Downloaded from www.ajronline.org by 91.74.11.197 on 10/18/15 from IP address 91.74.11.197. Copyright ARRS. For personal use only; all rights reserved
763
Detection
of Hepatic
Comparison Unenhanced
of Contrast-Enhanced CT, MR Imaging, and Iron Oxide-
Enhanced
MR
Metastases:
Imaging
Diagnostic accuracy of contrast-enhanced CT, unenhanced MR imaging, and MR images enhanced with superparamagnetic iron oxide was evaluated in 10 patients with histologically proved hepatic metastases. First, diagnostic performance of the imaging
Christian J. Fretz1’2 David D. Stark1 Charles E. Metz3 Guillermo Ralph Weissleder1 Jong-Her Shen3 Jack Wittenberg1 Joseph Simeone1 Joseph T. Ferrucci1
technique with respect to the ability of radiologists to recognize the presence or absence of a metastasis was measured by using receiver-operating-characteristic (ROC) analysis
of single images. Second, the total number of lesions (N = 108) detected by “complete” cT and MR examinations was counted. Finally, lesion-liver contrast-to-noise ratios (CNR) were measured in all MR sequences. The area under the ROC curve was .67 ± .03 for contrast-enhanced CT, .81 ± .07 for the unenhanced SE 260/14 sequence, and .92 ± .01 for the iron oxide-enhanced SE 1500/40 sequence. The enhanced SE 1500/40 sequence yielded significantly (p < .005) greater accuracy than did contrast-enhanced cT. The same sequence detected significantly (p < .05) more lesions than all other imaging techniques (19% more than the best unenhanced MR sequence and 36% more than contrast-enhanced ci). The enhanced SE 1500/40 sequence also yielded the
highest CNR value (19.5 ± 10.2) of all MR sequences. These results indicate that iron oxide-enhanced MR imaging technique for the detection of hepatic lesions. AJR 155:763-770,
October
is a superior
imaging
1990
Detection of hepatic metastases is critical to treatment planning at the time of diagnosis and during follow-up of a large number of cancer patients [1 -3]. Furthermore, hepatic resection for cancer has shown increased survival rates for selected patients [4-9]. Therefore, not only the presence or absence of hepatic lesions (in an all-or-none fashion), but also their number, location, and size may
Received February vision May 23, 1990.
26, 1 990; accepted
improve
This work was supported Cancer Society JFAA-163
in part by the American
and PDT-326 and the of Energy DE-FGO2-86ER60418.
U.S. Department I
after re-
Department
of Radiology,
Massachusetts
Gen-
eral Hospital and Harvard Medical School, Boston, MA 02114. 2 Present address: Institut f#{252}r Diagnostische Radiologie, Kantonsspital, CH-9007 St. Gallon, Switzerland. Address reprint requests to C. J. Fretz. 3
Department
Chicago, 4
of
Chicago,
Radiology,
The
University
of
IL 60637.
Present address: Nuclear Magnetic Resonance
Unit, University
influence therapy and patients’ outcome [3, 4, 7, CT has been considered the gold standard for because it has greater sensitivity and specificity [1 3]. Various contrast enhancement techniques
Hospital,
Nuevo
Leon,
0361 -803X/90/1 554-0763 0 American Roentgen Ray Society
Mexico.
10-12].
