F Zerrin Mary
Yetkin, Fischer,
E
MD MS
#{149} Victor
M. Haughton, M. Rao, PhD
#{149} Stephen
Multiple Sclerosis: for Diagnosis’ The specificity of magnetic resonance (MR) imaging in the diagnosis of multiple sclerosis (MS) has not been measured systematically. Conventional MR head images with sagittal localizer and axial multipleecho sequences with long repetition times were obtained in 92 patients with clinically verified MS (Schumacher criteria), 100 healthy volunteers, 60 subjects with hypertension, and eight patients with dementia. Two readers, without the aid of any clinical or demographic information, classified each of the 260 studies as MS or not MS. The readers classified the studies again after being supplied with the subjects’ ages and sex. True-negative and truepositive diagnoses of MS were tabulated. The specificity of the MR diagnosis of MS (true-negative results in proportion to all non-MS studies) was 95%-99% with all the control groups included. There is a small risk of misinterpreting incidental periventnicular white matter foci as plaques of MS in MR studies. Index
terms:
Sclerosis,
Brain,
multiple,
Radiology
MR
studies,
#{149} R.
Anne
Papke,
BSN
Specificity
M
resonance is acknowledged
(MR) imagas the
AGNETIC
ing
most sensitive imaging test for the detection of demyelinating plaques in the cerebral white matter. Its specificity is said to be unknown (i). Peniventriculan foci of high signal intensity, so-called unidentified bright objects on leukoaniosis, which may mesemble the plaques of multiple sclerosis (MS) in MR images, are incidental findings in up to 20% of patients or volunteers. Radiologists, using various criteria, distinguish unidentified bright objects and MS plaques in their reports. The accuracy of nadiologists’ differentiation between mcidental unidentified bright objects and plaques of MS has not been studied extensively. Therefore, we determined the specificity of the MR diagnosis of MS.
MATERIALS Specificity of blinded patients
AND
control
130.1214.
178:447-451
abnormalities
volunteers, From the Departments V.M.H., RAP.), Biostatistics
of Radiology (M.E.F.),
rology sin,
(S.M.R.), Froedtert
the
Medical
Memorial
College
Lutheran
(F.Z.Y., and Neuof Wiscon-
Hospital.
by
neurora-
chosen.
gust
20.
RO1
tutes
Supported
of Health.
Address
grants
from
RO1
the reprint
NS22913-03
National requests
Instito
1991
hypertension,
regarding
present
health
MS
and
medical history. No physical, laboratory, or neurologic examinations were performed. Any volunteer that admitted to neurologic or cardiac problems or to any present illness was excluded from the study.
V.M.H. © RSNA,
by
NS22128-05
with
A consecutive
60 patients
with
unselected
hypertension
recruited
from
MR
the
dementia
54-90 years. All evaluation, and a had been made.
Imaging
Reading
and patients with dementia. One hundred volunteers, selected to match the patients in age and sex, were questioned in detail
9200 W Wisconsin Ave. Milwaukee, WI 53226. Received June 13, 1990; revision requested July 13; revision received August 13; accepted Auand
patients
were
clinic. They were aged had undergone careful diagnosis of dementia
on the basis images of both
Ninety-two patients with the diagnosis of MS were selected from patients in the neurology service. In every case, Schumacher criteria for MS were satisfied. The patients ranged in age from 31 to 83 years. A control group consisted of healthy 1
disease
METHODS
subjects
were
out neurologic symptoms was recruited from the hypertension clinic by means of advertising in the hospital newsletter. Patients who weighed more than 250 lb (112.5 kg) or had a history of neurologic disease were excluded. The hypertensive patients were aged 26-75 years. Eight patients with the diagnosis of Alzheimer
All images were obtained on a 1.5-T imager with the quadrature head coil (Signa; GE Medical Systems, Milwaukee). A sagittal localizer image was obtained with a pulse sequence of 600/20 (repetition time msec/echo time msec). A series of 42 axial sections was then obtained with a section thickness of 5 mm; pulse sequences of 2,000-3,000 (depending on the heart rate)/20, 80; one signal averaged; a field of view of 20 cm; 1-mm skip; and cardiac gating.
