点击显示 收起
1 From the Department of Radiology, University of Bonn, Germany. Received September 29, 2000; revision requested November 22; final revision received July 30, 2001; accepted August
| ABSTRACT |
|---|
| Top ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION REFERENCES |
|---|
MATERIALS AND METHODS: Forty-three consecutive patients with chronic liver disease were enrolled in this study. The imaging protocol included unenhanced breath-hold T1-weighted fast field-echo sequences, unenhanced respiratory-triggered T2-weighted turbo spin-echo (SE) sequences, dynamic gadolinium-enhanced T1-weighted three-dimensional turbo field-echo sequences, and ferumoxides-enhanced T2-weighted turbo SE sequences. Images of each sequence and two sets of sequences (ferumoxides set and gadolinium set) were reviewed by four observers. The ferumoxides set included unenhanced T1- and T2-weighted images and ferumoxides-enhanced T2-weighted turbo SE MR images. The gadolinium set included unenhanced T1- and T2-weighted images and dynamic gadolinium-enhanced three-dimensional turbo field-echo MR images. In receiver operating characteristic (ROC) curve analysis, the sensitivity and accuracy of the sequences were compared in regard to the detection of all, small, and large HCCs.
RESULTS: Imaging performance was different with gadolinium- and ferumoxides-enhanced images in the detection of small and large HCCs. For detection of small HCCs, the sensitivity and accuracy with unenhanced and gadolinium-enhanced imaging (gadolinium set) were significantly (P = .017) superior to those with unenhanced and ferumoxides-enhanced imaging (ferumoxides set). The area under the composite ROC curves, or Az, for the gadolinium set and the ferumoxides set was 0.97 and 0.81, respectively. For large HCC, the ferumoxides set was superior compared with the gadolinium set, but this difference was not statistically significant. Analysis of all HCCs demonstrated no significant differences for gadolinium- and ferumoxides-enhanced imaging.
CONCLUSION: For the detection of early HCC, gadolinium-enhanced MR imaging is preferred to ferumoxides-enhanced MR imaging because the former demonstrated significantly greater accuracy in the detection of small HCCs.
Index terms: Gadolinium, 761.12143 • Liver neoplasms, MR, 761.121411, 761.121416, 761.12143 • Receiver operating characteristic (ROC) curve
| INTRODUCTION |
|---|
| Top ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION REFERENCES |
|---|
The use of contrast agents at MR imaging has improved the delineation of liver tumors. Dynamic MR imaging after bolus injection of gadolinium chelates is a valuable method for detection and characterization of liver tumors (6–9). Use of ferumoxides also has increased the sensitivity in lesion detection at MR imaging, especially in patients with liver metastases (10–14).
The detection of HCC with ferumoxides and gadolinium chelates is based on different mechanisms (10–14). The superparamagnetic iron oxide particles are taken up by the reticuloendothelial system, including the Kupffer cells of the liver. As the Kupffer cells are reduced in neoplasms, the contrast between lesions and surrounding liver tissue can be improved on T2-weighted MR images (14). However, the use of ferumoxides for the detection of HCCs in chronic liver diseases presents the problem that scarring and inflammation also reduce the hepatic uptake of ferumoxides in the surrounding liver (1,15). Although this limits the effectiveness of ferumoxides-enhanced images, some researchers (1,16) suggest that ferumoxides performs well in not all but most chronic liver diseases.
Detection of HCCs with gadolinium-based contrast material is based on the increased arterial blood supply in most of the HCCs, which results in increased detectability during the arterial phase of dynamic T1-weighted MR imaging. However, in patients with cirrhosis of the liver, it is still unclear whether ferumoxides- or gadolinium-enhanced imaging should be preferred for the detection of HCC. Tang et al (17) found gadolinium-enhanced imaging more valuable than ferumoxides-enhanced imaging, but Vogl et al (16) detected more HCC nodules with ferumoxides-enhanced MR imaging than with dynamic gadolinium-enhanced MR imaging. Because this difference might be caused by the size of the HCCs, the purpose of our study was to test the hypothesis that the accuracy of dynamic gadolinium- and ferumoxides-enhanced MR imaging might be different in small (
| MATERIALS AND METHODS |
|---|
| Top ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION REFERENCES |
|---|
Patients
The study included 43 consecutive patients (34 men, nine women; mean age, 60 years ± 13 [SD]; age range, 35–76 years) with chronic liver disease who were routinely referred to our department for gadolinium- and ferumoxides-enhanced MR imaging of the liver as part of the diagnostic work-up. Inclusion criteria were the presence of chronic liver disease and availability of follow-up studies of at least 1 year (except in patients who had received liver transplants). Twenty-seven patients had cirrhosis of the liver, and 16 patients had chronic hepatitis (hepatitis B, 10 patients; hepatitis C, four patients; hepatitis B and C, two patients). Thirty-two patients were referred for MR imaging examinations because of hepatic lesions that were suspected of being HCC at computed tomography (CT) (n = 21) or ultrasonography (US) (n = 11). In five patients, MR imaging was performed before transjugular intrahepatic portosystemic shunt placement; in three patients, MR imaging was performed before liver transplantation; and in three patients, tumor markers (-fetoprotein, 20 µg/L) were increased without findings of hepatic lesions at CT or US.
