Downloaded from www.ajronline.org by 155.33.148.196 on 02/16/15 from IP address 155.33.148.196. Copyright ARRS. For personal use only; all rights reserved
1251
Value of MR Imaging in Differentiating Benign from Malignant Soft-Tissue Masses: Study of 95 Lesions
Thomas H. Berquist1 Richard L. Ehman1 Bernard F. King1 Charles G. Hodgrnan1 Duane M. llstrup2
MR imaging has largely replaced CT as the technique of choice for preoperative staging of patients with soft-tissue masses. Whether MR imaging can be used to differentiate benign from malignant masses is controversial. Our experience suggests that MR imaging often can characterize soft-tissue masses accurately. To evaluate this question further, we studied 95 consecutive lesions (50 benign and 45 malignant). Consecutive cases were selected to simulate our clinical practice. Surgical proof was available for all masses except hematomas, for which clinical follow-up confirmed the diagnosis. MR images were interpreted twice by three radiologists. The first review was accomplished without any clinical history and the second review with clinical history. Reviewers were asked to classify the lesion as benign or malignant on the basis of their clinical knowledge and analysis of MR image features (size, lesion margin, signal homogeneity, and neurovascular or bone involvement). Although interpretation varied somewhat because of the experience of the reviewers, the specificity and accuracy of diagnosis averaged 90% for both benign and malignant lesions. Negative predictive value for malignancy averaged 94% among the three reviewers. MR imaging is the technique of choice for identification and characterization of softtissue masses. The nature of the lesion (benign vs malignant) can be determined in the majority of cases. AJR
Received
April 5, 1 990; accepted
June26, 1990. 1 Department
of Diagnostic
Clinic and Mayo Foundation,
after
Radiology,
Rochester,
revision
Section
Foundation,
of Biostatistics, Rochester,
Mayo
MN 55905.
0361-803X/90/1 556-1251 C American Roentgen Ray Society
Clinic
December
1990
MR imaging has become the premier method for identifying and staging softtissue masses of the musculoskeletal system. Although MR imaging is sensitive, many authors have reported that it is not useful in differentiating benign from malignant soft-tissue masses [1-S]. Our experience indicates that certain image features are useful in differentiating benign from malignant lesions. Therefore, we elected to perform a blind study of MR images of benign and malignant soft-tissue masses. Images were obtained and interpreted before biopsy or surgical data were obtained. Ninety-five (SO benign and 45 malignant) consecutive soft-tissue masses were studied to determine the usefulness of MR imaging features in differentiating benign from malignant lesions. Nonneoplastic lesions such as abscess and hematoma were included with benign tumors. The primary goal of this study was to determine the value of morphologic characteristics (e.g., lesion size, margins, signal intensity, and neurovascular involvement) in predicting the nature of soft-tissue masses. We also chose radiologists with different levels of experience with MR imaging to determine what effect experience has on the accuracy of interpretation.
Mayo
MN 55905.
Address reprint requests to T. H. Berquist. 2
155:1251-1255,
Materials
and Mayo
and
Methods
MR examinations biopsies The
were
specific
selected diagnoses
of 95 consecutive for
review.
are given
This in Table
and other lesions that mimic neoplasms.
patients series
with included
soft-tissue 45 malignant
masses and
who had not had 50 benign masses.
masses included hematomas, abscesses, Surgical proof was available in all cases except for
1 . Benign
BERQUIST
1252
TABLE
1: Diagnoses
of 95 Soft-Tissue
ET
AL.
