To assess the incremental value of hepatobiliary phase images in gadoxetate disodium-enhanced magnetic resonance imaging (MRI), and to compare diagnostic accuracy and lesion conspicuity on 10- and 20-minute delayed images for preoperative detection of hepatic metastases with subgroup analysis according to size and history of chemotherapy.
Materials and Methods:
Forty-six patients with 107 metastases who underwent surgery after gadoxetate disodium-enhanced MRI were evaluated. Four observers independently interpreted three sets: dynamic set comprising precontrast T1-, T2-weighted, and dynamic images; 10-minute set comprising dynamic set and 10-minute delayed; 20-minute set comprising 10-minute set and 20-minute delayed. Diagnostic accuracy was compared with subgroup analysis. Liver-to-lesion signal ratio (SR) was calculated using the region of interest method and compared.
Mean Az and sensitivities were significantly higher for 10- (Az = 0.894, sensitivity = 95.6%) and 20-minute (0.910, 97.2%) than dynamic set (0.813, 79.9%) (P < 0.001), with no significant difference between 10- and 20-minute sets (P = 0.140). In patients with small (≤1 cm) metastases and a history of chemotherapy, sensitivities were significantly higher with 10- (88.2%) and 20-minute (91.6%) sets than dynamic set (48.6%) (P < 0.001). SR was significantly higher for 10- and 20-minute delayed than precontrast and dynamic, with significantly higher SR on 20- than 10-minute delayed.
ACCURATE DETECTION AND LOCALIZATION of hepatic metastases is important for treatment planning and the success of therapeutic approaches for patients with primary extrahepatic malignancy (1, 2). magnetic resonance imaging (MRI) is preferred as a first-line modality for evaluating hepatic metastases (3), but its effectiveness is highly dependent on the use of optimal pulse sequences and the appropriate use of an MR contrast agent (4–6).
Gadoxetate disodium, a newly available hepatobiliary MR contrast material, has been shown to improve detection and characterization of hepatic metastases (7–9). The improvement of diagnostic accuracy by the use of gadoxetate disodium seems to rely on the availability of hepatobiliary phase imaging that provides marked improvement of lesion conspicuity due to increased lesion to liver contrast (10–12). The diagnostic efficacy of gadoxetate disodium-enhanced MRI for focal liver lesions has been demonstrated (13). To our knowledge, the added value of gadoxetate disodium-enhanced hepatobiliary phase imaging has not been specifically addressed in patients with hepatic metastasis. Although a scan delay of 20 minutes has been recommended as an optimal time for peak hepatic enhancement (13–18), some studies implied that a 10-minute delay time might be sufficient to achieve comparable results for the detection or characterization of focal liver lesions and shorter examination time (19–21). However, these studies were also performed for focal hepatic lesions with heterogeneous diseases. In addition, for the evaluation of patients with hepatic metastases, whether the efficacy of gadoxetate disodium-enhanced imaging might be affected by the lesion size and history of chemotherapy is yet to be determined.
Therefore, the aim of this study was to assess the incremental value of hepatobiliary phase images on gadoxetate disodium-enhanced MRI and to compare the diagnostic accuracy and lesion conspicuity on 10- and 20-minute delayed images for the preoperative detection of hepatic metastases from gastrointestinal malignancies with subgroup analysis according to size and history of chemotherapy.
MATERIALS AND METHODS
Patients and Reference Standard
The protocol for this retrospective study was approved by the Institutional Review Board of our institution and the requirement for informed consent was waived. In our radiology information database, we retrieved data of 265 consecutive patients who underwent preoperative gadoxetic acid-enhanced MRI between April 2008 and March 2009 for known or suspected hepatic metastases from gastrointestinal malignancies. Of these patients, 48 patients underwent hepatic resection. Of these 48 patients, two were excluded because too many (>20) newly formed metastases were observed at follow-up computed tomography (CT) examinations performed 30 and 55 days after surgery, and the sensitivity and false-negative rates could not be calculated. In addition, we included five patients who were randomly selected among 217 patients who showed suspicious features of hepatic metastasis on CT examination but not on gadoxetate disodium-enhanced MR examination or during a follow-up period of more than 1 year. We included these five patients to prevent the reviewers from assuming that all patients had hepatic metastases. Thus, we analyzed 46 patients with hepatic metastases (33 males and 13 females, mean age: 62.6 years). The average time interval between surgery and the last MR examination was 20 days (range 1–48 days).
