Detection of Hepatocellular

Carcinoma: Comparison of Low- and

High-Spatial-Resolution Dynamic MR
Images

Masayuki Kanematsu1 OBJECTIVE. The purpose of our study was to compare the diagnostic performance of

Hiroaki Hoshi1 low- and high-spatial-resolution gadolinium chelate-enhanced triphasic dynamic gradient-re-
Takamichi Murakami2 called echo (GRE) MR images in the detection of hepatocellular carcinoma.
MATERIALSAND METHODSTriphasic dynamic MR images obtained with low (256 x
Kyo Itoh3
1 28) and high (5 12 x 224) image matrices in 28 patients with 65 hepatocellular carcinomas
Masatoshi Hon2
(HCCs) were retrospectively analyzed. Image review was conducted on a segment-by-segment
Hiroshi Kond&
basis; a total of 2 15 liver segments, including 56 segments with tumor burden, were reviewed for
Hironobu Nakamura2 the presence of HCC by three independent radiologists. Detectability was evaluated with relative
sensitivity, specificity, and receiver operating characteristic (ROC) analysis. Image quality was
evaluated with rank order analysis.
RESULTS. Relative sensitivity was statistically significantly better with high-spatial-resolu-
tion images than with low-spatial-resolution images (p < .005). Relative specificity was statisti-
cally significantly better with low-spatial-resolution images than with high-spatial-resolution
images (p < .001). Diagnostic accuracy determined by ROC curve analysis was marginally
higher with high-spatial-resolution (area under ROC curve [AJ = .97) than with low-spatial-reso-
lution (A = .94, p < .09) images. Image quality was statistically significantly better with high-
spatial-resolution images (p < .005).
CONCLUSION. High-spatial-resolution dynamic GRE images were superior to low-
spatial-resolution images in sensitivity of detecting HCC and in image quality. Triphasic dy-
namic GRE imaging in the screening and follow-up programs of patients with suspected HCC
should be performed using high image matrices.

A ccurate
carcinoma
detection
(HCC)
of hepatocellular
in patients with
and Fujita
fulness
et al. [10] described
of dynamic GRE imaging
a potential
with the
use-
use
chronic liver damage is important of phased array multicoil and higher image ma-
in determining treatment options such as sur- trices (512 x 256 [frequency x phase-encoding
gical resection, transcatheter embolization, or matricesi). However, because of the prolonged
percutaneous ablation therapy. In addition to TE (6 msec) available on their GRE sequences
Ti- and T2-weighted MR imaging, the diag- and MR imaging systems. a limited number of
nostic usefulness of multiphasic dynamic gra- obtainable slices and longer breath-hold times
dient-recalled echo (GRE) MR imaging for were inevitable. Some recent commercially
hepatic tumor detection has been described available high-magnetic-field MR scanners
Received January 4, 1999: accepted after revision
April 19, 1999.
[1-5]. Although parenteral iron oxide com- with echoplanar capability have enabled a very
pounds have recently been shown to improve short TE, less than 2 msec, for GRE imaging
tDepartment of Radiology, Gifu University School of
Medicine, 40 Tsukasamachi, Gifu 500-8705, Japan. the detectability of focal hepatic lesions [6], even when using higher image matrices, result-
Address correspondence to M. Kanematsu. gadolinium chelate-enhanced multiphasic dy- ing in an increased number of obtainable slices
20epartment of Radiology, 2-15 Yamadaoka, Suita City, namic MR imaging remains important for the and in sufficient coverage of the whole liver
Osaka University Medical School, Osaka 565-0871, Japan. detection and characterization of focal liver during a tolerated respiratory suspension.
3Department of Radiology, Kyoto University Faculty of lesions [7, 8]. However, the use of higher image matrices
Medicine, 54 Seigoin-Kawaramachi, Sakyo-ku, Kyoto City, Dynamic MR images of the liver evaluated results in lower image contrast because of the
Kyoto 606-8507, Japan.
in previous studies were chiefly obtained using decreased number of protons in each image
AJR 1999:173:1207-1212
lower image matrices because of the limited ca- voxel, or in prolonged acquisition time. which
0361-803X/9a/1735-1207 pability of gradient strength, slew rate, and can increase image blur caused by motion.
© American Roentgen Ray Society pulse sequence architecture. Ohtomo et al. [9] Furthermore, longer acquisition time may not