TABLE
Masses
Type of Mass
AJR:155,
2: MR Features
of Soft-Tissue
No. (%)
r:5
Feature
Downloaded from www.ajronline.org by 155.33.148.196 on 02/16/15 from IP address 155.33.148.196. Copyright ARRS. For personal use only; all rights reserved
12 9
Cyst Hematoma Lipoma
8
Hemangioma
5
Desmoid Neuroma
4 4
Abscess
2
Malignant
Size (cm) 5 Margin Well
2 1 i i
aneurysm vein
Total Malignant
Irregular Signal
histiocytoma
0 0 6(13) 39 (87)
1 (2) 10(20) 14(28) 25 (50)
7 (i 5)
22 (44)
1 3 (29)
20 (40)
25 (56)
8 (16)
homogeneous
2 (5) 1 1 (24)
21 (42) i 7 (34)
Inhomogeneous
32 (71)
i 2 (24)
17
Liposarcoma Synovial sarcoma
7 6
Mesenchymal sarcoma Rhabdomyosarcoma
3 3
Chondrosarcoma Leiomyosarcoma Neurofibrosarcoma
2 2 2
the lesions (which were to be defined as smooth, partially well defined, or irregular) and the signal intensity and/or homogeneity (recorded on both Ti - and T2-weighted sequences). Also evaluated were neurovascular encasement or displacement, hemorrhage and/or edema in or around the lesion, and bone involvement. After each of
Fibrosarcoma
1
these features was evaluated,
Epithelioid
1
the lesion
as benign
or malignant
histologic
diagnosis.
Specific
(soft tissue)
sarcoma
Lymphoma Total
45
information
hematomas. In these patients, follow-up and hematomas confirmed the clinical was
not
available
to the
in both
upper
and
lower
with resolution diagnosis. This
the lower extremity: 29 (31 %) in 1 1 (i 2%) around the knee. The upper extremity and trunk. The 71 %) also were in the lower always immediately adjacent to
extremities.
MR images were obtained on either a 0.i5-T resistive system (25 patients) or a 1 .5-T Signa system (70 patients). Both Ti -weighted, 500/20 (TRITE), and T2-weighted, 2000/60, spin-echo (SE) sequences were available for each examination. Only pertinent images were selected for review. All patient data were removed from the
images, and all cases were randomized not
separated)
before
(benign and malignant
were
evaluation.
Images were interpreted blindly in separate sessions by three different radiologists. Two radiologists were fellows in cross-sectional imaging
and in the process
of taking
a 3-month
when they began their participation was
a consultant
imaging. in two
Images different
with
were settings.
5 years
MR
imaging
fellowship
in the study. The third radiologist
of experience
randomized
in musculoskeletal
and reviewed
In the first
part
of the
MR
by each radiologist study,
the
radiologists
were given images to review
without any clinical information or other imaging studies to compare with the MR study. After 3-6 months, the images were put in a different order and given to the interpreters a second
of other
evaluation.
imaging
studies
At this
were
able in only about one-third
time,
made
several recorded
the
clinical
available.
of cases because
technique of choice for soft-tissue The reviewers were instructed
were asked to categorize
and, when
diagnoses
possible,
to provide
a
for benign lesions such as
negative
for accuracy,
predictive
sensitivity,
specificity,
and positive and
values.
reviewers.