was measured readings of MR
and
#{149}
of MR
diologists with more than 2 years of expenience with MR imaging. For the patient group, subjects with clinically verified MS were chosen; for the control group, subjects who were not likely to have MS but might have incidental white matter
130.871
1991;
MD
series
but
with-
of Images
On each image all alphanumeric information, including name, age, sex, date of examination, and time of examination, that might permit the identification of the subject was concealed with black tape. The imaging studies from the patient group and the control group were placed in randomized sequence, and each study was given a code number. The study coordinator presented the images to the readers without any clinical or demographic information and asked the readers to determine whether the study was positive or negative for MS and to give the diagnosis as exactly as possible. The readers were then given the age and sex of the subject and asked again to determine whether the study was positive or negative for MS and to give the diagnosis as exactly as possible. The readers were not told the proportion of patients to control subjects or the types of controls included. Each reader read independently.
of Abbreviation:
MS
=
multiple
sclerosis.
447
Two experienced neuroradiologists using standard MR criteria classified abnormalities of white matter as MS plaques or incidental peniventnicular abnormalities. The readers evaluated the size, shape, number, and location of lesions pnimarily. They may have considered the appearance of the corpus callosum and brain stem, although they were not instructed to do so. MS cases were typically considered to have multiple oblong or ovoid foci of high signal intensity abutting the ventricles.
Statistical
Table 1 MR Findings:
All MS Patients “Subtle/Suspect
Reader
Sensitivity
Subjects
(Patients
and
MS”
(%)
Not MS
Specificity
“Subtle/Suspect
(%)
Sensitivity
MS”
(%)
=
MS
Specificity
(%)
1
76.09
96.39
80.43
96.39
2
68.48
98.81
77.17
95.24
Table 2 MR Findings:
All MS Patients
Methods
“Subtle/Suspect
Readings were classified as positive or negative, and the numbers of true- and false-positive and true- and false-negative results were tabulated. Statistical analyses were performed once with the equivocal findings (“possible MS” or “suggestive of MS”) classified as negative and again with them classified as positive. Specificity, the proportion of subjects without MS whose images were classified as negative, was determined. The specificity was determined for various control groups, including (a) all subjects without MS (ie, all volunteers and hypertensive and demented patients), (b) all subjects with disease but without MS (ie, the hypertensive and demented patients), (c) all the healthy volunteers, (d) all the control subjects aged more than 40 years, (e) all the control subjects with disease who were aged 30-40 years, and (f) all the control subjects that had focal abnormalities (unidentified bright objects) in the white matter. Agreement between readers was evaluated with the McNemar x2 test and the kappa statistic.
All Control
versus
Volunteers)
Reader
Sensitivity
versus MS”
(%)
All Not
Control
MS
Specificity
Subjects
with
Disease
“Subtle/Suspect
(%)
Sensitivity
(%)
MS”
=
MS
Specificity
(%)
1
76.09
95.45
80.43
95.45
2
68.48
98.53
77.17
94.12
Table 3 MR Findings:
All MS Patients “Subtle/Suspect
Reader
Sensitivity
All Volunteers
versus MS”
(%)
=
Not MS
Specificity
“Subtle/Suspect
(%)
Sensitivity
(%)
MS”
=
MS
Specificity
(%)
1
76.09
96.97
80.43
96.97
2
68.48
98.99
77.17
95.96
Table 4 MR Findings: Aged
More
than
MS Patients 40 Years “Subtle/Suspect
Reader
Aged MS”
(%)
Sensitivity
1
70.69
2
65.52
More
than 40 Years versus
Not MS
“Subtle/Suspect
(%)
Specificity
Control
98.31 100.00
Sensitivity
(%)
Subjects MS”
MS
Specificity
(%)
77.59
98.31
75.86
98.61
RESULTS A total
of 520 readings
for 260 pen-
sons were analyzed; two MR studies judged by the readers to be suboptimal were excluded from the study. Included were images of 92 MS patients, 60 hypertensive subjects, 100 volunteers, and eight patients with a diagnosis of dementia of the Alzheimer type. Foci of high signal intensity were present in the white matter of 40% of the hypertensive patients, 35% of volunteers, 90% of Alzheimer patients, and 90% of MS patients. Specificities for the blinded readers for various subgroups are in Tables 1-6. A group of control subjects with disease was defined as the hypertensive and demented patients who underwent imaging for this study; the presumably healthy subjects are the volunteers who under-
went
imaging.