AJR:173, November 1999 1207

 Kanematsu et al.

be optimal to obtain predominantly hepatic ar- and MR imaging 3-12 months (mean, 5.5 months) Image Analysis
tery phase images that are indispensable for after the initial MR imaging. Three radiologists who have served mainly as
the detection of hypervascular lesions. It is still The presence or absence of HCC was determined gastrointestinal radiologists for 7-17 years and who
by consensus of three radiologists on the basis of the have interpreted MR images of the liver as part of
unclear whether the dynamic GRE images ob-
findings on sonography, CT during arterial portogra- their daily clinical and research practices, were in-
tained with high image matrices (512 x 224,
phy and CT hepatic artenography combined, TI- and vited from other institutions to review the images.
referred to in our study as high-spatial-resolu-
‘P2-weighted MR imaging, dynamic MR imaging, They knew that the patients were referred for assess-
tion images) can replace dynamic GRE images ferumoxides-enhanced MR imaging, follow-up ment of possible HCC, but they did not know any
with low image matrices (256 x 128, low-spa- sonography, CT, or MR imaging; serologic tests; bi- other information about the patients’ histories. They
tial-resolution images). Therefore, we retro- opsy; or definitive surgery with intraoperative sonog- independently reviewed triphasic dynamic images
spectively compared the triphasic dynamic raphy. Thus, in the 28 patients, confirmation was obtained with low- and high-spatial resolution ORE
GRE images separately obtained with low and obtained for 65 HCCs (diameter, 6-SO mm [mean, sequences in the 28 patients.
high image matrices by assessing their diag- 21.7± 14.1 mm]). The image review was conducted on a segment-
nostic performance and image quality. by-segment basis-first, because one of the chief
MR Imaging determinants of hepatic resectability is the accurate
MR imaging was performed using a supercon- definition of the number of segments to be resected;
ducting magnet at 1.5 1 (Signa Horizon; General and second, because our objective was to assess the
Materials and Methods
Electric Medical Systems, Milwaukee, WI) and ability of the radiologists to detect HCC on images
Patient Study
phased array multicoil for the body (Torso-array coil: obtained with each imaging technique and not to lo-
During the 12-month period ofiune 1997 through General Electric Medical Systems). The system pro- calize the lesions. To prevent mislocation of the Ic-
May 1998. 37 patients who were strongly suspected vides an echoplanar capability with maximum gradi- sions by the radiologists, the hepatic segment
of having frank HCC on the basis of previously per- ent strength of 23 ml . m with a peak slew rate of numbering system of Couinaud [I 1] was drawn on
formed sonography, CT, or MR imaging underwent 77 ml . m msec1. For low- and high-spatial-res- the images by the study coordinator. The image re-
both low- and high-spatial-resolution dynamic MR olution gadolinium chelate-enhanced dynamic TI- view was performed in two separate sessions. Im-
imaging in our department separated by 3-day to 2- weighted GRE imaging, a fast multiplanar spoiled ages were reviewed in alphabetic order according to
week intervals. Informed consent for the MR imag- GRE acquisition under steady-state free precession the patient’s name, but the order in which images
ing examinations was obtained from all patients. One sequence (256 x 128 [frequency x phase-encoding obtained with the two imaging techniques were re-
patient with confluent fibrosis in cirrhosis was found matrices] and 512 x 224 image matrices, respectively; viewed was randomized. In other words, images
not to have HCC after the liver workup and was cx- 1’RIrE of 150/1.4 and 150/1.8, respectively; flip angle from all patients were reviewed at a single session,
cluded from the analysis because our study was con- of 110#{176};one signal acquisition) was used. The K- but only the images obtained with either of the two
ducted on a liver segment-by-segment basis. Eight space lines were filled sequentially from top to bot- imaging techniques in a given patient were reviewed
patients with HCC examined in the initial period of torn in the phase-encoding axis of the K-space, and at that session. The images obtained with the other
the study who had undergone transcatheter arterial the signal data obtained at the middle of acquisition technique were reviewed at the subsequent session.
chemoembolization or percutaneous ethanol ablation time were filled in the central K-space lines. The total To minimize learning bias, the name, age, identifica-
therapy MR imaging and who underwent MR
