We retrospectively reviewed our institutional database for liver MRIs conducted between January 2004 and September 2006 on patients suspected of having liver malignancy based on clinical and prior CT findings. We identified 391 patients with HCCs and 55 patients with liver metastases who underwent a double contrast MRI that consisted of a sequentially acquired gadolinium-enhanced and ferucarbotran-enhanced liver MRI. No institutional review board approval was required for retrospective review of medical records and images.

Of the 446 potential study patients, only those who met the following criteria were included in the study: (a) nodular HCC 2 cm or smaller in size that had been proven on histology or showed typical imaging findings on hepatic angiography, Lipiodol CT, and dynamic CT, (b) liver metastases 2 cm or smaller in size that had been proven on histology or showed typical imaging findings on three-phase dynamic CT and follow-up CT or MRI (rapid progression of disease), (c) follow-up contrast-enhanced CT or MRI performed for at least six months (range: 6-30 months). Consequently, the study population included 151 patients (114 men and 37 women; age range: 40-74 years): 119 patients with 157 HCCs and 32 patients with 98 metastases. The lesion distribution according to the confirmative diagnostic method is given in Table 1. Of 51 HCCs (from 38 patients) that were confirmed by pathologic analysis of surgical specimens, 45 lesions were grade I (n = 15) or II (n = 30) according to Edmondson's classification of HCC (11), and the remaining six were grade III. For the 81 patients with 106 HCCs who underwent transarterial chemoembolization, the final diagnosis was based on a combined interpretation of findings from all diagnostic procedures by two experienced radiologists in consensus. Diagnostic procedures included the following: image-guided biopsy (n = 12), characteristic imaging findings on three-phase multidetector CT (MDCT) and MRI (arterial hypervascularization and early washout on dynamic imaging and arterial hypervascularization plus hyperintensity on SPIO-enhanced MRI) (n = 106), and persistent dense compact nodular Lipiodol uptake of the detected lesions on MDCT and MRI after TACE (n = 106) (12), as well as aggravation (local recurrence or new growing HCCs) on follow-up CT or MRI (n = 19). Of all the patients with HCCs, four underwent ablation therapy, and two underwent chemoembolization before MRI. All patients with HCCs had liver cirrhosis (n = 116) or chronic hepatitis (n = 3) associated with viral hepatitis B. Seven of them also had viral hepatitis C. During the inclusion period for this retrospective study, there was no patient with small HCC (≤ 2.0 cm) who was without underlying liver parenchymal disease. The diagnosis of chronic hepatitis or cirrhosis was made by means of histology in 50 patients and by a combination of clinical course, blood chemistry tests (aspartate aminotransferase, alanine aminotransferase, alkaline phosphatase, bilirubin, albumin, and globulin), and typical CT and MRI findings in the remaining 69 patients. Based on the Child-Pugh classification, 98 patients were classified as Child class A, and the remaining 18 patients were classified as class B. Liver metastases arose from the following primary tumors: colorectal carcinoma (n = 25), gastric carcinoma (n = 5), and pancreatic carcinoma (n = 2). No patients with liver metastases had cirrhosis, and all had normal hepatic function test results, including prothrombin time and blood chemistry (serum albumin, total bilirubin level, and choline esterase).

All MRI examinations were performed using a 1.5 T unit (Magnetom Symphony; Siemens, Enlargen, Germany) with a combination of a phased array body coil for signal reception. Baseline MR images included a respiratory-triggered T2W-TSE sequence and a breath-hold T2^*W-GRE sequence. Respiratory-triggered T2W-TSE imaging was obtained using the following parameters: TR/TE of 3300-4200/76, echo train length of 13, 150° flip angle, matrix of 202 × 384, signal averages of two, and average scan time 3 minutes. Breath-hold T2^*W-GRE imaging was obtained using the following parameters: TR/TE of 180/12, 30° flip angle, matrix of 144 × 256, signal average of one, two sequential acquisitions, 20 slices, and average scan time 35 seconds (two stacks with 10 slices each in a 15-17-second breath-hold). A relatively long TE (12 ms) and lower flip angle (30°) were used for increasing the sensitivity to the field inhomogeneity induced by SPIO administration and for decreasing the T1 contrast, respectively (7, 13). For all sequences, a 6-7 mm slice thickness was used with a 10% intersection gap and a field of view of 35-40 cm, depending on the size of the liver.

In the first session, dynamic MRIs (volumetric interpolated breath-hold examination, VIBE: Siemens, Erlangen, Germany) with mSENSE were performed using the following parameters: TR/TE of 4.3/2.0, flip angle of 12°, bandwidth of 450 Hz/Px, matrix of 256 (read)×135 (phase)×40-46 (partition), effective slice thickness of 3.5-4 mm, and field of view of 32-35 cm. The determination of scan delay for the image acquisition timing was achieved using the test bolus technique, in which 1 mL of contrast media was injected, followed by a 20 mL saline flush, and the vessel of interest (the abdominal aorta) was then scanned approximately once per second. The mean delay time to peak aortic enhancement was 21 seconds (range: 17.0-23.0 sec). Thus, for a sequential mode of k-space acquisition (center lines of k-space are acquired during the middle of the acquisition time), the mean delay times (time interval between bolus administration and the start of image acquisition) for the early arterial, late arterial, portal, and equilibrium phase were 20, 30, 60, and 180 seconds, respectively. The acquisition time for each phase was 10-11 seconds. Gadopentetate dimeglumine (Magnevist®, Bayer HealthCare) was injected at a dosage of 0.1 mmol/kg body weight at a rate of 2 mL/sec. The contrast was injected into the antecubital vein using an automated injector (Spectris MR; Medrad Europe, Maastricht, The Netherlands), and a 20 mL saline flush followed the contrast injection.