detection than
of focal hepatic
lesions
sonography
or scintigraphy [1 4-1 8] have been proposed to
the diagnostic performance of CT, with sensitivity reported as 76and specificity reported as 70-99% [1 5, 20-22]. However, when pathologic inspection is used as a more exacting gold standard, the true sensitivity of contrast-enhanced CT for detection of individual hepatic lesions has recently been shown to be only 38% [23]. Several studies have concluded that unenhanced MR imaging can equal or exceed the accuracy of contrast-enhanced CT [1 9-21 , 23-25]. The introduction of superparamagnetic iron oxide as a tissue-specific MR contrast agent for the 96%
further
[1 5, 1 9-22]
reticuloendothelial
system
(RES)
appears
to offer
a further
detection of hepatic lesions [26]. In this study, we analyzed objectively the diagnostic performance of iodine-enhanced and iron oxide-enhanced MR. First, we conducted istic (ROC) analysis [27, 28] based on anatomically
compared
the number of lesions detected
improvement
in the
clinical data to compare CT, unenhanced MR,
a receiver-operating-charactermatched images; second,
in “complete”
CT and MR examinations;
we
FRETZ
764
Downloaded from www.ajronline.org by 91.74.11.197 on 10/18/15 from IP address 91.74.11.197. Copyright ARRS. For personal use only; all rights reserved
and third, we correlated these diagnostic results with quantitative measurements of lesion-liver contrast-to-noise ratios (CNR) calculated with standard techniques for MR image analysis.
Materials
ET AL.
AJR:155,
were repeated.
The relaxivity,
superparamagnetic netics,
Inc.,
iron
pharmacokinetics,
oxide
Cambridge,
formulation
MA)
have
and toxicity
(AMI-25,
been
1990
October
of this
Advanced
Mag-
elsewhere
described
[31 , 32]
Image Selection
and Methods
The ROC study was based on a total of 478 images representing Patients
54 anatomic
Ten patients years)
with
(four men, six women;
recent
41 -65 years of age; mean,
contrast-enhanced
oxide-enhanced
MR
CT,
examinations
and
unenhanced
MR,
acceptable
proof
and
of the
57 iron
pres-
ence or absence of hepatic lesions were selected for this study. Data from these patients concerning perfusion and retention imaging for lesion detection and characterization have been reported [29, 30]. No patients were excluded on the basis of image quality. All 10 patients
had
a biopsy-proved
primary
cancer
(four
colorectal
and
two
breast adenocarcinorna, and one each had lymphoma, carcinoid, islet cell carcinoma, and melanoma). The presence of malignant hepatic lesions in each patient was verified by surgical biopsy (seven), CT or sonographically guided percutaneous
needle
biopsy
(two),
and
autopsy
(one)
within
1 month
after CT and MR examination. examinations were performed
CT or unenhanced MR follow-up in five of the patients, and interval
growth
of individual
confirmed
the
presence
malignant
lesions.
The interval between CT and MR examinations ranged from 2 to 1 8 days, with a mean of 8 days (CT usually preceding MA) for seven patients.
Three
patients,
one
with
advanced
metastatic
disease
of
the liver and two patients each with three colonic metastases, had an interval of 35-64 days between tests, and in these three patients
there was no demonstrable
increase in size or number of lesions.
CT Scanning
CT examinations (four
patients:
Solon,
NJ),
General
Electric
Milwaukee,
on a Technicare DR (one
patient;
were
obtained
4 mm
was
used
in nine
in one
patient
nonionic
patient
or approximately
whereas
Nine
contrast
media
with
a flow CT
of 2 mI/sec scanning
coopergap
was
Injection
via peripheral
by using
a mechanical
30 sec after
initiated
enhanced
MA
were
(26 normal
matching
images
without
could
disease
anatomic
sections
enhanced
CT,
28 abnormal
MA,
and
Of these 478 images, 260
be identified.
and 21 8 with
and
unenhanced
disease.
A lesion
was
either
visible in retrospect with all imaging techniques on the same level or not visible on the anatomic section of interest or on the two adjacent sections.