before acquisition times were 16 and 26 sec for low- and tion number, and imaging parameters of each patient
imaging for evaluation of recurrent HCC were also high-spatial-resolution imaging, respectively. were masked, and the two reviewing sessions were
excluded. We excluded them because we suspected, Breath-hold low- and high-spatial-resolution dy- performed at 2-week intervals.
after this study was initiated, that statistically signifi- namic GRE images of the patients were obtained For each imaging technique, the radiologists re-
cant bias was introduced by imaging findings af- with a 75% field-of-view of35-26
rectangular x 29- corded the size and site (Couinaud segment) of visi-
fected by prior treatments. The study population 22 cm. Axial images obtained with a section
were ble abnormalities and indicated for each segment
consisted of the remaining 28 patients, who were 22 thickness of 8-10 mm and a 2- to 3-mm intersec- whether the presence of HCC could be ascertained.
men and six women 51-81 years old (mean, 67 tional gap to cover the entire liver. Dynamic GRE The radiologists assigned one of five confidence 1ev-
years). Of the 28 patients with HCC, 26 patients had images were obtained before and after an antecubital els (I = definitely absent, 2 = probably absent, 3 =
hepatic cirrhosis that was diagnosed on the basis of IV bolus injection of 0.1 mmol/kg of gadopentetate equivocal, 4 = probably present, 5 = definitely
definitive surgery in 13 patients, and on the basis of dimeglumine (Magnevist; Schering, Berlin, Ocr- present). When a lesion was located in two or more
morphologic findings at sonography, Cl’, or MR im- many) followed by 20 ml of flushed sterile saline. segments. the radiologist was asked to consider only
aging and liver function test results in 13 patients. Six The injection of contrast material and flushed sterile the segment that was mainly involved and to assess
patients had undergone liver resection because of saline was manually performed. The scan delays for the probability of another HCC in the other seg-
HCC 12-48 months (mean. 28.5 months) before this the hepatic artery phase were determined at 20 and ments. The radiologists were instructed to indicate a
study was conducted. 1\venty-four patients under- 14 sec after initiating contrast material injection with score of 1 when no focal signal intensity change was
went helical (1’ during arterial portography and CT low- and high-spatial-resolution imaging, respec- seen; a score of 3 when the signal intensity change
hepatic arteriography combined; and eight patients, tively, so the image data obtained approximately 27- was subtle, ill defined, and not circular or oval in
including patients who did not undergo angiographi- 28 sec after initiating contrast material injection shape; and a score of 5 when the signal intensity
cally assisted helical CT, underwent ferumoxides-en- were filled in the central lines for both im-
K-space change was discrete, well circumscribed, and circu-
hanced MR imaging within 2 weeks of MR imaging aging techniques. Therefore,
the scan delays after lar or oval in shape. Scores of 2 and 4 were assigned
for this study. Of the 28 patients with HCC, seven contrast material injection for triphasic imaging were on the basis ofthe radiologist’s subjective judgment.
subsequently underwent definitive surgery with intra- 20 see, 60 5cc, and 3 mm for low-spatial-resolution Furthermore, each radiologist evaluated the de-
operative sonography within 2 weeks ofMR imaging. imaging, and 14 sec. 60 see, and 3 mm for high-spa- gree of image quality in terms of image sharpness,
The remaining 21 patients underwent sonographically tial-resolution imaging, representing the hepatic at- image contrast, motion-related blur, and temporal
guided core needle aspiration biopsy of at least one tel)’, portal vein, and equilibrium phases, resolution of hepatic artery and portal vein phases
HCC for histologic proof within 2 weeks of MR im- respectively. For all MR imaging, a spatial presatura- usingafive-pointscale(1 =poor, 2 = fair, 3 =good,
aging: in these patients, other tumors with imaging tion pulse was used superior and inferior to the imag- 4 = very good, 5 = excellent). A “poor” score was
findings similar to those of the histologically exam- ing volume. A chemical shift selective fat-saturation assigned when the image could not be interpreted
ined lesions were considered to be the same disease. pulse was not used because the acquisition time and because of image degradation, “good” meant im-
All patients underwent follow-up sonography. CT, the number of slices were traded off. age degradation was present but did not markedly

