| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
1 Department of Radiology, Gifu University School of Medicine, 40 Tsukasamachi,
Gifu 500-8705, Japan.
2 Department of Nuclear Medicine, Kyoto University Faculty of Medicine, Kyoto
606-8501, Japan.
3 Department of Radiology, Osaka University School of Medicine, Osaka 565-0871,
Japan.
4 Department of Diagnostic Radiology, National Cancer Center Hospital, Tokyo
104-0045, Japan.
Received March 10, 2003;
accepted after revision July 23, 2003.
Address correspondence to M. Kanematsu.
Abstract
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
MATERIALS AND METHODS. Ferumoxides-enhanced MRIs obtained in 31 patients with 50 malignant hepatic tumors (33 hepatocellular carcinomas, 17 metastases) were reviewed retrospectively by three independent offsite radiologists. T1-weighted gradient-recalled echo images with TEs of 1.4 and 4.2 msec; T2*-weighted gradient-recalled echo images with TEs of 6, 8, and 10 msec; and T2-weighted fast spin-echo images of livers were randomly reviewed on a segment-by-segment basis. Observer performance was tested using the McNemar test and receiver operating characteristic analysis for the clustered data. Lesion-to-liver contrast-to-noise ratio was also assessed.
RESULTS. Mean lesion-to-liver contrast-to-noise ratios were negative and lower with gradient-recalled echo at 1.4 msec than with the other sequences. Sensitivity was higher (p < 0.05) with gradient-recalled echo at 6, 8, and 10 msec and fast spin-echo sequences (75–83%) than with gradient-recalled echo sequences at 1.4 and 4.2 msec (46–48%), and was higher (p < 0.05) with gradient-recalled echo sequence at 8 msec (83%) than with gradient-recalled echo at 6 msec and fast spin-echo sequences (75–78%). Specificity was comparably high with all sequences (95–98%). The area under the receiver operating characteristic curve (Az) was greater (p < 0.05) with gradient-recalled echo at 6, 8, and 10 msec and fast spin-echo sequences (Az = 0.91–0.93) than with gradient-recalled echo sequences at 1.4 and 4.2 msec (Az = 0.82–0.85).
CONCLUSION. In the detection of malignant hepatic tumors, gradient-recalled echo sequences at 8 msec showed the highest sensitivity and had an Az value and lesion-to-liver contrast-to-noise ratio comparable with values from gradient-recalled echo sequences at 6 and 10 msec and fast spin-echo sequences.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Researchers have compared imaging sequences such as conventional spin-echo, rapid acquisition with relaxation enhancement (referred to as fast or turbo spin-echo), and gradient-recalled echo sequences, but their results might have been influenced by differences in the MRI parameters or magnetic field strengths used [7, 8, 10–14]. Currently, however, T2*-weighted gradient-recalled echo and T2-weighted fast spin-echo sequences are often used in ferumoxides-enhanced MRI in many centers. More recently, some researchers [15–20] have reported the imaging findings or usefulness of T1-weighted imaging after administration of superparamagnetic iron oxide compounds.
Although a number of studies were conducted to assess the detectability of malignant hepatic tumors, few, to our knowledge, have described the optimal TE setting with the gradient-recalled echo sequence or have evaluated the detection ability of T1-weighted gradient-recalled echo imaging compared with T2*-weighted gradient-recalled echo and T2-weighted fast spin-echo imaging. We conducted such a study using a surgically proven patient population and dedicated statistical analyses.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Our surgeons performed partial hepatectomy on a segment-by-segment basis (using the Couinaud segment classification) whenever possible. They thoroughly explored the whole liver for tumors with visual inspection, palpation, and intraoperative sonography while referring to the radiologic reports. When they found suspected tumors outside the liver segments scheduled for resection, they performed intraoperative sonographically guided needle biopsy. Our pathologists thoroughly explored the resected specimens and intraoperative biopsy specimens, if any, for malignant tumors. Three on-site radiologists performed follow-up imaging on the patients after surgery with gadolinium-enhanced MRI after 6–12 months (mean, 8 months). When a patient had a recurrent tumor in the hepatic remnant, we performed a percutaneous needle biopsy to check for recurrent malignancy.