Sections
corresponding
in which
image
were
one
of
the
excluded
techniques
from
the
did
study.
not
This
have
was
a
often
the case for the most cephalic and caudal sections. By selecting matching anatomic sections instead of “complete” CT and MA examinations, we could avoid a potential bias against the CT breathhold imaging technique, which may omit or duplicate one or more anatomic sections because of slice misregistration [33]. In our study, unenhanced and enhanced MR images were taken at the same level, as the position of the center slice was marked on the patient’s skin, and table position is monitored electronically. No MR or CT images had to be excluded for reasons related to artifacts or poor technical quality. The proof for presence or absence of individual lesions was based on intraoperative or pathologic findings in five patients. Twenty-six of 54 anatomic sections were selected from these five patients. Consensus reading by two radiologists of all MA, sonography, and CT examinations
in conjunction
with all clinical
data and follow-up
imaging
examinations was used for the remaining five patients, contributing 28 sections to the ROC analysis. A lesion was considered real when the same defect in a given anatomic section was present on at least two different types of images, either on CT and MR (whether enhanced or not) or on unenhanced and enhanced MR images. Similar standards of proof have been used by previous investigators [19,
25].
Observer Performance Information
of
ionic and one 42 g of iodine per
received
0.6 g of iodine/kg).
patients.
on patient
(at least
for which
Systems, NJ), or
an intersection
patients
unit Iselin,
was 8 or 10 mm for
(dependent
patients,
patient.
iodinated
performed
slices
1 440HP Siemens,
GE Medical Hackensack,
Slice thickness
“Contiguous”
ation)
was
Somatom
2060 (one patient).
all examinations.
vein
performed
OH),
9800 (three patients; Elscint 2000 (one patient;
WI),
Technicare
were
sections
sections)
injector
in nine
injection obvious
began [22]. A drip infusion was used in one patient who had extensive metastatic liver disease.
films.
identifying
Individual
mixed
Experiments the patient
images
in random
were
order,
and
cut
was obscured from
on the hard-copy
standard
multiformat
film,
to
radiologists
with
presented
three
experience in abdominal CT and MR who had not previously seen these cases and who did not participate in the consensus reading of the examinations. The readers scored each image for the presence or absence
of focal
hepatic
to their observation
(1
=
lesions
definitely
and
assigned
or almost
a confidence
definitely possibly
level
absent; present;
2
=
possibly absent; 4 5 probably present; 6 definitely or almost definitely present). For each imaging method, a binormal ROC curve [28] was fitted
probably
absent;
3
=
=
=
=
MR Imaging
to All studies imaging
were
system
performed (Technicare,
echo (SE) images
on a 0.6-T Solon,
were acquired
an SE 260/1 4/1 0 (TR/TE/number
(25.1
OH).
MHz)
Baseline
with Ti -dependent
unenhanced
contrast
spin-
by using
averaged) sequence and with intermediate to T2-dependent contrast (SE 500/30/6, SE 1500/40,80/2) sequences. The whole liver could be imaged within 11 sections, of 4 mm. (nine in one
of signals
each
having
a thickness
of 1 5 mm
One
to two
hours
IV infusion
patients patient),
and
an intersection
of iron
oxide
gap particles
20 MmoI Fe/kg, and 5 mol Fe/kg were injected
received the
after
SE
500/30/6
and
SE
1 500/40,80/2
each
observer’s
estimation
superconducting
sequences
[34].
was determined specific binomial 35]. Composite three
readers
parameter
confidence
Diagnostic
rating
accuracy
data
by
maximum
of the various
likelihood
imaging
methods
by calculating the area (Ar) under each readerROC curve when it is plotted in the unit square [34, ROC curves to represent the performance of the
as a group
were
calculated
by averaging
the
binormal
values of the individual curves. Differences between the imaging methods in terms of the mean areas (As) under the ROC curves were analyzed statistically by using Student’s two-tailed t-test for paired data [36]. Differences between ROC curves of individual readers were tested for significance by
IRON
AJR:155, October 1990
TABLE
OXIDE-ENHANCED
MR
1: Receiver-Operating-Characteristic
OF
Estimates A
Imaging
METASTASES
765
for CT and MR Imaging
Index
Technique
Mean A Index
Reader 1
Downloaded from www.ajronline.org by 91.74.11.197 on 10/18/15 from IP address 91.74.11.197. Copyright ARRS. For personal use only; all rights reserved
HEPATIC
Reader 2
Reader 3
Contrast-enhanced CT Unenhanced MR imaging
.658
.646
.708
.670 ± .03
SE 260/14 SE 500/30 SE 1500/40 SE 1500/80 Enhanced MA imaging SE 500/30 SE 1500/40 SE 1500/80
.860 .774 .702 .737
DD .733 .701 .741
.760 .650 .431 .642
.810±07 .719 ± .06 .611 ± .16 .707 ± .06
.845’ .917c
.898a .914c
.701 .923
.81 5 ± .10’ .918 ±
.977’
.944’
.795
.905 ± .i0
Note.-DD A, index
.
