Journal of Orthopaedic Surgery 2016;24(2):216-21
A weighted scoring system to differentiate
malignant liposarcomas from benign lipomas
Shiyao Wang,1,2 Lester Wai Mon Chan,2 Xiaodong Tang,1 Chang Su,3 Chunfang Zhang,4 Kunkun Sun,5 Danhua
Shen,5 Hao Chen,6 Wei Guo1
Musculoskeletal Tumour Centre of Peking University People’s Hospital, China
Department of Orthopaedic Surgery, Tan Tock Seng Hospital, Singapore
3
Clinical Research Unit, Khoo Teck Pual Hospital, Singapore
4
Department of Statistics, Peking University People’s Hospital, China
5
Department of Pathology, Peking University People’s Hospital, China
6
Department of Radiology, Peking University People’s Hospital, China
1
2
ABSTRACT
Purpose. To construct a scoring system to differentiate
malignant liposarcomas from benign lipomas by
comparing their clinical and magnetic resonance
imaging (MRI) features.
Methods. Clinical and MRI features of 33 women and
33 men aged 17 to 83 (mean, 53) years who underwent
resection of malignant liposarcomas (n=32) or benign
lipomas (n=34) were reviewed.
Results. The 5 strongest predictors of liposarcoma
were male gender, larger tumour maximum
dimension, deep to fascia, thick non-fatty septum
or nodule, and internal cystic change. A weighted
scoring system was constructed using the 5 strongest
predictors as: Z score=10X1+X2+12X3+15X4+10X5,
respectively. A cut-off score of 35 was used; all 32
malignant liposarcomas and 4 of 34 benign lipomas
scored >35. The cut-off score of ≤35 could predict 30
of 66 lipomatous tumours as benign with a negative
predictive value of 100% (p<0.0001).
Conclusion. The 5 strongest clinical and MRI features
were identified to construct a scoring system to
differentiate malignant from benign lipomatous
tumours. Further validation in independent
populations is required.
Key words: lipoma; liposarcoma
INTRODUCTION
Lipomatous tumours are common mesenchymal
neoplasms of lipogenic differentiation and can be
benign or malignant with a wide range of histological
subtypes.1,2 Malignant liposarcomas necessitate wide
excision with chemotherapy and/or radiotherapy,
whereas benign lipomas can be treated with marginal
excision or observation.3,4 Borderline, low-grade
liposarcomas/atypical lipomatous tumours do not
appear to metastasise but have an increased risk of
local recurrence and a potential for dedifferentiation;
these should be treated as malignant tumours.
The National Institute for Health and Care
Excellence (NICE) guidelines recommend that any
tumour deep to fascia, >5 cm, rapidly increasing in
Address correspondence and reprint requests to: Dr Wei Guo, Musculoskeletal Tumour Centre of Peking University People’s
Hospital, China, 100044. Email: [email protected]
Vol. 24 No. 2, August 2016
A weighted scoring system to differentiate malignant liposarcomas from benign lipomas 217
size, or painful in a previously painless lump should
be regarded as malignant until proven otherwise.5
Magnetic resonance imaging (MRI) can differentiate
malignant from benign lipomatous tumours,5–8
particularly atypical lipomatous tumour/welldifferentiated liposarcoma from benign lipoma.9–12
(a)
(b)
(c)
Figure 1 Magnetic resonance imaging showing (a) a non-fatty nodule (≥1 cm) in a well-differentiated liposarcoma (asterisk),
(b) a non-fatty septum (>2 mm) in a well-differentiated liposarcoma (arrow), and (c) internal cystic change in a myxoid
liposarcoma.
Journal of Orthopaedic Surgery
218 S Wang et al.
This study aimed to construct a scoring system to
differentiate malignant liposarcomas from benign
lipomas by comparing their clinical and MRI features.