The majority of all lesions were in the thigh, 1 7 (1 8%) in the leg, and remainder of the lesions were in the majority of malignant lesions (32/45, extremity. Benign cysts were almost joints
the reviewers
hematoma or abscess were also requested when it was possible to provide one. Data were tabulated for both reviews for each of the three examiners. The final diagnoses (lesions benign vs malignant)
were evaluated posttraumatic of symptoms
for
(n=50)
intensity
Homogeneous Majority of mass
50 fibrous
irregular
Benign
(n=45)
defined
Partially
1
Angiolipoma
Malignant
1990
Masses
Benign
Myxoma Thrombosed Thrombosed Epidermoid
December
history
CT scans
and
results
were
MR imaging
availis the
masses at our institution. to categorize cases according
to
specific criteria. The size and location of each lesion were by each radiologist. Other features included the margins of
Results
For lesion size, as measured on MR images, the average of the three measurements was used. The sizes of lesions are summarized in Table 2. No malignant lesions were less than 3 cm in diameter. On the other hand, 22% of benign lesions were less than 3 cm in diameter. Thirteen percent of malignant lesions were 3-5 cm; 87% were larger than 5 cm. On the surface this may seem important; however, 50% of benign lesions also were larger than 5 cm. Forty-four percent of benign lesions had well-defined or sharp margins (Table 2). Slight irregularity was described in 40% of benign masses. Only eight (1 6%) showed total or nearly complete irregularity of margins. Most of the tumor margin was irregular in 38 malignant lesions (85%). However, sharp margins were noted in seven cases (15%). At least some inhomogeneous signal intensity was noted in 43 malignant lesions (95%) (Table 2). Homogeneous signal intensity was unusual in malignant lesions. Small areas of inhomogeneity were present in 1 2 benign lesions (24%). Neurovascular involvement was not common in either benign or malignant lesions; however, some displacement of neurovascular structures occurred in 32 malignant lesions (71 Three malignant lesions (7%) had neurovascular encasement. This was also observed with desmoid tumors (which are considered benign). Hemorrhage or edema or both %).
AJR:155,
Downloaded from www.ajronline.org by 155.33.148.196 on 02/16/15 from IP address 155.33.148.196. Copyright ARRS. For personal use only; all rights reserved
were
December
present
MR
1990
in 27%
of benign
and
54%
OF
of malignant
SOFT-TISSUE
lesions.
The prevalence of hemorrhage and edema in benign lesions was due to the number of hematomas (9/50) included in this group of patients. Bone involvement was noted with malignant lesions and desmoid tumors. Accuracy of interpretation varied somewhat among the reviewers. There were also differences in diagnostic ability related to the type of lesions (benign vs malignant) and availability of clinical data. The latter was especially important for the less experienced radiologists. Accuracy in predicting whether lesions were benign or malignant ranged from a low of 88% in the less experienced group to 93% for the physician experienced in MR imaging (Table 3). Radiologists were able to predict the histology of the lesions in up to 26% of cases, mostly for benign lesions, when no history was available. When
a history
was
available,
the
two
less
experienced
TABLE Masses
4: Observer
Performance
Benign Sensitivity Specificity Positive
Negative Accuracy Malignant Sensitivity Specificity
predictive
Physician
value
(%)
value (%)
(%) (%) (%)
Positive predictive Negative predictive Accuracy (%) experienced
of Soft-Tissue
No. A
(%) (%) predictive
for Detection
Ob server
Type of Mass/ Statistical Measure
a
changed their diagnoses 1 3% and 19% of the time overall. The more experienced physician changed the diagnosis only once on the basis of historical information. The less experienced physicians predicted the nature of benign lesions 38-40% of the time without clinical data and
1253
MASSES
value (%) value (%)
i
2
3
verage
90 90 92 88 90
82 83 95 83 88
90 96 96 90 93
88 90 94 87 90
90
96
96
94
90 88 92
82 93 95
96 90 96
90 87 94
90
88
93
90
in MR imaging.
reviewers
50-52%
of the time
when
history
was
available.
The
physician
more experienced in MR imaging accurately predicted the nature of a benign lesion in 74% of cases. The reviewers were not successful in predicting the histology of malignant lesions. The sensitivity, specificity, positive predictive value, negative predictive value, and accuracy were calculated for benign and malignant lesions (Table 4). The data for malignant lesions are particularly important. The specificity for malignant lesions ranged from 82% to 96%, with a negative predictive value of 92-96% and a positive predictive value of 88-90%.
Discussion An
increasingly
common
indication
for
MR
imaging
is to
determine the presence or absence of musculoskeletal masses. When a mass is found in such an examination, the immediate challenge is to characterize it. Several authors [3, 6-9] have noted that clinical data, especially patient age, lesion location, and duration of symptoms, are helpful in addition to image features for characterizing lesions. Some authors [5, 7, 1 0-i 2] have reported, however, that differentiation of benign and malignant lesions is inconsistent when MR features are used. Manipulation of Ti and T2 relaxation times has not improved specificity [1 , 6].