b C d
degenerate
=
higher
(p
.oi
data.
.05) than that of contrast-enhanced CT. Mean A, index higher (p < .05) than that of unenhanced SE 1500/40. A1 index higher (p < .01) than that of contrast-enhanced CT. Mean A, index higher (p < .005) than that of contrast-enhanced CT.
I-
/ ./
05
/
;
.
0 I
3.
w Lesion-Liver
CNR
I.
I-
Lesion-liver hanced
and
Fe/kg,
signal
CNR
tumor-to-liver
intensities
of liver, at least
noise including
at the same signal
quantitatively
to compare
MR images. In nine patients
enhanced
and background each image
was measured
anatomic
difference
to
the
tumor,
were measured
level to calculate scaled
standard
deviation
of
of pulse with the
statistically with one-factor analysis of variance(ANOVA)for repeated measures and Fisher’s protected least-significance difference (PLSD) [40].
IRON OXIDE.ENHANCED
II
on
detectability of lesions with different MR sequences [26]. CNR differences between the MR pulse sequences were analyzed
procedure
-
SE 1500/40
uNENHANcED5E260/14
-
-
CCNTRAST#{149}ENHANCEDCT
-
the CNR as the
background noise [38, 39]. CNR is an objective measure sequence performance that has been shown to correlate
multicomparison
/#{149}
receiving 20 mol
one representative
ghost artifacts
unen-
.
0.0
.
.
0.0
.
‘
.
05
FALSE
POSITIVE
. 10
FRACTION
Fig. 1.-Ocmposite receiver-operatlng-characteristic (ROC) curves indicate relative accuracies with which focal hepatic lesions are detected by contrast-enhanced CT, unenhanced MR (SE 260/14), and Iron oxideenhanced MR (SE 1500/40). Plotted data points represent specific ROC points of each reader for Iron oxide-enhanced MR (squares) and contrastenhanced CT (diamonds).
766
FRETZ
TABLE
2: Number
of Lesions
Detected
ET AL.
with Various
AJR:1 55, October
Imaging
Diameter
Techniques
by Size of Lesion
of Lesion
Technique
Total
Contrast-
Downloaded from www.ajronline.org by 91.74.11.197 on 10/18/15 from IP address 91.74.11.197. Copyright ARRS. For personal use only; all rights reserved
enhanced
SE 1500/40 SE 1500/80
SE SE
30
9 8
.
enhanced
mm
MR
500/30
enhanced
20-29
CT
Unenhanced MR SE 260/14 SE 500/30
Enhanced
1990
Imaging Technique MA
Unenhanced
SE SE SE
260/14 500/30 1500/40
SE 1500/80 Enhanced
SE SE SE
Contrast-to-Noise Ratio (mean ± SD) -13.3 -4.7
± ±
6.7a
3.6 ±
3.7
4.7 ±
5.9
18.7 ±
6.4b
3.2
MR
500/30 1500/40 1500/80
19.5 ± 10.2’ 15.5 ± Sib
Note-Lesion-liver contrast-to-noise ratio (CNR) calculated in nine patients receiving 20 1111O1 Fe/kg. An analysis of variance was performed. . CNR higher (p < .01) than that of unenhanced SE 1500/40. b CNR higher (p < .01) than that of unenhanced SE 500/30 and SE 1500/40,80.