A total of 107 metastases and 14 benign focal hepatic lesions were diagnosed according to the reference standard. For 106 metastases of 107 in 46 patients, diagnoses of metastases were based on histopathologic findings. The other one lesion was diagnosed by postoperative follow-up CT examinations which showed the presence of continuous growth. This lesion was considered a false-negative lesion on initial image interpretation by comparison with the preoperative images. In the remaining patients, there was no evidence of recurring metastasis in postoperative follow-up CT up to 3 months. Therefore, the number of metastases determined on pathologic reports was used as a reference standard for these patients. Among the 107 metastases, 90 were found in 41 colorectal cancer patients, and 17 in five gastric cancer patients. In our study there was no hepatic metastasis originating from neuroendocrine tumors that could show hypervascularity. The diameter of the metastases ranged from 0.2–8 cm (mean: 1.9 cm). The number of metastases per patient ranged from 1–8 (mean: 2.5): 18 patients with a solitary metastasis, 11 with two metastases, eight with three metastases, six with four metastases, one with five metastases, one with six metastases, and one with eight metastases. Of 46 patients who had hepatic metastases, 20 patients with 61 metastases had a history of chemotherapy prior to MRI, and the average time interval between the end of chemotherapy and the last MR examination was 14.5 days (range: 11–94 days).
Benign focal lesions consisted of 10 benign hepatic cysts and four hemangiomas. Diagnoses of 14 benign focal lesions in five patients were based on the absence of interval changes during serial follow-up of at least 6 months instead of pathologic confirmation. Benign lesions were counted only for lesions that were marked by at least one reader during the retrospective analysis.
All MRI examinations were performed on a 3.0-T MR scanner (Magnetom Trio a Tim; Syngo MR B15; Siemens Medical Solutions, Erlangen, Germany) using a six-element body phased-array coil on the anterior side of the patient and another six elements of the spine coil on the posterior side of the patient. To improve image quality, the integrated parallel imaging technique (iPAT) by means of generalized autocalibrating partially parallel acquisition (GRAPPA) with an acceleration factor of 2 was applied for all images in this study. Routine liver MR images were acquired in transverse planes using the following sequences: 2D dual-echo breath-hold T1-weighted spoiled GRE at opposed-phase (repetition time [TR] / effective echo time [TE], 140/1.22 msec) and in-phase (150/2.5) with a flip angle of 65°, one signal acquired, a matrix of 256 × 192, 7-mm slice thickness, a 0.7-mm gap and iPAT factor 2; a navigator-triggered T2-weighted turbo spin-echo (TSE) sequence with TR range of 3300–4900 and TE 73 msec, echo train length of 14, one signal acquired, a matrix of 320 × 179, superior and inferior spatial presaturation and chemical fat saturation, a 4-mm slice thickness and a 1-mm gap and iPAT factor 2; and a breath-hold, heavily T2-weighted half-Fourier acquisition turbo spin echo (HASTE) with a TR of 500 msec, a TE of 150 msec, a matrix of 320 × 179, 4-mm slice thickness, a 1-mm gap, and iPAT factor 2.
Dynamic T1-weighted MRI was performed using a transverse breath-hold 3D T1-weighted fat-suppressed spoiled gradient-recalled-echo sequence (volumetric interpolated breath-hold examination). The MR parameters included TR/TE 3.3/1.16, a flip angle of 13°, a 256 × 192 matrix, iPAT factor 2, one signal acquired, and 2-mm slice thickness using an interpolation technique. Dynamic imaging was performed following an intravenous bolus of gadoxetate disodium (Primovist, Bayer Schering Pharma, Germany; 0.025 mmol/kg of body weight) and a 20-mL saline flush at an injection rate of 1 or 2 mL/sec using an automatic infusion system (Spectris MR injection system, Medrad Europe, Maastricht, Netherlands). Four consecutive contrast-enhanced dynamic phases were obtained with 20–24-second durations for each scan. Scan delay for the arterial phase was usually 30–35 seconds, and was determined using a bolus-tracking technique or a test-bolus injection technique. Further dynamic images were obtained at 60–70 seconds (portal phase), 90–100 seconds (hepatic venous phase), and 120–150 seconds (equilibrium phase) after the injection. Additional hepatobiliary phase images were obtained at 10 and 20 minutes after injection of the contrast material.
The MR images were retrospectively and independently reviewed by four radiologists (M.J.K., M.S.P., J.Y.C. with 18, 11, and 10 years of experience in abdominal MRI, respectively, and H,T.J. with 2 years of experience in residency) who were aware of the fact that the patients had or were suspected of having hepatic metastases. The radiologists were unaware of all other information regarding the patients' history, laboratory results, findings from other imaging modalities, and the final diagnosis. They reviewed three sets of MR images with an interval of at least 2 weeks: 1) MR set 1 (dynamic set): precontrast T1-weighted images, precontrast T2-weighted images, and gadoxetic-enhanced dynamic images; 2) MR set 2 (10-min delay set): MR set 1 with 10-minute delayed phase images; 3) MR set 3 (20-min delay set): MR set 2 with 20-minute delayed phase images. We did not choose the combination of “precontrast T1- and T2-weighted and 10-minute delayed phase images” alone, because it seemed not practical. When we obtain 10-minute delayed images, we always read the images combined with dynamic imaging. Moreover, the purpose of this study was to determine the value of 10-minute and 20-minute delayed images in addition to precontrast and dynamic imaging.