1208 AJR:173, November 1999

 MR Imaging of Hepatocellular Carcinoma

preclude interpretation, and “excellent” meant the

image was virtually free from image degradation.
Scores of “fair” and “very good” were assigned ac-
cording to the radiologist’s subjective judgment.
Six of the 28 patients had undergone a prior
liver resection for HCC (one segment in four pa-
tients, two segments in one, and three segments in
one) before the study was performed. A total of
215 liver segments. including 56 segments with 65
HCCs. were reviewed.

StatisticalAnalysis
The relative sensitivity of each technique for de-
tecting HCC with three individual radiologists and
composite data was determined using the number
of segments assigned a score of 3 or greater (equiv-
ocal to definitely present) of the total number of 56
segments with HCC: and the relative specificity of
each technique was determined using the number
of segments assigned a score of I or 2 (definitely Note-Data are the numbers of segments of 56 se9ments with hepatocellular carcinoma assigned a score of 3-5 for sensitivity,
absent or probably absent) of the total of 159 seg- segments of 159 se9ments without hepatocellular carcinoma assigned a score of 1 or 2 for specificity, and sum ofthese segments I n =
215)for accuracy. Numbers in parentheses are the sensitivity, specificity, and accuracy expressed as percentages.
ments without HCC. The relative sensitivity and
aSpecificity was statistically significantly greater than that for high-spatial-resolution images I p < .001).
specificity were compared using the McNemar test.
bNo statistically significant difference between low- and high-spatial-resolution dynamic MR images was found.
The relative accuracy was compared with the chi-
CSensmvity was statistically significantly greater than thatfor low-spatial-resolution images I p < .005).
square test.
For each imaging technique, a binomial receiver
operating characteristic (ROC) curve was fitted to
Mean Values ofArea Under Recelv.rOp.ratingCharacterlstic Curve(A)
each radiologist’s confidence rating using a maxi-
with Low- and HIgh-SpatI*1-R..OIUtIOn Dynamic MR Images for
[ 12]. The diagnostic

- .
mum-likelihood estimation
tveallng Hepatocellular Carcinoma
accuracy of each imaging technique for each radi-
, , . I! : . .
ologist and their composite data were estimated by
calculating the area under the ROC curve (A) [131.
Differences between the ROC curves of individual
radiologists and composite ROC curves that com-
bined the performance ofall radiologists into a single
curve were tested using the area test (a univariate
score test of the difference between the areas under
the two ROC curves) 114, 15].
To assess interobserver variability in interpreting Note-Numbers are mean values ± 1 SD.
images. kappa statistics for multiple observers were
used to measure the degree of agreement. We used
difference in relative accuracy was noted be- ity was statistically significantly better with
nonweighted kappa statistics with binary data de-
tween the two techniques. high-spatial-resolution images than with low-
fined in terms of the presence (definitely present,
probably present. equivocal) or absence (probably The A index values for each radiologist spatial-resolution images (p < .005) (Table 3).
absent, definitely absent) of HCC in a liver segment. and the composite data of the two techniques
The degree of disagreement was not factored into the for detection of HCC are shown in Table 2.
calculation. A kappa value of up to .41) indicated pos- Diagnostic accuracy was statistically signifi- Discussion
itive but poor agreement, a value of .41-75 indicated cantly higher (p < 0.05) for high-spatial-reso- To increase spatial resolution by using high
good agreement, and a value of greater than .75 mdi- lution images than for low-spatial-resolution image matrices as are currently used with CT
cated excellent agreement.
images for one of the three observers. Diag- has been a challenge since MR imaging. one of
nostic accuracy determined using composite the major advances in clinical imaging, was in-
Results data was marginally higher for high-spatial- troduced. Signal-to-noise ratio, acquisition
Relative sensitivity, specificity, and accu- resolution images (A. = .97) than for low-spa- time, and number of obtainable slices are usu-
racy ofeach technique for detecting HCC with tial-resolution (A = .94, p < .09) images. The ally traded for increasing image matrices. Al-
three individual radiologists and the composite composite ROC curves generated from the though the trade-off in acquisition time and
data are shown in Table 1 . Relative sensitivity composite data of the three radiologists are number of obtainable slices can be solved by
was statistically significantly better with high- shown in Figure 3. using currently available fast multiplanar GRE
spatial-resolution images than with low-spa- The kappa values for the three radiologists sequences with shorter TEs (<2 msec) as we
tial-resolution images (p < .005) (Figs. 1 and were .83 and .84 for low- and high-spatial-reso- did in our study, the problem of the decrease in
2). In contrast, relative specificity was statisti- lution images. respectively. Excellent agree- signal-to-noise ratio cannot be disregarded. In
cally significantly better with low-spatial-reso- ment was obtained among the radiologists with our clinical study. however, low- and high-spa-
lution images than with high-spatial-resolution regard to the presence or absence of HCC in a tial-resolution images were comparable in terms
images (p < .001). No statistically significant given segment. The total degree of image qual- of image contrast, motion-related blur, and