The radiologists retrospectively reviewed the preoperative ferumoxides-enhanced MRIs, and when a tumor was judged in consensus to be visible on any MRI sequence, it was enrolled in the standard-of-reference tumors. Liver segments were considered to have no tumors in the statistical analyses if none were detected preoperatively or intraoperatively and if no new tumors were detected in the follow-up studies that were judged to be visible on preoperative MRIs. Thus, a sample of standard-of-reference tumors for statistical analysis was established by surgical records, pathology reports, follow-up radiology records, and retrospective MRI review.
The study population of 31 patients (27 men and four women; age range, 50–84 years; mean, 68 years) included 18 patients with 33 hepatocellular carcinomas (diameter, 4–150 mm; mean, 24.7 mm) and 13 with 17 metastases (diameter, 4–150 mm; mean, 24.7 mm) from colorectal (n = 6), gastric (n = 2), pancreatic (n = 2), biliary tract (n = 2), and duodenal (n = 1) carcinomas. In all, the histopathologically confirmed standard-of-reference lesions consisted of 50 malignant hepatic tumors (diameter, 5–80 mm; mean, 22.8 mm). Two tumors coexisted in the same liver segment in four segments in three patients with hepatocellular carcinomas and in six segments in three patients with metastases. No patient had three or more tumors in one segment.
Ferumoxides-Enhanced MRI
Ferumoxides (Feridex, Tanabe Pharmaceutical, Tokyo, Japan) were
administered at a dose of 10 µmol of iron per kilogram of body weight. The
ferumoxides suspension was diluted in 100 mL of 5% glucose solution and
administered IV over 30 min. Four patients had a slight back pain, but the
remaining patients had no adverse effects in this study. The MRI was begun
within 1 hr after the completion of administration of the ferumoxides
solution.
MRI was performed with a 1.5-T imager (Signa Horizon, General Electric Medical Systems, Milwaukee, WI). All MRIs were obtained in the axial plane with a phased array multicoil for the body, a section thickness of 8 mm with a 2-mm intersection gap, and a field of view of 32 x 24–29 x 22 cm. MRI was performed with five breath-hold gradient-recalled echo sequences with different TE settings (1.4, 4.2, 6, 8, and 10 msec) and a respiratory-triggered, fat-suppressed fast spin-echo sequence. The parameter settings of these sequences are summarized in Table 1.
|
Quantitative Image Analysis
The quantitative analysis was conducted with operator-defined
region-of-interest measurements of mean signal intensity of the hepatic
tumors, adjacent liver parenchyma, and background noise by one study
coordinator who studied 18 hepatocellular carcinomas (size range, 10–80
mm; mean, 28.8 mm) and 11 metastases (size range, 10–120 mm; mean, 34.5
mm) that were clearly visible for quantitative measurement with all sequences.
In the hepatic tumors, a circular region of interest was drawn to encompass as
much of the lesion as possible. The first SD of background noise was measured
in the phase-encoding direction outside the anterior abdominal wall to
calculate the lesion-to-liver contrast-to-noise ratio: the signal intensity of
the liver was subtracted from the signal intensity of the hepatic tumor, and
the remainder was divided by the first SD of the background noise.
Qualitative Image Analysis
Three gastrointestinal radiologists from other institutions who had no
knowledge of the patient information independently reviewed the MRIs on a
segment-by-segment basis in a random order. Each radiologist recorded the size
and site (Couinaud segment) for each visible lesion and allocated a confidence
level to each hepatic segment for the presence of malignant tumors (1 =
definitely absent, 2 = probably absent, 3 = indeterminate, 4 = probably
present, 5 = definitely present). A total of 248 liver segments were reviewed,
including 33 segments with hepatocellular carcinomas and 17 segments with
metastases. The radiologists were instructed to allocate a score of 1 when no
focal signal intensity change was seen; a score of 3 when the signal intensity
change was subtle, ill-defined, and not circular or oval; and a score of 5
when the signal intensity change was discrete, well-circumscribed, and
circular or oval. Scores of 2 and 4 were allocated according to the
radiologists' subjective judgment. At the time of image review, the
radiologists were aware that the sensitivities were calculated using the
number of segments allocated a confidence rating of 4 or 5 (i.e., probably
present or definitely present), and the specificities with the number of
segments with a rating of 1–3 (i.e., definitely absent, probably absent,
or indeterminate).