Each reader recorded the presence, size (maximum diameter), and the segmental location according to the Couinaud classification of the hepatic segmental anatomy (22). Schematic templates of the sectional images of the liver were provided for the readers so that they could mark the location, size, and the number of lesions in order to make an accurate correlation of the lesions detected by each reader. Each reader indicated the possibility of metastatic lesions using the following scoring system: 1) definitely not a metastatic lesion; 2) probably not a metastatic lesion; 3) possibly a metastatic lesion; 4) probably a metastatic lesion; and 5) definitely a metastatic lesion. Lesions awarded a score of 3, 4, or 5 were defined as metastatic lesions detected by each reader. A score of 1 was assigned retrospectively when a lesion was not identified by a reviewer. On noncontrast images, a slightly hyperintense lesion on T2-weighted images and a hypointense lesion on T1-weighted images were regarded as possibly metastatic lesions. On the gadoxetic-enhanced images, the absence of contrast agent uptake and the presence of a rim enhancement were regarded as signs of metastasis. The shapes, borders, and the internal textures of the lesions were also used as references. All images were reviewed using a local picture archiving and communication system (Centricity v. 1.5 or higher; GE Healthcare, Milwaukee, WI).
For quantitative analysis, we measured the liver-to-lesion signal ratio (SR), defined by an equation as follows: (signal intensity of the liver / signal intensity of the lesion). The region of interest (ROI) was drawn on the background liver and metastatic lesions on each MR sequence (precontrast, arterial, portal, venous, equilibrium, 10-minute delayed phase, and 20-minute delayed phase). The ROI was carefully drawn on the lesion to nearly match the size of the lesion, but not to cross the border. The average ROI size of the lesion was 451.5 ± 834.3 mm2 (range: 3.14–5024 mm2). The ROI on the liver was drawn on the parenchyma nearest to the lesion with a size of ≈150 mm2.
The mean ages of men and women in our sample population were compared using the independent-sample t-test. An alternative-free response receiver operating characteristic (ROC) curve was fitted to each reader's confidence scoring on the basis of retrospective interpretation. For each observer the accuracy of the diagnosis of metastasis, determined by the area under the alternative free-response receiver operating characteristic curve (Az), was calculated and compared using latent binomial alternative free-response receiver operating characteristic analysis (MedCalc, v. 10.1.2.0; Mariakerke, Belgium). In addition, the mean Az values for each MR set were calculated to evaluate overall accuracy and differences in the mean Az values were compared by paired t-test.
Sensitivity was defined as the number of true-positive lesions (assigned with a confidence level of 3, 4, or 5) divided by the number of hepatic metastases verified by histopathologic examinations and follow-up imaging. A false-positive diagnosis was defined as identification of one or more lesions in a patient by a reader at a confidence level of 3–5 that was not subsequently verified as a metastasis. Sensitivity and positive predictive value (PPV) were calculated on a lesion-by-lesion basis. Sensitivity and PPV of each MR set for detecting hepatic metastases and mean values were calculated. The McNemar test was used to compare the sensitivities of the three sets of MR images. Fisher's protected least significant difference test was used to test the differences in the positive predictive values between the MR sets.
We performed subgroup analyses to assess whether the diagnostic accuracy is affected by lesion size and history of chemotherapy prior to MRI. Sensitivity and PPV were calculated according to the lesion size (≤1 cm, >1 cm) and history of chemotherapy. In addition, cumulative sensitivity and PPV were calculated to present a summed result of all observers. Cumulative sensitivity was defined as the sum of the number of true-positive lesions divided by the quadruple number of hepatic metastases verified by a reference standard. Cumulative PPV was calculated by adding all observers' results.
The kappa (κ) test was used to assess the interobserver agreement in terms of detecting metastatic lesions. The degrees of agreement were categorized as follows: κ value <0, poor; κ 0–0.20, slight agreement; κ 0.21–0.40, fair agreement; κ 0.41–0.60, moderate agreement; κ 0.61–0.80, substantial agreement; and κ 0.81–1, almost perfect agreement.
Comparisons of the liver-to-lesion SRs among MR sequences (precontrast, arterial, portal, venous, equilibrium, 10-minute delayed phase, and 20-minute delayed phase) were performed using one-way analyses of variance (ANOVA) with post-hoc Scheffe's tests. Comparisons between chemotherapy and nonchemotherapy groups and between 10-minute and 20-minute delayed phases in each group were performed using paired t-test. Unpaired t-tests were performed to compare SRs between patients with and without history of chemotherapy for both 10-minute and 20-minute delayed phase images. All statistical computations other than ROC analysis were performed using statistical software (SPSS, v. 17.0.1, Chicago, IL), and P-values <0.05 were considered statistically significant.