MR Imaging Examinations
All MR imaging examinations were performed with a 1.5-T imaging system (ACS-NT; Philips Medical Systems, Best, the Netherlands), with a maximum gradient strength of 23 mT/m and a rise time of 0.2 msec, by using a body coil for transmission and reception of the signal. At our institution, we prefer to use the body coil for MR imaging of the liver because phased-array coils produce inhomogeneous images of the liver, as the signal-to-noise ratio changes with the distance from the surface coil. Unenhanced images included breath-hold T1-weighted fast field-echo sequences (repetition time msec/echo time msec, 130/4.5; flip angle, 80°; field of view, 34 cm; matrix, 128 x 256; number of sections, 24; section thickness, 8 mm; number of signals acquired, one) and respiratory-triggered T2-weighted turbo spin-echo (SE) sequences (2,000/60; flip angle, 90°; field of view, 34 cm; matrix, 179 x 256; number of sections, 24; section thickness, 8 mm; number of signals acquired, four; echo train length, 17).
Within 1 week, gadolinium-enhanced dynamic T1-weighted three-dimensional (3D) turbo field-echo and ferumoxides-enhanced T2-weighted turbo SE MR imaging were performed in the same patient.
The dynamic T1-weighted 3D turbo field-echo sequences (7/3; flip angle, 20°; field of view, 34 cm; matrix, 128 x 256; number of sections, 20; section thickness, 9 mm; number of signals acquired, one) were acquired during a single breath hold of 19 seconds before and 30, 60, and 180 seconds after rapid (flow rate, 5 mL/sec) bolus injection of 20 mL of gadopentetate dimeglumine (Magnevist; Schering, Berlin, Germany) with an automated injector (Spectris MR; Medrad Europe, Maastricht, the Netherlands). The injection was followed by a 20-mL saline flush. The 9-mm section thickness of the 3D turbo field-echo sequence was adapted to the T2-weighted sequences with a section thickness of 8 mm and a gap of 0.8 mm. The aim was to avoid the influence of different section thicknesses in each of the sequences. At the time of the study, bolus timing software was not available at our institution. Administration of prebolus to determine the transit time was not performed, because small HCC might have been obscured by a small amount of gadopentetate dimeglumine.
Ferumoxides-enhanced MR imaging was performed after infusion of 0.075 mL of ferumoxides solution (Endorem; Guerbet, Sulzbach, Germany) per kilogram of body weight diluted in 100 mL of 5% glucose solution during 30 minutes. After the infusion was completed, a respiratory-triggered T2-weighted turbo SE sequence was performed, with the same parameters as were used at unenhanced MR imaging.
Lesion Confirmation
Confirmation of the diagnosis and determination of the number and sizes of the liver lesions were based on combined interpretation of findings from all diagnostic procedures by two experienced radiologists (D.P., B.K.) in consensus who did not take part in the receiver operating characteristic (ROC) analysis. The diagnostic procedures included the following: histologic analysis; determination of -fetoprotein levels; and US, dual helical CT, MR imaging, and digital subtraction angiography with CT arteriography and with CT portography. In all patients (except two patients who received liver transplants), follow-up study findings during a minimum of 12 months (range, 12–26 months) were available. Histologic confirmation was available by means of subsequent percutaneous biopsy in 14 patients, intraoperative biopsy in eight patients, or liver transplantation in two patients. In six patients without histologic confirmation, follow-up study findings were used to confirm the diagnosis of HCC as follows: In one patient with chronic hepatitis C, an increase in tumor size from 3 to 6 cm was demonstrated at follow-up MR imaging and from 1 to 5.8 cm at follow-up CT, and the -fetoprotein level increased from 52 to 168 µg/L. In four patients with chronic liver disease, an increase in the tumor size of at least 100% was seen at follow-up MR imaging and dual helical CT, and portal vein infiltration of the tumor was seen at CT during arterial portography. One patient with chronic hepatitis B had an 8-cm tumor in the liver that increased to 10 cm at follow-up, and the -fetoprotein level was 421 µg/L. Thirteen patients had no HCC. Two of 13 patients had a liver cyst, and nine of 13 had multiple regenerative nodules. Three of 13 patients had no regenerative nodules or other focal lesions. The diagnosis was proved with follow-up study findings that demonstrated no manifestation of an HCC during at least 1 year.