Statistical Analysis
Analysis of variance and multiple comparisons using the Scheffé
criterion [21] were performed
to compare the mean lesion-to-liver contrast-to-noise ratios. Sensitivity for
tumor detection was determined by the number of segments assigned a score of 4
or 5 (i.e., probably present or definitely present) of the total number of
segments with tumors. Likewise, specificity was determined by the number of
segments assigned a score of 1, 2, or 3 (i.e., definitely absent, probably
absent, or indeterminate) of the total number of segments without tumors.
We compared sensitivities and specificities using an extension of the McNemar test for clustered data (Clusterpro, Cleveland Clinic Foundation, Cleveland, OH) [22]. Observer performance was examined by calculating the areas under curves (Az) with receiver operating characteristic analysis for clustered data. For each imaging sequence, a nonparametric receiver operating characteristic curve for clustered data was constructed for each radiologist and for pooled rating data from the three radiologists (Cluster.For, Cleveland Clinic Foundation). Differences between two receiver operating characteristic curves for clustered data were tested by comparing areas with a Z test that took into account the correlation between imaging techniques on each segment and cluster of segments in patients (Clusterbi.For, Cleveland Clinic Foundation).
The kappa statistic for multiple observers was used to measure the degree of agreement and assess inter-observer variability in interpreting images [23]. A kappa value of up to 0.20 stood for slight agreement, 0.21–0.40 for fair agreement, 0.41–0.60 for moderate agreement, 0.61–0.80 for substantial agreement, and 0.81 or greater for almost perfect agreement.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
The mean lesion-to-liver contrast-to-noise ratios in hepatocellular carcinomas and metastases with each sequence are shown in Figure 1. The mean lesion-to-liver contrast-to-noise ratio in hepatocellular carcinomas with gradient-recalled echo sequences at a TE of 1.4 msec showed a negative value and was significantly lower than those with gradient-recalled echo at a TE of 4.2 msec or longer and the fast spin-echo sequence (p < 0.05) (Fig. 2A, 2B, 2C, 2D, 2E, 2F). For metastases, the mean lesion-to-liver contrast-to-noise ratio with gradient-recalled echo sequence at a TE of 1.4 msec also showed a negative value and was significantly lower (p < 0.05) than those with gradient-recalled echo at a TE of 6 msec or longer and the fast spin-echo sequences (Fig. 3A, 3B, 3C, 3D, 3E, 3F).
|
|
|
|
|
|
|
|
|
|
|
|
|
The mean sensitivities for tumor detection are shown in Figure 4. For hepatocellular carcinomas, metastases, and tumors overall, the sensitivity was significantly greater (p < 0.05) with gradient-recalled echo at TEs of 6, 8, and 10 msec and fast spin-echo sequences than with gradient-recalled echo sequences at TEs of 1.4 and 4.2 msec. For metastases and tumors overall, the sensitivity was significantly greater (p < 0.05) with the gradient-recalled echo sequence at a TE of 8 msec than with gradient-recalled echo at a TE of 6 msec and the fast spin-echo sequences. The minimum size of tumors detected by the radiologists working in blinded fashion with any sequences of MRIs was 8 mm in hepatocellular carcinomas and 7 mm in metastases. The mean specificities are shown in Figure 5. No significant differences appeared among sequence specificities.
|
|
The mean Az values for tumor detection are shown in Figure 6. For hepatocellular carcinomas, metastases, and tumors overall, the Az value was greater (p < 0.05) with gradient-recalled echo at TEs of 6, 8, and 10 msec and fast spin-echo sequences than with the gradient-recalled echo sequence at a TE of 1.4 msec. For hepatocellular carcinomas and tumors overall, the Az value was greater (p < 0.05) with gradient-recalled echo at TEs of 6, 8, and 10 msec and fast spin-echo sequences than with the gradient-recalled echo sequence at a TE of 4.2 msec.
|
The multiple-observer kappa values ranged from 0.77 to 0.87 (mean, 0.82) for hepatocellular carcinomas, from 0.77 to 0.87 (mean, 0.82) for metastases, and from 0.63 to 0.88 (mean, 0.81) for tumors overall. Substantial or almost perfect agreement was reached among the three radiologists.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
For hepatocellular carcinomas, however, the mean lesion-to-liver contrast-to-noise ratio with the gradient-recalled echo sequence at a TE of 1.4 msec was closer to the baseline, which meant that hepatocellular carcinomas were more isointense to the liver than were the metastases. This quantitative result explained the lower sensitivity and Az value for hepatocellular carcinomas compared with metastases with this sequence. The reasons that hepatocellular carcinomas were less conspicuous with T1-weighted gradient-recalled echo sequences were twofold: First, the tissue components in hepatocellular carcinomas were more analogous to those in normal liver than those in metastases. Second, cirrhosis that often accompanies hepatocellular carcinomas disturbed ferumoxides uptake in the liver and limited the increase in signal intensity of the surrounding liver.