AJR:173, November 1999 1209

 Kanematsu et al.

temporal resolution of the hepatic artery and

portal vein phases subjectively judged by the
three radiologists, whereas high-spatial-resolu-
tion images were superior to low-spatial-reso-
lution images with respect to image sharpness.
These results may indicate that dynamic GRE
imaging with high image matrices can increase
spatial resolution without any significant trade-
off in image quality necessary for efficient dy-
namic MR imaging for detecting HCC.
Regarding the detection of HCC, high-spa-
tial-resolution images showed somewhat higher
diagnostic accuracy determined by ROC analy-
sis and statistically significantly greater relative
sensitivity than did low-spatial-resolution im-
ages. We inferred that the sensitivity for detect-
ing HCC excelled in high-spatial-resolution
imaging-first. because image contrast neces-
sat), for detection of small hypervascular lesions
was reserved despite the use of high image
matrices: and second, because high image ma-
trices helped depict detailed configurations of
HCCs (Figs. 1 and 2).
Meanwhile, relative specificity was statisti-
cally significantly lower with high-spatial-reso-
lution images than with low-spatial-resolution
images. This result may indicate that the radiol-
ogists judged more nonpathologic focal signal
intensity changes as true lesions with high-
spatial-resolution images. This result has two
possible explanations. First, the high-spatial-
resolution images visualized better than low-
spatial-resolution images small perfusion
Fig. 1-70-year-old woman with cirrhosis and biopsy-proven well-differentiated hepatocellular carcinoma (ar-
abnormalities in the hepatic artery phase such
rows A and C).
A-D, Gadolinium chelate-enhanced dynamic gradient-recalled echo images obtained with low image matrices as are commonly seen on helical CT hepatic
(256 x 128 [frequency x phase-encoding matrices]) at hepatic artery (A) and equilibrium (B) phases and images arteriography [16, 17]. Second, the radiolo-
with high image matrices (512 x 224) at hepatic artery (C) and equilibrium (0) phases show conspicuity and con- gists invited from other institutions were not
figurations of tumor. Note better delineation on C and D than on A and B. Also note that peripheral portal vein
accustomed to interpreting high-spatial-resolu-
branches (arrows, B and 0) are more clearly shown on D than on B
tion dynamic GRE images. However, the low
specificity for tumor detection may be improved
while keeping high sensitivity if radiologists are
used to interpreting high-spatial-resolution
images, because high-spatial-resolution im-
ages provide more morphologic information.
Conclusively, we believe that the superiority
in both tumor sensitivity and image quality
with high-spatial-resolution images encour-
ages the practical use of this imaging technique
in the screening and follow-up programs of pa-
tients with suspected HCC, provided that more
accurate examinations (such as angiographi-
cally assisted helical CF or ferumoxides-en-
hanced MR imaging) are performed in the
preoperative workup of surgical candidates.
Fig. 2.-65-year-old man with cirrhosis and biopsy-proven poorly differentiated hepatocellular carcinoma (arrow, B). All patients evaluated in our study were co-
A and B. Gadolinium chelate-enhanced dynamic gradient-recalled echo images obtained at hepatic artery phase with
low (256 x 128, A) and high (512 x 224, B) image matrices show tumor enhancement Note that enhancement is more operative to respiratory suspension for 26 sec
pronounced on B than on A. and successfully held their breath during data

1210 AJR:173, November 1999

 MR Imaging of Hepatocellular Carcinoma

transplantation for patients with hepatic malig-

nancy has not been commonly performed.
We included patients with strongly sus-
pected frank HCC on previous examinations in
the study population, and all patients had HCC
anyway, resulting in an unusually high rate of
0 positive biopsy results compared with those of
previously reported studies. Because such in-
clusion criteria eventually excluded patients
U
with premalignant or borderline hepatic nod-
.
ules in cirrhosis (although the clinical signifi-