TABLE
3: Accuracy
in Predicting No. (%) (n
Observer
No. Accuracy
Benign =
Lesions
of Diagnoses
95)
Exact Histology
Changed with History i3 1 ga 1
1
85(90)
i5(16)
2
84 (88)
25 (26)
3b
88 (93)
24 (25)
a Diagnosis interpretation b Physician
and Malignant
less accurate after history was changed. experienced in MA imaging.
in 1 1 % of
patients
when
initial
Our results provide support for using MR imaging to characterize soft-tissue masses. By using traditional radiologic criteria, such as homogeneity and boundary characteristics, experienced and inexperienced observers were able to identify malignant lesions with an average sensitivity of 94% in this series. The negative predictive value of an interpretation that a mass was not malignant also was high (94%). Classic benign and malignant lesion characteristics are illustrated in Figures 1 and 2, respectively. Benign lesions (Fig. 1) tend to be well marginated, have homogeneous signal intensity, and do not encase neurovascular structures or invade bone. Malignant lesions (Fig. 2) generally have irregular margins and inhomogeneous signal and more often encase neurovascular structures and involve bone [3, 6]. The benign lesions that most often were classified incorrectly as malignant in this series were desmoid tumors and necrotic benign neoplasms (neurofibromas). A single hemangioma was considered malignant by one observer (Fig. 3); this was a solid inhomogeneous cavernous hemangioma [13, 1 4]. The difficulty of correctly classifying desmoid tumors is not surprising in view of the infiltrative, locally aggressive nature of the lesions [3, 6]. Many of the benign lesions (ganglionic cysts, lipomas, hemangiomas, neuromas [because of location], and most hematomas) were diagnosed correctly on the basis of imaging features alone [1 3, 15-i 7]. When the MR images were interpreted with knowledge of the clinical findings, the specific histologic diagnosis for benign lesions could be predicted in 38-40% of cases by the inexperienced reviewers and in 74% by the more experienced MR imaging consultant. These percentages are higher than the 24% reported by Kransdorf et al. [18]. The malignant lesion most commonly misclassified as benign was synovial sarcoma. Three of the six synovial sarcomas in this group of lesions had smooth margins and a cystic appearance or mimicked hematomas (Fig. 4). Errors in diagnosis were made in more than half of the observations. The results show that historical data can be helpful in lesion classification. The accuracy improved for the less experienced reviewers when history was available. The more experienced
BERQUIST
1254
ET AL.
AJR:155,
December
1990
Fig. 1.-Benign neuroma. A, Axial Ti-weighted MR image (SE 500/20) shows a well-marginated homogeneous lowintensity mass. B, Axial T2-weighted MR image (SE 2000/60)
shows uniform high signal intensity. Margination
Downloaded from www.ajronline.org by 155.33.148.196 on 02/16/15 from IP address 155.33.148.196. Copyright ARRS. For personal use only; all rights reserved
and signal intensity are typical of a benign lesion. Change in signal intensity between the two images indicates lesion is not a lipoma.
Fig. 2.-Fibrosarcoma of forearm. A, Axial Ti-weighted MR image (SE 500/20) shows mass with mixed or inhomogeneous signal intensity. B, Axial
T2-weighted
MR image
(SE
2000/60)
shows mass has irregular margins and inhomogeneous signal intensity. These features are typical of a malignant mass and appreciated more easily on T2-weighted image.
A, Axial Ti-weighted MR image (SE 500/20) shows scattered areas of increased signal intensity due to blood. B, Axial T2-weighted MR image shows areas of high and low intensity (arrowheads) due to paramagnetic effects of blood. Inhomogeneous signal intensity and irregular margins may cause
Fig. 4.-Synovial sarcoma in gluteal region. Coronal SE 500/20 MR image of pelvis shows wellmarginated cystic lesion with appearance similar to that of hematoma. Three of six synovial sarcomas were considered benign because of this ap-
this lesion
pearance.
