To assess the diagnostic performance of gadobenate dimeglumine-enhanced 3D gradient echo (3D-GRE) magnetic resonance imaging (MRI) for the detection of hepatic hypovascular metastases.
To assess the diagnostic performance of gadobenate dimeglumine-enhanced 3D gradient echo (3D-GRE) magnetic resonance imaging (MRI) for the detection of hepatic hypovascular metastases.
We retrospectively analyzed the initial radiologic reports of MRI of 41 patients with suspected hepatic metastases. Seventy-nine metastatic lesions were confirmed by histopathology or intraoperative ultrasound (IOUS). The sensitivity and positive predictive values for the diagnosis of hepatic metastasis were determined among each MRI set (hepatobiliary phase, precontrast images, dynamic imaging). The diagnostic performance of dynamic image set and combined dynamic and hepatobiliary image set was also evaluated by two radiologists using alternative free response receiver operating characteristic (ROC) analysis.
The overall detection rate and positive predictive value of MR were 96.2% (76/79) and 96.2% (76/79), respectively. Images obtained with hepatobiliary phase 3D-GRE showed a significantly better detection rate compared to those with precontrast sequences or dynamic imaging (P = 0.008 and 0.016, respectively). Regarding lesions 1 cm or smaller, the detection rate was 90.3% (28/31). Each reader showed a higher Az value of the combined hepatobiliary image set than those of the dynamic image set.
3D-GRE MRI using a hepatobiliary contrast agent is an accurate tool in the detection of hepatic hypovascular metastases and improves detection rate compared with precontrast and dynamic imaging. J. Magn. Reson. Imaging 2010;31:571–578. © 2010 Wiley-Liss, Inc.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
HEPATIC RESECTION of metastasis improves the long-term survival of patients with colorectal cancer (1–4). The success of surgery or tumor ablation therapies mainly depends on the accuracy of the preoperative detection of the exact number and location of metastatic lesions. A number of imaging modalities, including computed tomography (CT), CT during arteriography, and contrast-enhanced magnetic resonance imaging (MRI) are available for the detection of hepatic metastases (5–10). Among those, SPIO-enhanced MRI has been reported to be an accurate noninvasive examination technique for the evaluation of hepatic metastasis (6, 7, 11, 12); however, it was also shown to have a limitation in the detection of hepatic metastases smaller than 1 cm (5, 11, 13). Furthermore, gadolinium-enhanced dynamic imaging has also been reported to have difficulty in detecting and characterizing small hepatic metastases (14, 15).
Gadobenate dimeglumine (MultiHance, Gd-BOPTA; Bracco, Milan, Italy)-enhanced MRI has been reported to improve the detection of focal hepatic lesions due to its high T1-relaxing effect and hepatocyte-binding capability (16–19). Specifically, this hepatobiliary contrast agent provides not only dynamic imaging capability, but also information from hepatic parenchymal enhancement during the hepatobiliary phase. Currently available 3D gradient echo (3D-GRE) liver imaging in combination with parallel acquisition techniques has made it possible to image the entire liver with high spatial and temporal resolution (16, 20–25). Thus, we postulated that MRI with 3D-GRE and hepatobiliary contrast agent might be a useful imaging technique for detecting liver metastasis.
The purpose of this study, therefore, was to evaluate the diagnostic accuracy of gadobenate dimeglumine-enhanced 3D-GRE MRI in the detection of hepatic hypovascular metastasis and then to determine whether combined dynamic and hepatobiliary phase imaging improves the performance of detecting metastases.
The protocol for this study was approved by the Institutional Review Board of our institution and informed consent for this retrospective study was waived. From June 2005 to March 2007, 160 consecutive patients with hepatic metastases from extrahepatic abdominal cancers underwent Gd-BOPTA-enhanced liver MRI which included hepatobiliary-phase images. Among them, 41 patients (24 men and 17 women) who underwent hepatic resection for the management of hepatic metastases were included in this retrospective study. A total of 119 patients were excluded from our study because they had undergone chemotherapy or were not eligible for surgery. The mean age was 58 years (range, 35–78 years). A total of 79 metastases were detected in the reference standard: 71 lesions from colorectal cancer in 35 patients, five lesions from stomach cancer in four patients, one lesion from breast cancer in one patient, and two lesions from renal leiomyosarcoma in one patient.