I
cance of such nodules is still controversial).
our conclusion may be limited to the detection
offrank HCC.
We did not perform quantitative analysis
for images obtained with the two different
techniques using the operator-defined region-
of-interest measurements of mean signal in-
tensity in the liver, spleen, or hepatic lesions,
or background noise, because we estimated
that the signal-to-noise ratio theoretically de-
False-Positive Fraction
creased approximately 62% by increasing the
image matrices from 256 x 128 to 5 12 x 224.
Fig.3.-Composite receiveroperating characteristic (ROC) curves generated from pooled data ofthree indepen- Furthermore, it was difficult to efficiently
dent radiologists for lesion detection in patients with hepatocellular carcinoma )HCC) show observers’ confi- place regions of interest encompassing lesions
dence in detection of HCC with high-spatial-resolution (0) (A = .91) and low-spatial-resolution (0) (A = .94)
throughout multiphase dynamic images be-
dynamic gradient-recalled echo (GRE) images. Note that diagnostic accuracy with high-spatial-resolution dy-
namic GRE images is somewhat higher than that with low-spatial-resolution dynamic GRE images, although dif- cause the HCCs evaluated in the study were
ference between two ROC curves did not reach statistical significance. relatively small, and some HCCs showed a
nodule-in-nodule appearance with heteroge-
neous enhancement or enhancing fibrous cap-
#{149}w’:1u-1mag. Quality ofLow- and HIgh.Spadal Resolution Dynamic MR Images sules. We deduced that satisfactory tumor
detectability and image quality with high-
MR Pulse Sequence
Quality Factor spatial-resolution images were to be achieved
Low Spatial Resolution High Spatial Resolution despite the theoretically significant decrease

Image sharpness 3.9 ± 0.5 4.4 ± 0.6a in signal-to-noise ratio, probably owing to the
strong TI-shortening effect of gadopentetate
lma9e contrast 3.8 ± 0.6 3.7 ± 0.6
dimeglumine injected as a bolus, excellent
Motion-related blur 4.2 ± 0.5 4.3 ± 0.6
Tl-weighted contrast with the fast multipla-
Temporal resolution of hepatic artery and portal vein 4.0 ± 0.7 4.0 ± 0.6
nat spoiled GRE acquisition under steady-
phases
state free precession sequence. and the use of
Total 4.0±0.6 4107b phased array multicoil.
Note-Numbers are mean values ± 1 SD. Data are based on observers’ scores of degree of image quality 11 = poor, 2 = fair, We did not use a test bolus injection [18] or
3 = good, 4 = very good, 5 = excellent). triggering software [19] to determine an opti-
iMean score was statistically significantly greater than that for low-spatial-resolution images I p < .00011.
mal scan delay for the hepatic artery phase be-
bMean score was statistically significantly greater than that for low-spatial-resolution images Ip < .005).
cause we needed to achieve a high patient
throughput in the clinical setting. and because
acquisition. However, a 26-sec breath-hold is definitive surgery, and underwent imaging triggering software was not available at the
not always tolerated by elderly patients or by within 3-12 months to confirm all other le- time of our study. Other potential limitations of
patients with impaired respiratory function or sions to also be malignant from their growth. this study are reading order bias and recall bias.
substantial ascites. The use of somewhat lower Although it is ideal that histologic proof be ob- However, because two reading sessions were
phase-encoding matrices than 224 while keep- mined from all lesions by evaluating explanted conducted in a random order for the pulse se-
ing 5 12 frequency-encoding matrices may effi- liver or by at least definitive surgery with intra- quences, the intervals between the two sessions
ciently decrease data acquisition time while operative sonography, it is becoming more in- were at least 2 weeks, and the lesions evaluated
maintaining satisfactory spatial resolution. frequent to obtain surgical proof in patients with were relatively small. we believe these biases
Our study has some limitations. Only seven HCC because small HCC lesions are often were minimal.
patients underwent definitive surgery with in- treated with transcatheter arterial chemoembo- In conclusion, gadolinium chelate-enhanced
traoperative sonography; the remaining 21 pa- lization or percutaneous ablation, especially in triphasic dynamic GRE imaging using high im-
tients with HCC underwent biopsy of at least cirrhotic patients whose functional reserve is age matrices (512 x 224) resulted in higher sen-
one HCC lesion for histologic proof instead of severely impaired, and because whole-liver sitivity for detecting HCC than did imaging with

