lesions (13, 15). Pelvic
MR imaging with use of a phased-may coil unques-
Tumor diameter (cm) tionably improves image quality. Normal ovaries can
Fig. 5. If y is the probability of malignancy, the equation is logit be detected in the majority of patients, and fine inter-
(y)=-2.7+0.04x ,+1.8x2+1.4x,+0.5x,+1.6x,. For each of the val- nal architectural details can be visualized with high
ues of x2 (massive ascites), x3 (bilaterality), x, (internal archi-
tecture), x5 (thick wall or solid), a regression curve can be
resolution fast SE T2-weighted images. In body-coil
drawn for logit (y) versus size (x,). The figure illustrates esti- MR imaging, a Gd-DTPA-enhanced study is manda-
mated probability of malignancy as a function of size and se- tory for detecting fine internal architectural details.
lected values of (x2,x3. x,, x5). In phased-array-coil MR imaging, on the other hand,
high resolution T2-weighted images may elimiate
the need for contrast enhancement.
each of the values of xl, x2, x3, x4, x5; a regression Small cystic or epithelial neoplasms were better
curve can be drawn for logit (y) versus size. For ex- characterized by transvaginal ultrasonography
ample, corresponding to x2=0 (massive ascites ab- owing to its ability to resolve the internal architec-
seht), x3=0(unilateral),x4=1 (heterogeneousinternal tural details such as the focal mural wall thickness
architecture), x5=0 (thin wall), the equation is logit or a solid component protruding from a predomi-
(y)= -2.2+0.04x1. Therefore the curve y= l/(l+e - nantly cystic mass. A scoring system based on the
[2.2+0.04x]) as a function of tumor size is obtained. morphological characteristics of a mass at ultra-
If the tumor is 30 cm in diameter, the estimated sonography can, with reasonable accuracy, differen-
probability of malignancy is calculated by giving tiate benign from malignant masses (6, 12). In these
x,=30 to the equation, and calculated as 27% (Fig.
1). Figs 2,3 and 4 are examples of calculations using
this model for malignant tumors. Fig. 5 illustrates
the estimated probability of malignancy as a func-
tion of size and selected values of (x2,xg,x4, x5).
In a retrospective application of this model, max- cn 20
imum accuracy in discriminating between benign .- 16
5
(I)
and malignant was obtained at a cut-off point of I2 12
0.49 (between 0.47 and 0.51) (Fig. 6). When this w-
0
model was prospectively applied to the 75 cases 0
8
found at other hospitals, 50 tumors (43 benign and 7
= 4
malignant) had an estimated probability of malig-
nancy of 0.49 or less, and 25 tumors (4 benign and 0
21 malignant) more than 0.49. The mean point value 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
obtained was 0.26f0.17 for the benign masses and
Estimated probability of malignancy
0.70f0.25 for the malignant tumors. The model had
an accuracy of 86% (sensitivity 89%, specificity Fig. 6. In retrospective calculation, with the logistic regression
model for all tumors, maximum accuracy in discriminating be-
89%) in distinguishing between benign and malig- tween benign and malignant lesions was obtained at a cut-off
nant sonographically indeterminate lesions at a cut- point of 0.49 (between 0.47 and 0.51). 0 Benign lesion. W Ma-
off point of 0.49. lignant lesion.
576
MR IMAGING OF OVARIAN TUMORS
studies, however, each variable was arbitrarily benign or not. However, a combination of ap-
scored. In addition, false-positive predictive diag- proaches, such as the one developed in this study,
noses were frequently seen in mature cystic terato- will probably be able to improve the differentiation
mas, fibrothecomas, and less frequently, endometri- of benign and malignant ovarian and adnexal lesions.
omas, because these diseases have variable sono-
graphic appearances. MR imaging can definitely REFERENCES
differentiate these benign lesions, because these tu-
1. BROWN D. L., FRATES M. C., LAING F. C. et al. : Ovarianmasses.
mors usually have characteristic MR appearances. Can benign and malignantlesions be differentiatedwith color
MR imaging can facilitate the differentiation of and pulse Doppler US? Radiology 190 (1994), 333.
the nature of the mass: simple fluid, atypical fluid, 2. Cox D. R.: The analysis of binary data. Metheuen, London
blood, solid tissue, fibrous tissue, fat, etc, but these 1970.
features do not usually lead to a specific diagnosis. 3. FISHER L. D. & VAN BELLE G.: Discrimination and classifica-
tion. In: Biostatistics. A methodology for the health scien-
MR findings useful for the prediction of malignancy ces, p. 630. Edited by L. D. Fisher & G. van Belle. Wiley-
in sonographically indeterminate masses have been International Science, New York 1993.