MATERIALS AND METHODS
This study was approved by the ethics committee of
our hospitals. Records of 33 women and 33 men aged
17 to 83 (mean, 53) years who underwent resection
of malignant liposarcomas (n=32) or benign lipomas
(n=34) from 2004 to 2012 were retrieved. Patients
with recurrent lipomatous tumour or inadequate
MRI were excluded. Histological diagnosis was made
by consensus of 2 pathologists based on the World
Health Organization criteria.13 Clinical features were
reviewed by an orthopaedic oncology surgeon, and
MRI features were reviewed by a radiologist.
MRI features included the margin (well-defined
vs. partially ill-defined/ill-defined), the homogeneity
of signal intensity (homogeneous vs. inhomogeneous),
bright signal intensity on T1-weighted fast spin echo
images (entire lesion vs. partial/none), bright signal
intensity on T2-weighted fast spin echo with fat
suppression images (entire/partial lesion vs. none),
non-fatty septum or nodule (no/thin vs. thick nonfatty septum [>2 mm] or nodule [>1cm]), internal
cystic change (yes vs. no), non-fatty content (>50% vs.
≤50%), no fat signal (yes vs. no) [Fig. 1].
Patients with liposarcoma were compared with
those with lipoma using the unpaired Student’s t
test for normally distributed continuous variables, or
the Fisher’s exact test for categorical variables. A p
value of <0.05 was considered statistically significant.
In a univariate analysis, significant variables (p<0.1)
were entered into a multiple logistic regression model
using a backward selection method to determine the
independent predictors. An odds ratios of >1 and
<1 indicated an increased and decreased odds of
liposarcoma, respectively.
The strongest predictors were chosen to
construct the scoring system. The weighting of the
predictors was determined by the coefficient (B) as: Z
score=B1X1+…+BkXk. Before that, the discriminating
power of the chosen score model was estimated and
a receiver operating characteristic (ROC) curve was
plotted. The ROC curve indicated the relationship
of sensitivity and specificity (i.e. the cut-off score for
differentiating malignant from benign lipomatous
tumours). An area under the curve (AUC) of 1.0
indicated a perfect discrimination, whereas an AUC
of 0.5 indicated no discrimination.14 Another ROC
curve was plotted to determine the cut-off for tumour
size with maximum sensitivity and specificity using
Table 1
Univariate analysis of variables between liposarcoma and
lipoma
Variable
Gender
Male
Female
Age (years)
Location
Extremity
Trunk
Depth
Subcutaneous
Deep to fascia
Maximum dimension (cm)
Margin
Well-defined
Partially ill-defined/ill-defined
Homogeneity of signal intensity
Homogeneous
Inhomogeneous
Bright signal intensity on T1weighted fast spin echo images
Entire lesion
Partial or none
Bright signal intensity on T2weighted fast spin echo with fat
suppression images
None
Entire lesion or partial
Non-fatty septum or nodule
No or thin septum or nodule
<1 cm
Thick/nodular ≥1 cm septum
Internal cystic change
None
Present
Non-fatty content
≤50%
>50%
No fat signal
No
Yes
Lipo- Lipoma
sarcoma (n=34)
(n=32)
24
8
56.53±
16.21
9
25
50.79±
11.73
24
8
21
13
1
31
14.72±
5.50
9
25
8.00±
3.48
21
11
27
7
0
32
23
11
0
32
23
11
0
32
19
15
3
25
29
9
19
13
32
2
25
7
31
3
24
8
34
0
p
Value
<0.001
0.107
0.297
0.013
0.014
0.272
<0.001
<0.001
<0.001
<0.001
0.001
0.18
0.005
the Youden index15 (i.e. sensitivity+specificity-1).