At the time of surgery, all patients underwent thorough exploration before hepatic resection. The extent of hepatic disease was assessed by means of bimanual palpation, followed by intraoperative ultrasound (IOUS) after complete hepatic mobilization. Each IOUS examination was performed by a gastrointestinal radiologist (one of two radiologists with at least 6 years of experience in IOUS) or a surgeon (10 years of experience in IOUS) using the Aloka 2000 system (Aloka, Tokyo, Japan) with a 7.5-MHz transducer tailored for IOUS procedures. The liver was evaluated with knowledge of the MRI findings for the number of lesions, hepatic segmental localization, and the relationship of the lesions to the hepatic veins, inferior vena cava (IVC), portal vein branches, and hepatic hilum. In the 41 patients, 79 hepatic metastases were diagnosed by histopathologic examinations following surgical resections (n = 55) or IOUS before radiofrequency ablation (n = 24). The number of lesions for each patient ranged from one to five (mean = 2.0), and lesion size ranged from 4–81 mm (mean = 20.5 mm). Thirty-one lesions were ≤1 cm in diameter, and 48 were > 1cm in diameter. The time interval between MRI and surgery ranged from one to 41 days (mean = 13.9 days). Follow-up CT after surgery or intraoperative radiofrequency ablation within 6 months was also used as a reference.
All MRI examinations were performed using a 1.5 T imaging system (Achieva 1.5T Nova Dua, Philips Medical Systems, Best, Netherlands) equipped with a phased-array coil (Synergy; Philips Medical Systems). Antiperistaltic agents or oral contrast agents were not used.
The MRI protocol consisted of a breath-hold transverse T1-weighted in- and out-of-phase 2D gradient echo (GRE) sequence (TR/in-phase TE, 180/4.6 msec; out-of-phase TE = 2.3 msec; flip angle, 90°; field of view [FOV], 32–38 × 25–29 cm; matrix, 256 × 256; section thickness, 7 mm; slice spacing, 7.7 mm; one signal acquired; number of slices = 24); T2-weighted single-shot turbo spin echo (TR/TE, 452/80; FOV, 32–36 × 25–29 cm; matrix, 288 × 230; section thickness, 7 mm; slice spacing, 5 mm; scan slices were overlapped by 2 mm using an interleaved acquisition technique) with spectral fat suppression and respiratory triggering technique; and a breath-hold transverse 3D-GRE (TR/TE, 4.7/2.3 msec; flip angle, 15°; FOV, 32–36 × 25–36 cm; matrix, 320 × 224; section thickness, 3 mm; no gap; acquisition time, 23 sec).
Contrast-enhanced MRI was performed using a breath-hold 3D-GRE sequence following an intravenous bolus of 0.1 mmol Gd-BOPTA (MultiHance, Bracco) per kg of body weight followed by a saline flush of 30 mL. This sequence was repeated four times with data acquisition in the hepatic arterial, portal venous, equilibrium, and hepatobiliary phases. To determine the optimal timing for the hepatic arterial phase, fluoroscopic bolus detection technique was used in all patients (Bolus-Trak, Philips Medical Systems). Portal venous and equilibrium phase images were obtained ≈20–30 seconds after the acquisition of the prior phase images. Hepatobiliary phase images were obtained with a mean delay time of 138 minutes, ranging from 107–195 minutes. An automatic injection system (Spectris MR injection system, Medrad Europe, Maastricht, Netherlands) operating at an injection rate of 2 mL/sec was used. The actual sequence was started manually when the fluoroscopic sequence revealed that the contrast material bolus had reached the abdominal aorta.
Interpretations from original MR reports and surgical reports were recorded retrospectively by a single author (J.Y.C.) not involved in the clinical interpretation. The original MR reports were made by one of two experienced radiologists (26 cases by M.J.K. and 15 cases by M.S.P.), each with greater than 8 years of experience in hepatobiliary imaging. Information included the site of metastatic lesions (Couinaud segment), size (maximum diameter), and surgical results. The results were collected and compared with pathologic reports on a lesion-by-lesion basis.