limited by the lack of explicit criteria for identifying 4. FLEISCHER A. C., ROGERS W. H., RAOB. K., KEPPLED. M. &
such masses. The superior resolution of high-field JONESH. W.: Transvaginal color Doppler sonography of
MR imaging has allowed us to develop a computer- ovarian masses with pathologic correlation. Ultrasound Ob-
stet. Gynecol. 1 (1991), 275.
assisted diagnosis system based on the morphology 5. KURJAK A., ZALUD I. & ALFIREVIC Z.: Evaluation of adnexal
of ovarian lesions. In our multivariate analysis, sig- masses with transvaginal color ultrasound. J. Ultrasound
nificant parameters for histological characterization Med. 10 (1991), 295.
were: a) thick wall or solid vs cystic appearance; b) 6. LERNER J. P., TIMOR-TRITSCH I. E., FEDERMAN A. & ABRAM-
OVICH G.: Transvaginal ultrasonographic characterizationof
internal architecture; c) bilaterality; and d) massive
ovarian masses with an improved, weighted scoring sys-
ascites. In our study, the signal intensity of the tu- tem. Am. J. Obstet. Gynecol. 170 (1994), 81.
mor did not have a predictive value in distinguish- 7. MORROW C. P. & TOWNSEND D. E. (also editors): Tumors of
ing between benign and malignant lesions. This is the ovary. General considerations, classification, the ad-
partly because the majority of benign masses, which nexal mass. In: Synopsis of gynecologic oncology, p. 23 1.
have characteristic signal intensities such as simple Wiley, New York 1987.
8. NISHIMURA K., TOGASHI K., ITOHK. et al.: Endometrial cysts
cysts, endometrioma or mature cystic teratoma, of the ovary. MR imaging. Radiology 162 (1987), 315.
were not included in the analysis. 9. NYBERG D. A., PORTER B. A., OLDSM. O., OLSON D. O., AN-
The data obtained from the logistic regression DERSEN R. & WESBY G. E.: MR imaging of hemorrhagic ad-
analysis may indicate the possibility of malignancy. nexal masses. J. Comput. Assist. Tomogr. 11 (1987), 664.
Because both malignant and benign cases are 10. OSMERS R., VOLKSEN M. & SHAUER A.: Vaginosonography
for early detection of endometrial carcinoma? Lancet 335
present at all probability levels in this computer- (1990), 1569.
assisted model, we still need a biopsy in the majority 11. RICCIO T. J., ADAMS H., MUNZING D. E. et al.: Magnetic reso-
of cases. However, the information of probability of nance imaging as an adjunct to sonography in the evalua-
malignancy will be of great help before surgery. The tion of the female pelvis. Magn. Reson. Imaging 8 (1990),
model for predicting malignancy may be applicable 699.
12. SASSONE A. M., TIMOR-TRITSCH I. E., ARTNER A. et al.:
to transvaginal ultrasound findings because all the Transvaginal sonographic characterization of ovarian dis-
parameters used can be obtained with transvaginal ease. Evaluation of new scoring system to predict ovarian
ultrasound. It should be noted, however, that para- malignancy. Obstet. Gynecol. 78 (1991), 70.
meters that are useful in a highly selected population 13. STEVENS S . K., HRICAK H. & STERN J. L.: Ovarian lesions.
referred for surgery and with a high prevalence of Detection and characterization with gadolinium-enhanced
MR imaging at 1.5 T. Radiology 181 (1991), 481.
ovarian cancers can prove to be poor predictors in 4. TAYLOR K. J. W. & SCHWARTZ P. E.: Screening for early
the general population in which the prevalence of ovarian cancer. Radiology 192 (1994), 1.
malignancy is very low and the prevalence of physi- 5. THURNHER S., HODLER J., BAERS., MARINCEK B. & VON
ological cysts is high (14). Therefore this model SCHULTHESS G. K.: Gadolinium-DOTA enhanced MR imag-
should be useful only in highly selected sonographi- ing of adnexal tumors. J. Comput. Assist. Tomogr. 14
(1990), 939.
cally indeterminate ovarian lesions. 6. TOGASHI K., NISHIMIJRA K. & ITOHK. et al.: Ovarian cystic
Conclusion: A distinction can be made in most teratomas. MR imaging. Radiology 162 (1987), 669.
cases between benign and malignant ovarian lesions 7. TOGASHI K., NISHIMURA K., KIMURA I. et al.: Endometrial
by means of parameters obtained from unenhanced cysts. Diagnosis with MR imaging. Radiology 180 (1991),
and contrast-enhanced MR imaging. In our quest to 73.
8. WEINER Z., THALER I., BECKD., ROTTEM S . , DEUTSCH M. &
accurately predict the histological characterization of BRANDES J.: Differentiating malignant from benign ovarian
a mass, we have come far but we still find it difficult tumors with transvaginal color flow imaging. Obstet.
to identify certain lesions of complex appearance as Gynecol. 79 (1992), 159.
577