In the second session, ferucarbotran (SHU-555-A; Resovist®, Bayer HealthCare)-enhanced imaging was performed immediately after completion of the dynamic MR examination. Ferucarbotran-enhanced imaging was comprised of the respiratory-triggered T2-weighted TSE sequence and the breath-hold T2^*-weighted GRE sequence for the baseline MRI. Patients up to 60 kg body weight received 0.9 mL of ferucarbotran, and patients above 60 kg body weight received 1.4 mL of ferucarbotran (8-12 µmol of iron per kilogram of body weight). Ferucarbotran was rapidly injected intravenously through a 5 µm filter and was followed by a 20 mL saline flush; imaging then commenced approximately 10 minutes after the intravenous injection of contrast agent.

Two gastrointestinal radiologists (one on-site reviewer and one off-site reviewer), who had at least six years experience interpreting liver imaging in their daily clinical practice, reviewed the images independently and separately. All images were reviewed on a 2000×2000 Picture Archiving and Communication Systems (PACS; Marotech, Seoul, Korea) monitor. The radiologists were aware of the overall goal of the study before the reading session, and they knew that the patients were at risk for HCC or metastasis, but were unaware of the presence or location of any liver lesions or of the results of other imaging studies. They independently reviewed three sets of MR images in two reading sessions: (a) unenhanced T1- and T2-weighted imaging, gadolinium-enhanced dynamic imaging, and ferucarbotran-enhanced T2^*W-GRE imaging (the T2^*W-GRE set), (b) unenhanced T1- and T2-weighted imaging, gadolinium-enhanced dynamic imaging, and ferucarbotran-enhanced T2W-TSE imaging (the T2W-TSE set), and (c) a combination of the T2^*W-GRE set and T2W-TSE set (the combined set). In the first session, either the T2^*W-GRE set or the T2W-TSE set was randomly presented with no specific patient order. In the second session, the remaining T2^*W-GRE set or T2W-TSE set was randomly presented, and the T2^*W-GRE or T2W-TSE reviewed in the first reading recession was added for a combined review. To minimize any learning bias, there was at least a two-week interval between the two reading sessions.

The criteria for diagnosing HCC or metastasis on the ferucarbotran-enhanced MRI were defined as a focal, discrete, nodular high signal intensity area relative to adjacent liver parenchyma (lower than the signal intensity of CSF or gallbladder on a T2W-TSE sequence to exclude the cyst) on both T2^*W-GRE and T2W-TSE images. On gadolinium-enhanced MRI, early enhancement with rapid washout (typical of HCC) and irregular peripheral enhancement with a peripheral washout on delayed phase images (typical of metastases) were used as the criteria for diagnosing malignancy. For each eligible lesion, one of four confidence levels was assigned to each decision as follows: "1" as 'probably not present'; "2" as 'possibly present'; "3" as 'probably present'; and "4" as 'definitely present'. Lesions that showed only arterial hypervascularization were regarded as category 2 if they showed no delayed capsular enhancement, mosaic pattern, or moderate hyperintensity on T2-weighted imaging (category 3). During gadolinium-enhanced MRI, 115 of the total 157 HCCs showed early enhancement with rapid washout, and the remaining 42 appeared as hyperintense nodules during arterial phase imaging and were occult during portal and equilibrium phase imaging. To achieve an accurate correlation between the findings of the scored lesions and the findings of the reference standard, each observer made markings over all lesions with arrows on the PACS monitor. After the two observers conducted the two review sessions, the study coordinator, who was not involved in image interpretation, compared the scoring results of each of the observers to a gold standard, and then a possible explanation was formulated for any false-positive or false-negative findings.

Based on the reviews submitted by the two observers, an alternative-free response receiver operating characteristic (ROC) curve analysis was performed on a lesion-by-lesion basis (14). For each imaging set (T2^*W-GRE images, T2W-TSE images, and combined), an alternative-free response ROC curve was fitted to each observer's confidence rating data using a maximum likelihood estimation program (ROCKIT 0.9B; courtesy of Metz CE, University of Chicago, IL, 1998) (15). The area under the alternative-free response ROC curve (Az) was calculated to determine the diagnostic accuracy of each image, each observer, and lesion type. The differences between the imaging sets with regard to the area under the alternative-free response ROC curves were compared using a two-tailed Student's t test for paired data. The sensitivities for each image and for each observer, as well as according to the lesion type, were then calculated. The sensitivity was defined as the number of true positive diagnoses, using a confidence level of 3 or 4. The sensitivities and positive predictive values for each image set were then compared using the McNemar test. A two-tailed p value of less than 0.05 was considered statistically significantly. We also calculated the 95% confidence interval (CI) to determine a range of plausible sensitivity differences.

In order to assess interobserver agreement for evaluating the two images and the combined approach, we calculated the kappa statistic for multiple observers (16). Kappa values of less than 0.20 indicated positive, but poor, agreement, those from 0.21-0.40 indicated fair agreement, those from 0.41-0.60 indicated moderate agreement, those from 0.61-0.80 indicated good agreement, and those greater than 0.81 indicated excellent agreement.