On the basis of confirmation, 77 HCCs were present in 30 patients. The mean number of HCCs per patient was 3.0 ± 1.6, and the maximum number in a patient was eight. Thirty-one of the 77 lesions were 1.5 cm or smaller in diameter, and 46 lesions were larger than 1.5 cm in diameter (20 lesions were 1.5–3.0 cm, 18 lesions were 3.1–6.0 cm, and eight lesions were larger than 6.0 cm in diameter).
MR Image Analysis
The retrospective reviewing procedure in the ROC analysis was performed in three sessions, and each observer had no knowledge of any clinical information. To limit learning bias, the interval between sessions was at least 3 months, and images were randomly assigned to each reader and to each session.
In the first session, the observers reviewed images obtained with each sequence separately. The images of all unenhanced and contrast material–enhanced sequences were presented to each observer (J.T., R.B., R.C., S.F.) in a randomized fashion, so only one image obtained with one sequence could be used for scoring.
In the second session, a set of images (ferumoxides set) that included unenhanced and ferumoxides-enhanced images (unenhanced T1-weighted fast field-echo, unenhanced T2-weighted turbo SE, and ferumoxides-enhanced T2-weighted turbo SE MR images) of each lesion were presented to the observers for scoring.
In the third session, the observers read a set of images (gadolinium set) that included unenhanced and gadolinium-enhanced images (unenhanced T1-weighted fast field-echo, unenhanced T2-weighted turbo SE, and dynamic gadolinium-enhanced T1-weighted 3D turbo field-echo MR images).
Each observer recorded suspected lesions and assigned each lesion a confidence rating score on the basis of a five-point scale as follows: 5, HCC definitely not present; 4, HCC probably not present; 3, undetermined; 2, HCC probably present; and 1, HCC definitely present. Overall, each reader made 474 decisions for the presence of HCC on the presented images. Prospectively, the observers were aware that only those lesions with a score of 1 or 2 would be included in separate sensitivity estimations.
In addition, two radiologists (D.P., B.K.) who did not take part in the ROC analysis evaluated all lesions that were not identified by any observer on the images of the gadolinium set or the ferumoxides set for potential explanations about why the lesions were missed.
Statistical Analysis
For all sequences and for each observer, alternative-free-response ROC curves were plotted for all, small, and large lesions. With use of ROC evaluation software (ROCKIT 0.9B; C. Metz, University of Chicago, Ill), the diagnostic accuracy of each imaging sequence was determined by calculating the area under the ROC curve (Az). Differences between ROC curves were tested for significance (P
Composite ROC curves were used to represent the performance of all observers as a group and were calculated by averaging the scores assigned by each of the observers for each sequence and for each set of images, as demonstrated in previously published articles (14,17).
The number of lesions assigned a score of 1 (HCC definitely present) or 2 (HCC probably present) by each reviewer was regarded as the number of correctly diagnosed lesions, and sensitivity values were calculated on this basis. Differences in sensitivity values in the detection of HCCs were tested for significance (P
The statistic was used to measure the degree of agreement among the observers, and values greater than 0 were considered to indicate positive correlation. Data were as follows: values between 0 and 0.30 were considered to indicate a positive but poor correlation; values between 0.31 and 0.60, good correlation; values between 0.61 and 0.90, very good correlation; and values greater than 0.90, excellent correlation.
| RESULTS |
|---|
| Top ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION REFERENCES |
|---|
|
|
The analyzed interobserver variability demonstrated good to very good agreement among the four observers for all sequences (Table 1).