To compare the sensitivity of each MR sequence, two other radiologists (J.Y.C. and J.S.L.) retrospectively reviewed the precontrast T1- and T2-weighted images, dynamic images, and hepatobiliary phase images of each patient with knowledge of the reference standard, and provided a possible explanation for the causes of the false-positive and false-negative findings. Retrospective interpretation for detecting metastases was also performed by two radiologists (7 and 5 years of experience) to assess the diagnostic performance of MRI. The readers reviewed dynamic image set (precontrast T1- and T2-weighted images and dynamic images) first and then combined the hepatobiliary image set (precontrast images, dynamic image set, and hepatobiliary phase) with an interval of 4 weeks. The readers recorded the presence or absence of metastatic hepatic lesions and assigned the following confidence levels to their observations: 1, not metastasis; 2, possibly not metastasis; 3, indeterminate; 4, probably metastasis; 5, definitely metastasis. The readers were aware that for statistical analysis purposes only scores of 4 and 5 would be considered as the presence of metastases. The readers also referred to the segment in which they suspected the metastasis. A slightly hyperintense lesion on T2-weighted images and a hypointense lesion on T1-weighted images were regarded as possibly metastatic lesions. The shape, border, enhancement pattern, and the internal texture of the lesions were also used as references. All images were reviewed using a local picture archiving and communication system (PACS; GE Medical Systems, Milwaukee, WI).
Sensitivity and positive predictive value (PPV) of MR for detecting hepatic metastases were calculated according to the lesion size on the basis of original MR reports. The McNemar test was used for statistical significance of any difference among MR sequences (precontrast, dynamic, and HBP sequences). An alternative-free response receiver operating characteristic (ROC) curve was fitted to each reader's confidence scoring on the basis of retrospective interpretation. The diagnostic accuracies of the dynamic image set and the combined hepatobiliary image set were determined by calculating the area under each reader-specific ROC curve (Az). Each reader's performance in diagnosing hepatic metastases was assessed using the area (Az) under the ROC curve. All statistical computations were performed using statistical software (MedCalc Software, v. 6.15.000; Mariakerke, Belgium) and the results were considered to indicate a statistically significant difference if associated with a P-value of less than 0.05. Interobserver variability for detecting hepatic metastases was not applicable because kappa values were not reliable due to a heavy concentration of scores in one category (26, 27).
Sensitivity was defined as the number of correct diagnoses of metastatic lesions divided by the number of hepatic metastases identified at IOUS, histopathologic examinations, and follow-up imaging. A false-positive diagnosis was defined as when one or more lesions were identified in a patient by a reviewer when no lesion was present. Sensitivities and PPV were calculated on a lesion-by-lesion basis.
All metastatic lesions were hypovascular on dynamic MRI. The number of metastases in each patient ranged from 1–5 (mean 2.0). MRI correctly depicted 76 metastatic lesions (Fig. 1). The overall detection rate of metastatic lesions at MR was 96.2% (76 of 79), and the positive predictive value was 96.2% (76 of 79) (Table 1). Regarding lesions 1 cm in diameter or smaller, the detection rate at MR was 90.3% (28 of 31). 3D-GRE images obtained at the hepatobiliary phase showed a significantly better detection rate compared to those with precontrast or dynamic images (P = 0.008 and 0.016, respectively) (Table 1). Six metastatic lesions in four patients were detected only on hepatobiliary phase images (Fig. 2). All of these lesions were metastases from colorectal cancer and five metastatic lesions were smaller than 1 cm. One metastatic lesion observed on hepatobiliary phase images but not seen on precontrast and dynamic MR images was not detected on IOUS (Fig. 3). This lesion was subsequently confirmed as a true metastatic lesion by surgical resection of the lesion after it had grown. Nine metastatic lesions were smaller than 1 cm and the other three lesions were between 1–2 cm.