Positive inversion of mean lesion-to-liver contrast-to-noise ratios occurred with gradient-recalled echo sequences at a TE of 4.2 msec or longer for both hepatocellular carcinomas and metastases. At a TE of 4.2 msec, the lesion-to-liver contrast-to-noise ratio in hepatocellular carcinomas was twice as high as that in metastases. Conversely, however, the sensitivity and Az value at a TE of 4.2 msec were somewhat greater with metastases than with hepatocellular carcinomas. The cause was unclear but may relate to the fact that sensitivities and Az values tended to be greater with metastases than with hepatocellular carcinomas with all sequences in our study, suggesting the probable superiority of ferumoxides-enhanced MRI with metastases over hepatocellular carcinomas.
Some recent studies [15–20] reported that T1-weighted images obtained after administration of superparamagnetic iron oxide compounds are useful in the differentiation of malignant tumors from benign hepatic lesions: Hepatic cysts appear distinctly hypointense because of the long T1 relaxation time that they originally have; cavernous hemangiomas appear moderately hyperintense because of the T1 shortening effect of superparamagnetic iron oxide compounds pooled in the cavernous vessels; and focal nodular hyperplasia can also appear mildly hyperintense because of the T1 shortening effect when the contrast agent is taken up in the residual reticuloendothelial systems. Studies suggest that T1-weighted gradient-recalled echo imaging at shorter TEs does not necessarily improve tumor detection but may contribute to improved tissue characterization.
Gradient-recalled echo sequences at TEs of 6, 8, and 10 msec showed high mean lesion-to-liver contrast-to-noise ratios and high sensitivities and Az values for both hepatocellular carcinomas and metastases. The sensitivity with the gradient-recalled echo sequence at a TE of 8 msec was significantly higher than with other sequences for metastases and tumors overall and tended to be higher than the sensitivity of the gradient-recalled echo sequence at a TE of 10 msec for hepatocellular carcinomas, metastases, and tumors overall. The Az values for the different sequences and times were not significantly different, but it was suggested that a TE of 8 msec for ferumoxides-enhanced T2*-weighted gradient-recalled echo imaging may be optimal because of its sensitivity [15–20].
In our clinical practice, we occasionally find that the peripheries of
hepatic tumors are blurred on the gradient-recalled echo images at a TE of 10
msec because this sequence is susceptible to internal magnetic heterogeneity
and prone to the blooming effect
[12,
13]. We suspect that small
tumors (
5 mm) may be missed with T2*-weighted gradient-recalled echo
sequences with long TEs for the same reason. Meanwhile, the fast spin-echo
sequence is more robust to the internal magnetic homogeneity than the
gradient-recalled echo sequence and may have a greater chance to detect tiny
tumors. Further research will be necessary to clarify this.
The fast spin-echo sequence was inferior to the gradient-recalled echo sequence at a TE of 8 msec in sensitivity for metastases and tumors overall, but was comparable in Az value with gradient-recalled echo sequences at TEs of 6, 8, and 10 msec for hepatocellular carcinomas, metastases, and tumors overall. The respiratory-triggered T2-weighted fast spin-echo sequence offered high lesion-to-liver contrast-to-noise ratio, high spatial resolution derived from its large imaging matrices, and less blooming effect because of the robustness of the T2* effect; small rounded areas of hyperintensity caused by hepatic cysts, hemangiomas, or hepatic vessels were more efficiently differentiated from small malignant tumors [12]. Hepatic cysts are virtually isointense and not seen on T2*-weighted gradient-recalled echo images with long TEs but are distinctly hyperintense on fast spin-echo images (Fig. 3A, 3B, 3C, 3D, 3E, 3F). Cavernous hemangiomas are often as hyperintense as solid malignant tumors on T2*-weighted gradient-recalled echo and fast spin-echo images, but T1-weighted gradient-recalled echo images with short TEs show them as hyperintense and are useful for characterizing them. We believe that the additional use of fast spin-echo and T1-weighted gradient-recalled echo sequences improves the characterization.