diagnosis subjects, unteers,
448
The
specificity
of the
of MS when all control including patients and volwere included was 95%-99%
Radiology
#{149}
Table 5 MR Findings Aged
30-40
MS Patients Years “Subtle/Suspect
Reader
Sensitivity
(%)
Aged
30-40
MS”
Years versus
Not
Specificity
MS
Control
Subjects
“Subtle/Suspect
(%)
Sensitivity
(%)
with MS”
Disease =
MS
Specificity
1
82.76
100.00
82.76
100.00
2
72.41
100.00
75.86
iOO.00
(Table cases
i), read
depending as subtle
were
classified
on whether the or suspect MS
as positive
on negative
for MS. When data for the healthy volunteers were eliminated and only those of the hypertensive and demented patients were used, the specificity was 94%-99% (Table 2). If only data for the healthy volunteers were used, the specificity was 96%-99% (Table 3). For each of these calculations, readings performed without age or sex information were used. The sensitivity of the diagnosis of MS ranged from 68% (when the find-
(%)
ings of cases read as suspected MS were considered negative) to 80% (when these findings were considened positive). When the demognaphic information was supplied to readens, the sensitivity improved to 80% (suspected MS classified as normal) and 83% (suspected MS classified as MS). Within the subgroups included in the study, specificity was 81%-iOO%. If the MS patient and control groups were limited to persons aged more than 40 years, the specificity was 98%-iOO% whereas sensitivity was
February
1991
Table 6 MR Findings: Abnormalities
All MS Patients “Subtle/Suspect
Reader 1 2
Table 7 False-Positive
All Control
MS”
(%)
Sensitivity
versus
Not
MS
Specificity
76.09 68.48
83.78 95.35
Readings
of MS in Subjects Positive
Classification Control
of
Subjects
Sex, Age (y)
Sex and Age Withheld (n 13)
Subjects
(%)
without
Known
83.78 81.40
Neurologic
Sex and Age Known (n 7)
Healthy
F, 44
2
2
Healthy
F, 40
2
2
F, 76
2
0
Hypertension
F, 54
1
1
Alzheimerdisease
M, 79
Hypertension
F, 70
1 i
0 0
Hypertension
F, 75
1
0
Healthy
F, 56
1
0
Large, confluent periatrial lesions(Fig la-id) Slightly hyperintense peniventricular lesion Multiple periventricular lesions (Fig li-il) Periventnicular disease, little cortical atrophy Multiple peniventricular
were read as positive one reader and only had no information
sex.
Theme
readings among
Volume
178
Number
#{149}
2
le-lh)
Dawson fingers simulated Multiple periventricular lesions Cerebellar and periventricular lesions White matter lesions
were for
the
for MS by only when the reader about age and
i3 false-positive
MS in control subjects two readers when age
sex information
was
withheld
and
and
seven false-positive readings when age and sex were known. Readers were able to identify some cases mimicking MS as true-negative mesuits with the criteria used (Fig 2a2f).
For all MS patients and control subjects, the agreement between readers was excellent (P .78 for McNemam test, K = 0.88). False-positive readings, listed in Table 7, were made for nine subjects. Three MR studies of healthy volunteers were read as positive for MS by both readers when they had no demographic information and again when they were supplied the age and sex (Fig ia-id, ieih). One set of images for a 54-yearold hypertensive woman was read as positive for MS by one reader, both without and with the demographic information (Fig u-il). Images for one 76-year-old demented patient were read as positive for MS by both readers when the images were interpreted blindly but were read as negative for MS when the age and sex were known. Four additional studies
Disease
Comments
lesions(Fig
x2
(%)
Specificity
Ratings
2
6).