|Initial interpretation||HBP||Precontrast T1&T2||Dynamic|
|All lesions||Sensitivity||76/79 (96.2%)||76/79 (96.2%)||68/79 (86.1%)||69/79 (87.3%)|
|PPV||76/79 (96.2%)||76/79 (96.2%)||68/68 (100%)||69/70 (98.6%)|
|Lesions ≤ 1 cm in diameter||Sensitivity||28/31 (90.3%)||28/31 (90.3%)||22/31 (70.9%)||22/31 (70.9%)|
|PPV||28/30 (93.3%)||28/30 (93.3%)||22/22 (100%)||22/23 (95.6%)|
|Lesions > 1 cm in diameter||Sensitivity||48/48 (100%)||48/48 (100%)||46/48 (95.8%)||47/48 (97.9%)|
|PPV||48/48 (100%)||48/48 (100%)||46/46 (100%)||47/47 (100%)|
The false-positive and false-negative rates on MR examinations were 3.6% (3/82) and 3.6% (3/82), respectively. False-positive findings were caused by small hepatic vessels (n = 2) and a presumed benign nodule (n = 1). The causes of false-positive findings were determined based on retrospective review of MR images by consensus of two radiologists and the presumed benign nodule was not detected on either IOUS or 6 months follow-up CT. False-negative findings were caused by small hepatic surface location (n = 1; 0.3 cm), and lesion size less than 0.5 cm (two lesions from rectal cancer; subcapsular location of left lobe of the liver).
The area under the curve (Az value), sensitivity, and PPV of dynamic image set and hepatobiliary image set on retrospective review of the two readers are given in Table 2. Each reader showed a higher Az value in the interpretation of the combined hepatobiliary image set than in the interpretation of the dynamic image set (P = 0.010 for reader 1 and P = 0.027 for reader 2). The detection rates of dynamic image set for both readers were 82.8% for reader 1 and 78.5% for reader 2 while those of combined hepatobiliary image set were 92.4% (73/79) and 92.4% (73/79) for both readers. For lesions smaller than 1 cm, the detection rate was 83.8% (26/31) for reader 1 and 80.6% (25/31) for reader 2.
|Reader 1||Reader 2|
|Dynamic image set||Combined hepatobiliary image set||Dynamic image set||Combined hepatobiliary image set|
|Az value||0.890 [0.822,0.939]||0.941 [0.885,0.975]||0.863 [0.783, 0.921]||0.911 [0.841,0.958]|
|Sensitivity||82.8% (65/79)||92.4% (73/79)||78.5% (62/79)||92.4% (73/79)|
|PPV||95.6% (65/68)||94.8% (73/77)||96.8% (62/64)||98.6% (73/74)|
With an overall detection rate of 96.2%, the results of our study showed that 3D-GRE MRI using a hepatobiliary contrast agent is more accurate than precontrast and dynamic MRI in the detection of hepatic metastases. In detecting hepatic metastases, there is a wide range in sensitivity depending on the imaging modality and technique used. The reported sensitivities for detecting hepatic metastases ranged from 56%–98% for IOUS, 39%–92% for CT, 69%–98% for MRI, and 86%–99% for fluoro-2-D-glucose positron emission tomography (FDG PET) (5, 6, 8, 28–35). Although it has been reported that FDG PET has the highest sensitivity on a per-patient basis, MRI using a contrast agent has the highest sensitivity on a lesion-by-lesion basis with sensitivities ranging from 81%–98% (8, 28–30, 33, 35, 36). In the present study the sensitivity for the detection of hepatic metastases has one of the highest values among noninvasive imaging methods even with a high-quality reference standard.
The high sensitivity achieved on MRI in our study can be attributed to the combined use of 3D-GRE sequence and a hepatobiliary contrast agent. Without hepatobiliary phase imaging, the accuracy of MRI for the detection of metastases would be reduced to 87.3% in our study. Although gadolinium-based extracellular agents are by far the most common contrast agents used in hepatic MRI and useful for detecting hypervascular metastases, dynamic imaging using these agents has a limitation in the detection of hypovascular metastases (14, 15). However, in the hepatobiliary phase using gadobenate dimeglumine-enhanced MR, both hyper- and hypovascular metastases show hypointensity in the surrounding hepatic parenchymal enhancement. Therefore, hepatobiliary phase MR allows significantly high detection rates with increased confidence, even for lesions smaller than 1 cm in diameter. The sensitivity on gadobenate dimeglumine-enhanced hepatobiliary phase imaging was significantly higher than that on dynamic imaging in both initial and retrospective interpretation. This was because gadobenate dimeglumine-enhanced hepatobiliary phase imaging provided improved lesion-to-liver contrast and delineation of a lesion by showing a clearly demarcated margin in the strongly enhancing background liver parenchyma.