RESULTS
Of the 32 liposarcomas, 16 were atypical lipomatous
tumours/well-differentiated liposarcomas, 10 were
myxoid cell liposarcomas, one was pleomorphic
liposarcoma, one was dedifferentiated liposarcoma,
and 4 were mixed-type liposarcomas (3 pleomorphic/
myxoid and one well-differentiated/myxoid
liposarcomas). Of the 34 lipomas, 16 were typical
Vol. 24 No. 2, August 2016
A weighted scoring system to differentiate malignant liposarcomas from benign lipomas 219
Table 2
Logistic regression analysis for 5 strongest predictors after backward selection
B (95% CI)
OR (95% CI)
p Value
2.40 (-0.26–5.08)
0.33 (0.01–0.67)
3.90 (0.23–7.57)
4.87 (1.64–8.10)
3.63 (0.37–6.90)
11.11 (0.76–161.52)
1.40 (1.01–1.95)
49.68 (1.26–1945.56)
130.67 (5.17–3296.79)
37.98 (1.44–995.36)
0.078
0.045
0.037
0.003
0.029
Variable
Male gender
Tumour maximum dimension in cm
Tumour deep to fascia
Thick non-fatty septum (>2 mm) or nodule (>1 cm)
Internal cystic change
lipomas, 12 were intramuscular or intermuscular
lipomas, and 6 were angiolipomas.
Patients with liposarcoma differed to those with
lipoma in terms of gender, tumour depth, tumour
maximum dimension, homogeneity of signal
intensity, bright signal intensity on T1-weighted
fast spin echo images, bright signal intensity on T2weighted fast spin echo with fat suppression images,
non-fatty septum or nodule, internal cystic change,
and no fat signal (Table 1). The 5 strongest predictors
of liposarcoma were male gender, larger tumour
maximum dimension, deep to fascia, thick non-fatty
septum or nodule, and internal cystic change. They
were entered in a multiple logistic regression analysis
(Table 2) for construction of a weighted scoring
1.0
0.8
0.6
Sensitivity
AUC=0.980
0.4
0.2
system to exclude malignant lipomatous tumours.
An ROC curve was plotted to determine the
sensitivity and specificity of the scoring system.
The AUC was 0.980 indicating strong power for
the scoring system to discriminate malignant from
benign lipomatous tumours (Fig. 2). Another ROC
curve was plotted to determine the sensitivity and
specificity of tumour maximum dimension. The
maximum Youden index was 0.669 with sensitivity of
0.875 and specificity of 0.794, and its corresponding
tumour maximum dimension was 10.02 cm (Fig. 3). A
cut-off of 10 cm was used, but the tumour maximum
dimension as a continuous variable was found to be
more accurate.
The scoring system was constructed using the
5 strongest predictors as: Z score=2.40X1+0.33X2+
3.90X3+4.87X4+3.63X5, where X1 denotes gender
(1=male / 0=female), X2 tumour maximum
dimension in cm, X3 tumour depth (1=deep to fascia
/ 0=subcutaneous), X4 non-fatty septum or nodule
(1=thick septum [>2 mm] or nodule [≥1 cm] / 0=no
or thin), and X5 internal cystic change (1=yes /
0=no). The coefficients were modified to be integerfriendly and weighting was added to corresponding
predictors without changing the diagnosis outcome
as: Z score=10X1+X2+12X3+15X4+10X5. The weighted
score of all tumours was plotted and a cut-off score
of 35 was used; all 32 malignant lipomatous tumours
and 4 of 34 benign lipomatous tumours scored >35
(Fig. 4). The cut-off score of ≤35 could predict 30 of
66 lipomatous tumours as benign with a negative
predictive value of 100% (p<0.0001).
DISCUSSION
0
0
0.2
0.4
0.6
1-Specificity
0.8
1.0
Figure 2 A receiver operating characteristic curve is plotted
to determine the sensitivity and specificity of the scoring
system to differentiate malignant from benign lipomatous
tumour. The area under the curve (AUC) of 0.980 indicates
strong discriminating power.
To plan the appropriate resection (marginal or
wide), a biopsy should be performed to establish the
diagnosis. Nonetheless, biopsy is time- and resourceconsuming and may generate anxiety. The biopsied
tissue may be non-representative and the result
may be false positive or negative. If a diagnosis can
be made based on clinical and imaging findings, a
Journal of Orthopaedic Surgery
220 S Wang et al.
(a)
(b)
1.0
0.8
Size=10.02, maximum Youden index=0.669
0.7
0.8
0.6
Youden index
Sensitivity
0.6
0.4
0.5
0.4
0.3
0.2
0.2
0
0
0
0.2
0.4
0.6
1-Specificity
0.8
1.0
2.70
4.49
4.94
5.47
6.12
7.00
7.68
8.04
8.87
9.00
9.91
10.36
10.63
10.89
11.64
12.76
13.75
15.05
15.88
17.58
22.40
25.50
0.1
Size of lesion (cm)
Figure 3 (a) A receiver operating characteristic curve is plotted to determine the sensitivity and specificity of tumour maximum
dimension. (b) The Youden index corresponds to every tumour maximum dimension. The maximum Youden index is 0.669 with
sensitivity of 0.875 and specificity of 0.794, and its corresponding tumour maximum dimension is 10.02 cm.