MS
80.43 77.17
2
66%-78% (Table 4). If the MS patients and controls were limited to those aged 30-40 years and the control group consisted of only the persons with disease (n = 7), the specificity was 100% and the sensitivity was 72%-83% (Table 5). If the control group was limited to the volunteers (n 35), hypertensive patients (n 25), and Alzheimer patients (n = 7) who had abnormal white matter foci, the specificity was 8i%-95% (Table
Matter
MS”
(%)
Sensitivity
F, 40
disease
White
“Subtle/Suspect
Healthy
Alzheimer
with
DISCUSSION The purpose of this study was to estimate the specificity of MR imaging in the diagnosis of MS, that is, the probability that the MR images of healthy subjects will be read as negative for MS. Ideally, a prospective study of a clinic population would provide the best estimate of the accuracy of MR imaging. Problems in yenifying the normal and diseased cases in such a population negated such a study. Therefore, as an alternative,
we selected
a group
of clinically
yen-
fied MS patients for our proved cases and healthy volunteers as our control subjects. Hypertensive or demented persons who were unlikely to have MS but likely to have white matter abnormalities at MR imaging were
also included. Persons with symptoms mimicking MS were excluded because of the difficulty of confirming the diagnosis. The patients we selected may have had more advanced findings than a group of patients meferned for MR imaging as part of an MS evaluation. The control group was not designed for the purpose of measuring misdiagnoses or misciassifications. The major source of false-positive diagnoses of MS with MR imaging is incidental white matter changes (unidentified bright objects, peniventniculan hypenintensities, leukoaniosis) present in healthy volunteers and patients. As our study demonstrates, the types of non-MS cases read together with the MS cases affect the specificity of the diagnosis of MS. The risk of a false-positive diagnosis of MS is related to the frequency and pattern of the nonspecific changes. Some diseases, such as encephalitis, cerebral vasculitis, on acute disseminated encephalomyelitis, may also be misdiagnosed as MS (2). The prevalence of nonspecific incidental white matter changes is up to 20% among healthy persons (3), higher among hypertensive patients (4), and 30%100% among demented patients (5,6). The specificity of 8i%-95% measured in our study among those control subjects that had white matter changes very likely is an estimate of the specificity of MR imaging in the most adverse situation. The specificity measured among the healthy volunteens represents the most favorable situation. The study shows that great-
en specificity
is achieved
when
the
subject’s age and sex are considered. The possible methodologic errors in this study are primarily in the ciassification of patients and control subjects. The verification of MS cases with Schumacher criteria is accepted in research because anatomic diagnoses are not routinely obtained (7). Schumacher criteria are the most conservative for selecting MS patients. They are based on clinical course and findings, not on labomatory findings. Poser criteria involve laboratory evidence but rely pnimanily on clinical findings. Although the health of the volunteers was unvenified, the probability of including a case of MS in our control group is a fraction of a percent. Bias in case selection is not likely to affect the mesults: The specificity calculation is not based on the number or type of MS cases. The sensitivity reached in this study (66%-83%, depending on the
Radiology
449
#{149}
a.
b.
e.
f.
(2,769/80)
MR images that show confluent images
weighted
show
(2,800/80)
control 90%
g.
J.
1.
Figure 1. 80) images
group in
previous
were read paraatrial
multiple
images
used) reports
is lower (1,8,9).
k.
multiple
than
the Blind-
one blinded study the overall specificity of MR imaging (not specifically for MS) was 81% (10). In an unblinded study, the specificity of MR imaging for MS was 59%-100%, depending on the criteria used (11). Even with the most stringent criteria suggested in that study, one of our control subjects and one of the control
Radiology
#{149}
lesions
periventricular
ed studies tend to have lower sensitivity. Although several investigators measured the sensitivity of MR imaging in the diagnosis of MS (1,8,9), few have reported its specificity. In
450
h.