In addition to the improved lesion contrast of gadobenate dimeglumine-enhanced hepatobiliary phase imaging, the 3D technique may play an important role in the detection of small hepatic lesions. Because 3D-GRE sequence provides the ability to acquire images with thin sections (<5 mm), no gaps, fat saturation, and high signal-to-noise-ratio, high sensitivity for detecting subcentimeter sized lesions can be achieved (16, 18, 20–22). Specifically, for the gadobenate dimeglumine-enhanced images in the present study, a slice thickness of 3 mm was used without intersection gap, which contributes to the detection of subcentimeter metastatic lesions because of less partial-volume averaging (24). However, this assumption was not validated in this study because we did not compare 2D and 3D imaging intraindividually. Although SPIO-enhanced MRI has been regarded as the most accurate noninvasive imaging technique in the evaluation of hepatic metastases (7, 37–40), the imaging sequence for SPIO-enhanced MRI has been reported to be relatively thick, with a slice thickness ranging from 7–10 mm, and a 1–2-mm interslice gap, possibly affecting the sensitivity for detecting small lesions. We believe that 3D-GRE MRI using gadobenate dimeglumine has an advantage over SPIO-enhanced T2-weighted imaging with regard to section thickness and partial volume averaging in small hepatic metastases.
A standard of reference is important for assessing the diagnostic performance of imaging modalities. In our study, all lesions were confirmed by surgery, IOUS, and histopathologic examination, which are by far the most reliable standards of reference. Using this standard of reference, the detection rate for metastatic lesions 1 cm in diameter or smaller at MR was 93.3% (28 of 31), which is higher than the reported detection rate for lesions smaller than 1 cm ranging from 30%–83% (5, 11, 29, 35, 41). According to our data, the combined use of hepatobiliary contrast agent and 3D technique will enable this detection threshold to be lowered.
Although gadobenate dimeglumine-enhanced hepatobiliary phase imaging was the most sensitive sequence for detecting hepatic metastases, acquisition of hepatobiliary phase imaging together with dynamic imaging and T2-weighted imaging are recommended for the evaluation of hepatic metastases. Because not all subcentimeter sized lesions detected on hepatobiliary phase imaging are metastases, dynamic imaging and T2-weighted sequence are important for characterization of the focal lesions. In our study, three subcentimeter sized lesions (3/82 findings; 3.6%) which were not demonstrated on dynamic imaging and T2-weighted sequence were falsely positive on hepatobiliary phase images. False-negative results occur mainly with hepatic surface location and small lesions less than 5 mm in size because MRI still has a limited spatial resolution at this time.
The main disadvantage for the acquisition of hepatobiliary phase images on gadobenate dimeglumine-enhanced MR examinations is the relatively long delay time (usually 1–3 hours) required between dynamic and hepatobiliary imaging (17, 42). However, acquisition of hepatobiliary phase images seems necessary for the preoperative evaluation of patients who are candidates for hepatic resection with suspected hepatic metastases because the exact number and location of the lesion is critical. In terms of scan delay, Gd-EOB-DTPA (Primovist, Bayer Schering Pharma, Berlin), a recently approved hepatobiliary contrast agent, may provide hepatobiliary phase images with a shorter delay time of 20 minutes and may facilitate patient throughput in busy practices.
Our study has several limitations. First, histologic proof was not available for lesions treated with radiofrequency ablation. However, false-positive findings would be minimal because radiofrequency ablation was performed on the basis of typical appearances on IOUS. Second, there may have been inadvertent selection bias, because patients with limited numbers of metastatic disease underwent surgery, limiting our patient group to surgical patients. However, this method of patient selection was relevant to our study because our intention was to focus on improving the diagnostic accuracy for preoperative evaluation of hepatic metastases.
In conclusion, 3D-GRE MRI using a hepatobiliary contrast agent is an accurate tool in the detection of hepatic metastases and improves detection rate compared with precontrast and dynamic imaging in planning hepatic resection.