80
70
Z score
60
Malignant
50
40
Cut-off: 35
30
20
Benign
10
0
1 2 3 4 5 6 7 8 910111213141516171819202122232425262728293031323334
Patient sequence
Figure 4 The weighted score line for malignant (above)
and benign (below) lipomatous tumours is plotted. A cutoff score of 35 is used; all 32 malignant liposarcomas and
4 of 34 benign lipomas scored >35. The cut-off score of ≤35
can predict 30 of 66 lipomatous tumours as benign with a
negative predictive value of 100% (p<0.0001).
biopsy is not required.
Many studies have reported correlation
between imaging findings and pathological
diagnosis.1,3,4,7,8,10,11,16 In our study, the tumour
maximum dimension rather than multiple
dimensions or total volume was used; its contribution
to the scoring system was weak compared with other
variables, although the NICE guideline recommends
that a tumour of >5 cm should be regarded as
malignant until proven otherwise. In our study using
the Youden index, a cut-off for tumour size of 10 cm
resulted in a slightly lower sensitivity and much high
specificity for malignancy.17,18 Tumour size was more
accurate as a continuous variable than as a categorical
variable.
Many MRI features are related to each other. A
lesion with homogenous signal intensity throughout
does not have a thick septum or nodule. Our primary
aim was to exclude malignancy; the scoring system
should identify as many benign tumours as possible
(high specificity) with a low false negative rate (high
negative predictive value). Sensitivity was a secondary
concern as the scoring system is intended to rule out
rather than rule in malignancy. The absence of a nonfatty septum or nodule and internal cystic change
had the highest specificity for benign tumour and
good negative predictive value while maintaining
reasonable sensitivity. Inhomogeneous signal
intensity and the absence of bright signal intensity on
T1-weighted fast spin echo images had the highest
sensitivity for liposarcoma, but their specificity and
negative predictive values were inferior to those of a
thick septum or nodule and cystic degeneration.
Vol. 24 No. 2, August 2016
A weighted scoring system to differentiate malignant liposarcomas from benign lipomas 221
Thick non-fatty septum or nodule was present
in the myxoid, pleomorphic, dedifferentiated, and
mixed-type liposarcomas. This feature was also
present in 9 of 34 benign lipomas in our study, as the
lipomas may contain areas of fat necrosis or fibrosis.10
Internal cystic change is a typical feature of myxoid
liposarcoma; nonetheless benign angiolipoma may
also display the similar feature.
In a study of a scoring system to differentiate
lipomas from atypical lipomatous tumours based
on tumour diameter, depth, MRI septa, and MRI
enhancement,12 the sample size of 12 atypical
lipomatous tumours and 48 lipomas was too small
to allow multivariate analysis or weighting. No
factor was significant in the univariate analysis. In a
study of an algorithm for general surgeons to decide
when to refer a patient with lipomatous tumour to
a specialist,19 factors included were patient age ≥55
years, tumour size ≥10 cm, extremity location, and
previous resection; MRI features were not taken into
account.
Limitations of our study were its retrospective
nature and inclusion of a single ethnic population.
CONCLUSION
The 5 strongest clinical and MRI features were
identified to construct a scoring system to differentiate
malignant from benign lipomatous tumours. Further
validation in independent populations is required.
DISCLOSURE
No conflicts of interest were declared by the authors.
REFERENCES
1. Drevelegas A, Pilavaki M, Chourmouzi D. Lipomatous tumours of soft tissue: MR appearance with histological correlation.
Eur J Radiol 2004;50:257–67.