AJR:173, November 1999 1211

 Kanematsu et al,

low image matrices (256 x 128). Image quality 4. Peterson MS, Baron RL, Murakaini T Hepatic ma- I I. Couinaud C. Lefoie: Eudes anatomiques et chirur-
was better with high-spatial-resolution images lignancies: usefulness of acquisition of multiple ar- gicales. Paris: Masson, 1957
terial and portal venous phase images at dynamic 12. Metz CE. ROC methodology in radiologic imaging.
than with low-spatial-resolution images. We be-
gadolinium-enhanced MR imaging. Radiology investRadiol 1986;21 :720-723
lieve these results encourage the practical use of
1996;20l :337-345 13. Hanley IA, McNeil BJ. The meaning and use of the
high-spatial-resolution dynamic MR imaging in 5. Larson RE, Semelka RC, Bagley AS, Molina PL, area under a receiver operating characteristic
the screening and follow-up programs of pa- Brown ED, Lee JK. Hypervascular malignant liver (ROC)curve. Radiology 1992;143:29-36
tients with suspected HCC. lesions: comparison of various MR imaging pulse 14. Metz CE. Some practical issues ofthe experimental
sequences and dynamic CT. Radiology 1994;l92: design and data analysis in radiological ROC stud-
393-399 ies. Invest Radiol 1989;24:234-245
6. Sen#{233}terre
E, Taourel P. BouvierY, et al. Detection 15. Fretz Ci, Stark DD, Metz CE, et al. Detection of he-
Acknowledgement
of hepatic metastasis: ferumoxides-enhanced MR patic metastases: comparison of contrast-enhanced
We thank Kazuyuki Uchiumi of General imaging versus unenhanced MR imaging and CT CT, unenhanced MR imaging, and iron oxide-en-
Electric Yokogawa Medical Systems, Japan, during arterial portography. Radiology 1996:200: hanced MR imaging.AJR 1990;155:763-770
for technical advice. 785-792 16. Kanematsu M, Hoshi H, ImaedaT, et al. Nonpatbo-
7. VoglTJ, Stupavsky A, Pegios W, et al. Hepatocellu- logical focal enhancements on spiral CF hepatic an-
lar carcinoma evaluation with dynamic and static giography. Abdom Imaging 1997;22:55-59
gadobenate dimeglumine-enhanced MR imaging 17. Kanematsu M, Hoshi H, Sone Y, Mochizuki R,
References and histopathologic correlation. Radiology 1997; Kato M, Yokoyama R. Detection of hepatic tumors:
1. Yamashita Y, Hatanaka Y, Yamamoto H, et al. Dif- 205:721-728 arterial-phase MR imaging versus spiral CT arteri-
ferential diagnosis of focal liver lesions: role of 8. Blakeborough A, Ward I, Wilson D, et al. Hepatic ography. Abdom Imaging 199823:416-42l
spin-echo and contrast-enhanced dynamic MR im- lesion detection at MR imaging: a comparative 18. Earls JP, Rofsky NM. DeCorato DR. Krinsky OA,
aging. Radiology 1994:193:59-65 study with four sequences. Radiology 1997;203: Weinreb IC. Hepatic arterial-phase dynamic gado-
2. Low RN. Francis LR, Hertkens Ri, et al. Fast multi- 759-765 linium-enhanced MR imaging: optimization with a
planar spoiled gradient-recalled imaging of the 9. Ohtomo K, MatsuokaY, Abe M, et al. High-resolu- test examination and a power injector. Radiology
liver: pulse sequence optimization and comparison tion MR imaging evaluation of hepatocellularcarci- 1997;202:268-273
with spin-echo MR imaging. AiR 1993:160:501- noma.Abdom Imaging 1997;22:182-186 19. Foo TK, Saranathan M, Prince MR. Chenevert TL.
509 10. Fujita 1, Honjo K, Ito K, Matsumoto T, Matsunaga Automated detection of bolus arrival and initiation
3. Ito K, Choji T. NakadaT, Nakanishi 1, Kurokawa F, N, Hamm B. High-resolution dynamic MR imag- of data acquisition in fast, three-dimensional, gado-
Okita K. Multislice dynamic MRI of hepatic tu- ing of hepatocellular carcinoma with a phased-ar- linium-enhanced MR angiography. Radiology 1997;
mors. J ComputAssist Tomogr 1993:17:390-396 ray body coil. RadioGraphics 1997;l7:315-331 203:275-280

1212 AJR:173, November 1999