Fig. 3.-Cavernous
hemangioma
to be confused
in proximal
with a malignant
forearm.
mass.
AJA:155,
December
MR
1990
OF
SOFT-TISSUE
Downloaded from www.ajronline.org by 155.33.148.196 on 02/16/15 from IP address 155.33.148.196. Copyright ARRS. For personal use only; all rights reserved
MR imaging reviewer relied on history minimally, and overall specificity of 96% for all lesions did not change tween the first and second portions of the study.
the be-
REFERENCES 1 . Beltran J, Simon DC, Katz W, Weis LD. Increased MR signal intensity in skeletal muscle adjacent to malignant tumors: pathologic correlation and clinical
relevance.
Radiology
1987;1 62:251-255
2. Bland KI, McCoy DM, Kinard RE, Copeland EM III. Application of magnetic resonance imaging and computerized tomography as an adjunct to the surgical management of soft tissue sarcomas. Ann Surg 1987;205: 473-481
3. Berquist TH. Magnetic resonance imaging of musculoskeletal
neoplasms.
Clin Orthop i989;244:101-118
4. Sundaram M, McGuire MH, Herbold DR, Beshany SE, Fletcher JW. High signal intensity soft tissue masses on Ti weighted pulsing sequences. Skeletal Radio! 1987;16:30-36 5. Sundaram M, McGuire MH, Herbold DR. Magnetic resonance imaging of soft tissue masses: an evaluation of fifty-three histologically proven tumors. Magn Reson Imaging i988;6:237-248 6. Berquist TH, Ehman AL, Richardson ML. Magnetic resonance of the musculoskeletal system. New York: Raven, 1987:85-108 7. Dooms GC, Hricak H, Sollitto RA, Higgins CB. Lipomatous tumors and tumors with fatty component: MR imaging potential and comparison of
MASSES
1255
MA and T results. Radiology 1985;157:479-483 8. Weekes AG, Berquist TH, McLeod AA, Zimmer WD. Magnetic resonance imaging of soft-tissue Magn Reson Imaging
tumors:
comparison
with
computed
tomography.
1985;3:345-352 9. Weekes AG, McLeod RA, Reiman HM, Pritchard DJ. CT of soft-tissue neoplasms. AJR 1985;144:355-360 10. Petasnick JP, Turner DA, charters JR. Gitelis 5, Zachanas CE. Soft-tissue masses of the locomotor system: comparison of MR imaging and CT. Radiology
1986;160: 125-133 M, McGuire MH, Schajowicz F. Soft-tissue masses: histologic basis for decreased signal (short T2) on T2-weighted MR images. AiR 1987;148: 1247-1 250 12. Totty WG, Murphy WA, Lee JKT. Soft-tissue tumors: MA imaging. Radio!ogy 1986;160:135-141
1 1 . Sundaram
13. Cohen EK, Kressel HY, Perosio T, et al. MR imaging of soft-tissue hemangiomas: correlation with pathologic findings. AJR 1988;150: 10791081 14. Cohen JM, Weinreb JC, Redman H. Arteriovenous malformations of the extremities:
MR imaging.
Radiology
1986;158:475-479
15. Cohen MD, McGuire W, Cory DA, Smith JA. MR appearance 16.
of blood and blood products: an in vitro study. AiR 1986;146: 1293-1 297 Ehman AL, Berquist TH. Magnetic resonance imaging of musculoskeletal trauma. Radio! Clin North Am 1986;24(2):291-319
17. Swensen SJ, Keller PL, Berquist TH, McLeod RA, Stephens DH. Magnetic resonance
imaging
of hemorrhage.
AiR
1985;145:921-927
18. Kransdorf MJ, Jelinek JS, Moser AP Jr, et al. Soft-tissue using MR imaging. AJR 1989;153:541-547
masses: diagnosis