ROC Analysis for Small Lesions
In the group of small lesions, the mean area under the composite ROC curves with the pooled data of the four observers was as follows: unenhanced T1-weighted fast field-echo, Az = 0.75; unenhanced T2-weighted turbo SE, Az = 0.77; gadolinium-enhanced T1-weighted 3D turbo field-echo, Az = 0.83; ferumoxides-enhanced T2-weighted turbo SE, Az = 0.78; ferumoxides set, Az = 0.81; gadolinium set, Az = 0.97 (Fig 2).
|
ROC Analysis for Large Lesions
For large lesions, the results of the ROC analysis were different, compared with the results of the analysis for small lesions. The best accuracy for large lesions was achieved with the ferumoxides set with Az of 0.99, followed by the gadolinium set with Az of 0.98. The mean areas under the composite ROC curves for single sequences were as follows: unenhanced T1-weighted fast field echo, Az = 0.85; unenhanced T2-weighted turbo SE, Az = 0.92; gadolinium-enhanced T1-weighted 3D turbo field-echo, Az = 0.94; and ferumoxides-enhanced T2-weighted turbo SE, Az = 0.94 (Fig 3). The difference between the Az values of the ferumoxides set and those of the gadolinium set was not significant (P = .639).
|
|
|
|
|
|
|
|
|
|
| DISCUSSION |
|---|
| Top ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION REFERENCES |
|---|
Tang et al (17) found gadolinium-enhanced MR imaging more valuable than ferumoxides-enhanced MR imaging in the detection of HCCs by using T2*-weighted gradient-echo and fat-suppressed T2-weighted turbo SE sequences for ferumoxides-enhanced MR imaging; in 53 patients with HCCs, significantly more lesions were identified on gadolinium-enhanced MR images compared with ferumoxides-enhanced MR images (97 of 103 vs 80 of 103, P < .01).
In contrast, Vogl et al (16) detected more HCC nodules with ferumoxides-enhanced MR imaging than with dynamic gadolinium-enhanced MR imaging. More recently, Ward et al (18) found that by using imaging with both contrast materials, gadolinium and ferumoxides in double-contrast MR imaging, compared with only ferumoxides-enhanced imaging, the diagnosis of HCC could be improved. In their study, the sensitivity was substantially more increased for small lesions (<1 cm), for which sensitivity increased from 14% to 46%, than for larger lesions (1 cm), for which sensitivity increased from 81% to 91% with the addition of gadolinium-enhanced imaging.
The setup of our study reflected clinical practice closely, because the use of alternative-free-response ROC analysis allowed positional information to be recorded and enabled all of the readers’ scores to be correlated with the actual lesion present. As with conventional ROC methods, the area under the alternative-free-response ROC curve indicates the performance or accuracy of a sequence (14).
The detection of small HCCs is of special clinical importance. In patients with chronic liver diseases, transplantation is the most effective treatment for HCC if the diagnosis is made at an early stage. The decision to proceed to transplantation is based on the number and size of lesions. Solitary small HCCs are usually treated successfully with transplantation; however, patients with solitary 2–5-cm lesions have a reduced 4-year survival rate of 75% (19). Transplantation is of no benefit when the lesions are diffuse or multiple.
The following factors may have contributed to our findings of substantially superior accuracy with dynamic gadolinium-enhanced MR imaging in small HCCs. First, well-differentiated small HCCs are known to show active uptake of iron oxide particles, which reduces lesion conspicuity on ferumoxides-enhanced images as reported by Vogl et al (16). In our study, this was seen in four of eight lesions that were not identified by any observer with the ferumoxides set. Second, the reduced uptake of iron oxide particles in cirrhotic liver parenchyma due to decreased activity of Kupffer cells might obscure small hepatic lesions more often than it would obscure large lesions (15). Third, smaller well-differentiated HCCs have a mainly arterial blood supply, which results in higher conspicuity in the arterial phase of dynamic gadolinium-enhanced MR imaging (20–22).
A limitation of our study is that not all HCCs were histologically proved. This is a known shortcoming in most comparative studies (23). However, histologic confirmation of HCC in liver cirrhosis may be difficult (21). The major problem is that US- or CT-guided biopsy is impossible for nodules that are visible on only MR images. Therefore, the diagnosis was based on long-term follow-up studies of at least 1 year.
In conclusion, different accuracy values of gadolinium- and ferumoxides-enhanced MR imaging were determined in the detection of small and large HCCs in chronic liver diseases. For small lesions, gadolinium-enhanced imaging was significantly superior to ferumoxides-enhanced imaging, and for large lesions and in the overall performance of the ROC analysis for all lesions, no significant differences were identified between gadolinium- and ferumoxides-enhanced MR imaging. Because gadolinium is less expensive and the gadolinium-enhanced MR imaging procedure is less time-consuming, dynamic gadolinium-enhanced MR imaging should be preferred for early detection of HCC.
| REFERENCES |
|---|
| Top ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION REFERENCES |
|---|