Downloaded from www.ajronline.org by 91.74.11.197 on 10/18/15 from IP address 91.74.11.197. Copyright ARRS. For personal use only; all rights reserved
AJR:155,
IRON
October1990
D
OXIDE-ENHANCED
MR
OF
HEPATIC
METASTASES
E
Fig. 2.-Metastatic
colonic
767
F
cancer.
A, Unenhanced CT scan. No lesion visible. B, Iodine-enhanced CT scan. Note good opacification of hepatic vessels. Streaklike artifacts partially obscure left lobe of liver. No lesion visible. C, SE 260/14 MR image obtained before iron oxide injection shows one lesion (curved arrow) in left lobe of liver. D and E, SE 500/30 (D) and SE 1500/40 (E) MR images obtained before iron oxide injection. Lesion in left lobe of liver not visible. F, SE 1500/40 MR image obtained after injection of 20 MmoI Fe/kg of AMI-25. Lesion in left lobe (curved arrow) confirmed; additional lesion (straight arrow) visible in right lobe.
phagocytic ability, and therefore the relaxation times of tumors remain virtually unchanged after iron oxide administration. The resulting loss of signal intensity from liver, with tumor unchanged, increases tumor-liver contrast (Figs. 3B-
3E). Traditionally,
calculation
of sensitivity
and
specificity
was
used to compare new imaging techniques [1 5, 1 9, 25]. However, there may arise a dilemma in which one technique provides higher sensitivity, but lower specificity, and the relative capacities of the two techniques cannot be determined [28]. Instead, ROC analysis provides a more meaningful approach to assess diagnostic performance of different techniques [27, 28]. In this study of iron oxide-enhanced MR imaging, the SE 1 500/40 sequence yielded a significantly (p < .005) higher mean A value than did contrast-enhanced CT. The three enhanced MR sequences reached higher mean A2 values than did all unenhanced MR sequences. Inspection
of images
proved
shows
detection
that
improved
of small lesions
lesion
conspicuity
and
(Fig. 2) correspond
im-
to the
higher A values for the iron oxide-enhanced MR images. Our ROC study design may even underestimate the true superiority of iron oxide-enhanced MR imaging, because the demonstration of more than one lesion in a given image did not affect the reader’s scoring (Fig. 3). The number of lesions detected by the enhanced SE 1500/ 40 sequence was 1 9% and 36% higher than the corresponding numbers for the best unenhanced MR sequence (SE 260/ 14) and contrast-enhanced CT, respectively. The improved contrast between liver and lesion after injection of iron oxide allowed a higher detection rate for all enhanced sequences
compared
with their unenhanced
counterparts.
Notably,
small
lesions (5-20 mm in diameter) previously missed on unenhanced MR or enhanced CT images were frequently visualized by enhanced MR (Table 2 and Fig. 3). This clinical result
FRETZ
Downloaded from www.ajronline.org by 91.74.11.197 on 10/18/15 from IP address 91.74.11.197. Copyright ARRS. For personal use only; all rights reserved
768
Fig. 3.-Metastatic
islet cell carcinoma.
External
biliary drainage
ET AL.
AJR:155, October 1990
with air in biliary tree. Ascites.
Portal hypertension
with esophagogastric
vances.
A, Iodine-enhanced CT scan shows subcapsular hypervascular metastasis in right lobe of liver (arrows). B and C, SE 500/30 (B) and SE 1500/40 (C) MR images obtained before iron oxide injection. Metastasis is hypointense in B (arrows), not visible in C because of isointensity with surrounding liver. D and E, SE 500/30 (D) and SE 1500/40 (E) MR images obtained after injection of 20 Mmol Fe/kg of AMI-25. Liver signal intensity decreased in both images resulting in better delineation of 2.5-cm subcapsular metastasis. In addition, two small (