The image review and statistical analyses were conducted on a segment-by-segment basis first, because our surgeons performed partial hepatectomy on Couinaud's segment-by-segment basis whenever possible. Second, one of the chief determinants of hepatic resectability was the accurate definition of the number of segments to be resected. Third, our objective was not to localize lesions but to compare the ability of radiologists to detect liver tumors and to define liver segments to be resected on preoperative MRIs obtained with each imaging technique. It may be a statistical issue that multiple lesions existed in one liver segment; in our study two hepatocellular carcinomas coexisted in the same segment in four segments in three patients, two metastases coexisted in the same segment in six segments in three patients, but no patient had three or more tumors in one segment. All six patients with multiple tumors in one segment underwent partial hepatectomy on Couinaud's segment-by-segment basis for the resection of tumors. We believe that such multiplicity in a segment can be ignored as long as the number of such segments is small and a partial hepatectomy based on Couinaud's classification system is used.
The validity of standard-of-reference segments with no tumor may be limited. We had 198 liver segments without tumors; all these segments were intraoperatively explored for tumors by visual inspection, palpation, and intraoperative sonography performed with reference to the radiology reports, and all were followed up in 6–12 months after partial hepatectomy. Nine new tumors appeared in the follow-up studies, and three of them were enrolled in the standard-of-reference tumors for statistical analysis because they were judged to be visible even on the preoperative MRIs.
Although we used dedicated criteria to establish the standard-of-reference tumor group, it is possible that some of the recurrent tumors not enrolled in the standard-of-reference tumors had already existed at the time of preoperative MRI. The Az values may be inflated in our study because the number of false-negative segments may have been underestimated, and any lesions missed on all sequences reduce Az values but do not affect comparisons among the pulse sequences.
This study had some other limitations. We did not evaluate the detectability of tumors by combining different imaging sequences. However, we suggest that a combination of T2*-weighted gradient-recalled echo, T2-weighted fast spin-echo, and T1-weighted gradient-recalled echo sequences would improve tissue characterization and ensure tumor detection on ferumoxides-enhanced MRI, and such a combination would not require a lengthy examination.
Although the cohort in our study was not large, we tried to compensate for that by using dedicated statistical procedures. In our study, we administered a dose of 10 µmol of iron per kilogram of body weight, but we have no evidence that this dose is optimal. A 15-µmol dose is commercially available in European countries. We suspect that the higher dose may improve the detection of hepatic tumors.
Finally, the diagnostic value of ferumoxides-enhanced MRI for detection of hepatocellular carcinomas may be limited because ferumoxides uptake is hampered in cirrhotic livers with impaired function [24], and some well-differentiated hepatocellular carcinomas themselves take up ferumoxides. Gadolinium-enhanced MRI may be more useful, particularly in the diagnosis of hepatocellular carcinoma [25, 26]. Nevertheless, superparamagnetic iron oxide-enhanced MRI seems particularly useful for the detection of metastases [9, 27]. Ferucarbotran, a new type of bolus-injectable superparamagnetic iron oxide [28] currently commercially available in many countries, significantly reduces the examination time and allows acquisition of unenhanced images in the clinical setting.
In conclusion, we found the greatest sensitivity for malignant hepatic tumors was achieved using the gradient-recalled echo sequences at a TE of 8 msec, and we confirmed no significant difference in lesion-to-liver contrast-to-noise ratio and Az value with the receiver operator characteristics analysis for gradient-recalled echo at 6, 8, and 10 msec and fast spin-echo sequences. T1-weighted gradient-recalled echo sequences at TEs of 1.4 and 4.2 msec did not show high lesion detectability, although they may have improved tissue characterization.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
This article has been cited by other articles:
|
|
 |
|
  R. F. Hanna, N. Kased, S. W. Kwan, A. C. Gamst, A. C. Santosa, T. Hassanein, and C. B. Sirlin Double-Contrast MRI for Accurate Staging of Hepatocellular Carcinoma in Patients with Cirrhosis Am. J. Roentgenol., January 1, 2008; 190(1): 47 - 57. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
M.-J. Kim, J. H. Kim, J. Y. Choi, S. H. Park, J.-J. Chung, K. W. Kim, and D. G. Mitchell Optimal TE for SPIO-enhanced gradient-recalled echo MRI for the detection of focal hepatic lesions. Am. J. Roentgenol., September 1, 2006; 187(3): W255 - W266. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