2. Dei Tos AP. Liposarcomas: diagnostic pitfalls and new insights. Histopathology. 2014;64:38–52.
3. Murphey MD, Carroll JF, Flemming DJ, Pope TL, Gannon FH, Kransdorf MJ. From the archives of the AFIP: benign
musculoskeletal lipomatous lesions. Radiographics 2004;24:1433–66.
4. Murphey MD, Arcara LK, Fanburg-Smith J. From the archives of the AFIP: imaging of musculoskeletal liposarcoma with
radiologic-pathologic correlation. Radiographics 2005;25:1371–95.
5. http://www.ncin.org.uk/publications/data_briefings/liposarcoma_incidence_and_survival_rates_in_england
6. El Ouni F, Jemni H, Trabelsi A, Ben Maitig M, Arifa N, Ben Rhouma K, et al. Liposarcoma of the extremities: MR imaging
features and their correlation with pathologic data. Orthop Traumatol Surg Res 2010;96:876–83.
7. Kransdorf MJ, Moser RP Jr, Meis JM, Meyer CA. Fat-containing soft-tissue masses of the extremities. Radiographics
1991;11:81–106.
8. Sung MS, Kang HS, Suh JS, Lee JH, Park JM, Kim JY, et al. Myxoidliposarcoma: appearance at MR imaging with histologic
correlation. Radiographics 2000;20:1007–19.
9. Galant J, Martí-Bonmatí L, Sáez F, Soler R, Alcalá-Santaella R, Navarro M. The value of fat-suppressed T2 or STIR sequences
in distinguishing lipoma from well-differentiated liposarcoma. EurRadiol 2003;13:337–43.
10. Brisson M, Kashima T, Delaney D, Tirabosco R, Clarke A, Cro S, et al. MRI characteristics of lipoma and atypical lipomatous
tumour/well-differentiated liposarcoma: retrospective comparison with histology and MDM2 gene amplification. Skeletal
Radiol 2013;42:635–47.
11. Jaovisidha S, Suvikapakornkul Y, Woratanarat P, Subhadrabandhu T, Nartthanarung A, Siriwongpairat P. MR imaging of fatcontaining tumours: the distinction between lipoma and liposarcoma. Singapore Med J 2010;51:418–23.
12. Nagano S, Yokouchi M, Setoguchi T, Ishidou Y, Sasaki H, Shimada H, et al. Differentiation of lipoma and atypical lipomatous
tumour by a scoring system: implication of increased vascularity on pathogenesis of liposarcoma. BMC Musculoskelet
Disord 2015;16:36.
13. Fletcher CD, Bridge JA, Hogendoorn PC, Mertens F. WHO classification of tumours of soft tissue and bone. Pathology and
genetics of tumours of soft tissue and bone. 4th ed. Lyon: IARC Press; 2013.
14. Apfel CC, Greim CA, Haubitz I, Goepfert C, Usadel J, Sefrin P, et al. A risk score to predict the probability of postoperative
vomiting in adults. Acta Anaesthesiol Scand 1998;42:495–501.
15. Youden WJ. Index for rating diagnostic tests. Cancer 1950;3:32–5.
16. Peterson JJ, Kransdorf MJ, Bancroft LW, O’Connor MI. Malignant fatty tumours: classification, clinical course, imaging
appearance and treatment. Skeletal Radiol 2003;32:493–503.
17. Gelineck J, Keller J, Jensen OM, Nielsen OS, Christensen T. Evaluation of lipomatous soft tissue tumours by MR imaging.
Acta Radiol 1994;35:367–70.
18. Datir A, James SL, Ali K, Lee J, Ahmad M, Saifuddin A. MRI of soft-tissue masses: the relationship between lesion size,
depth, and diagnosis. Clin Radiol 2008;63:373–8.
19. Fisher SB, Baxter KJ, Staley CA 3rd, Fisher KE, Monson DK, Murray DR, et al. The general surgeon’s quandary: atypical
lipomatous tumour vs lipoma, who needs a surgical oncologist? J Am Coll Surg 2013;217:881–8.