1.
as positive for MS in control subjects. (a-c) Proton-density lesions in a 40-year-old female volunteer. (e, g) Proton-density
periventricular
show
d.
C.
subjects (9)
would
in
a 40-year-old
female
hyperintense
white
reported have
by received
(2,308/25) and (b, d) T2-weighted (2,308/ (2,769/20) and (f, h) T2-weighted volunteer. (i, k) Proton-density (2,800/30) and (j, 1) T2matter lesions in a 54-year-old hypertensive woman.
Ormerod MR
et al diag-
noses of MS. Although the specificity of MR imaging has been thought to diminish among patients aged more than 40 years, we did not find a lowen specificity among the older subjects. Our study indicates that, despite a high incidence of nonspecific cerebral white matter abnormalities, the diagnosis of MS with MR imaging has a high degree of specificity if age and sex information are used and if questionable cases are considered negative. However, up to 2%-4% of
healthy
subjects
have
high-intensity
images from na.
pemiventniculan
abnormalities
MR
on
that cannot be differentiated MS on the basis of present U
cnite-
References 1.
Asbury
AK,
Herndon
et al.
Use
in the
diagnosis
Cy
of
statement
Sclerosis
RM,
magnetic
McFarland
resonance
of multiple of the
Society.
sclerosis:
National Magn
HF, imaging
poli-
Multiple
Reson
Med
1986;
3:821-822. 2.
Miller MR
litis
3.
DH, brain
Ormerod scanning
differentiation 515. Neuroradiology Awad JA, Spetzler
IEC,
Gibson
in patients
A, et al. with
vascu-
from multiple sclero1987; 29:226-23 1. RF. Hodak JA, et al.
February
1991
4.
Incidental lesions noted on magnetic resonance imaging of the brain: prevalence and clinical significance in various age groups. Neurosurgery 1987; 20:222-227. Bradley WG, Waluch V. Brandt-Zawadzky M, et al. Patchy periventricular lesions in elderly:
a common
NMR imaging. 1984; 1:35-41. 5.
6.
b.
C. 7.
George
9.
10.
d. Figure
2.
(a, c, e) Proton
of a hypertensive
read
as negative
control
e. density subject
for MS.
(2,000/20) with
f. and (b, d, f) T2-weighted
nonspecific
periventnicular
(2,000/70) lesions.
The
MR images images
were
11.
DeLeon
MJ,
during
Med Kalnin
Imaging A, et al.
Leukoencephalopathy in normal and pathologic aging. II. MRI of brain lucencies. AJNR 1986; 7:567-570. Bowen BC, Barker WW, Loewenstein DA, et al. MR signal abnormalities in memory disorder and dementia. AJNR 1990; 11:283-290. Paty DW, Oger JJF, Kastrukoff LF. MRI in the
8.
AE,
observation
Noninvasive
diagnosis
of MS:
a prospective
study with comparison tion, evoked potentials, ing, and CT. Neurology
of clinical evaluaoligoclonal bond1988; 38:180-184.
Baker
Kispert
HL,
Berquist
TH,
DB,
et al.
Magnetic resonance imaging in a routine clinical setting. Mayo Clin Proc 1985; 60:75-90. Ormerod EIC, Miller DH, McDonald WI, et al. The role of NMR imaging in the assessment of multiple sclerosis and isolated neurological lesions. Brain 1987; 110:1579-1616. Haughton VM, Rimm AA, Sobocinski KA, et al. A blinded clinical comparison of MR imaging and CT in neuroradiology. Radiology 1986; 160:751-755. Fazekas F, Offenbacher H, Fuchs 5, et al. Criteria for an increased specificity of MR interpretation in elderly subjects with suspected multiple sclerosis. Neurology 1988; 38:1822-1825.
Volume
178
Number
#{149}
2
Radiology
451
#{149}