Analyses of All Lesions
For all-lesion analysis, the calculated Az values for each observer with the three MR sets are shown in Table 1. All four observers achieved significantly higher Az values with MR set 3 (20-min delay, mean Az = 0.910 ± 0.013, mean ± one standard deviation) than with MR set 1 (dynamic set, mean Az = 0.813 ± 0.021), (P < 0.001). Two observers achieved significantly higher diagnostic accuracy with MR set 2 (10 min delay) than with MR set 1, and the mean Az values of the four observers were significantly higher for MR set 2 (0.894 ± 0.015, P < 0.001) than for MR set 1. There were no significant differences in Az values between MR set 2 and MR set 3 for any observer.
Table 1. Calculated Az Values for Each MR Set for All Lesions
For all 107 metastatic lesions, each observer's sensitivity and PPV for each MR set are presented in Table 2. All observers showed significantly higher sensitivities with MR set 2 (mean, 95.6%) and MR set 3 (mean, 97.2%) than with MR set 1 (mean, 79.9%) (P < 0.05 for both). There were no significant differences in the sensitivity between MR set 2 and MR set 3 for all observers (P = 0.656). The mean PPVs of all MR sets were not significantly different for any observers.
Table 2. Sensitivities and PPV for Detecting Hepatic Metastasis for Each MR Set
The cumulative sensitivities and PPVs according to the lesion size and history of chemotherapy before MRI are shown in Table 3 by observer and MR set. For 49 small (≤1 cm) metastases, sensitivities were significantly higher with MR set 2 (88.2%, P < 0.001) and MR set 3 (91.6%, P < 0.001) than with MR set 1 (48.6%) (Fig. 1). There was no significant difference in sensitivity between MR set 2 and set 3 (P = 0.289). For larger (>1 cm) metastases, sensitivities with each MR set were not significantly different. PPVs for all observers were not significantly different among the three MR sets for lesions ≤1 cm and lesions >1 cm.
Table 3. Sensitivity and PPV According to Lesion Size and Chemotherapy Prior to MRI
Twenty of 46 patients with metastases had undergone chemotherapy prior to preoperative MR examinations. For 61 metastases in those patients, sensitivities were significantly higher with MR set 2 (95.5%, P < 0.001) and MR set 3 (96.3%, P < 0.001) than with MR set 1 (77.9%). There was no significant difference in sensitivity between MR set 2 and MR set 3 (P = 0.625) (Fig. 2). Similarly, for 46 metastases in patients without history of chemotherapy, sensitivities were significantly higher with MR set 2 (95.1%, P < 0.001) and MR set 3 (97.8%, P < 0.001) than with MR set 1 (82.0%), and there was no significant difference in sensitivity between MR set 2 and MR set 3 (P = 0.218). PPV for all observers was not significantly different among the three MR sets regardless of previous chemotherapy.
Quantitative analysis and interobserver agreement
The means and ranges of the liver-to-lesion SRs on each phase image are presented in Fig. 3. Liver-to-lesion SR was significantly higher on 10-minute (2.154 ± 0.488) and 20-minute (2.393 ± 0.567) delayed phase images than on precontrast (1.365 ± 0.256), arterial (1.281 ± 0.443), portal (1.677 ± 0.515), hepatic venous (1.686 ± 0.444), and equilibrium (1.735 ± 0.445) phase images, with significantly higher SR on 20-minute delayed images than on 10-minute delayed images. With the exception of arterial phase, all dynamic images showed significantly higher liver-to-lesion SR than precontrast phase images. Comparisons of liver-to-lesion SR between 10-minute and 20-minute delayed phase images according to history of chemotherapy are presented in Fig. 4. Mean liver-to-lesion SR was significantly higher in patients without previous chemotherapy (2.285 ± 0.468) than in those with chemotherapy (2.058 ± 0.484, P = 0.016) on 10-minute delayed images, but was comparable between the two groups on 20-minute delayed images (2.315 ± 0.603 in chemotherapy group and 2.498 ± 0.501 in nonchemotherapy group; P = 0.097). κ values for interobserver agreement for each MR set were estimated as 0.537–0.908, representing moderate to good interobserver agreement (Table 4).
Table 4. Interobserver Agreement for Each MR Set
MR set 1
MR set 2
MR set 3
MR set 1, precontrast T1- and T2-weighted and dynamic; MR set 2, dynamic set with 10-min delayed images; MR set 3, 10-min delay set with 20-min delayed images.
Data are κ values.
The results of our study indicate that the detection of hepatic metastases from gastrointestinal malignancies can be significantly improved by adding gadoxetate disodium-enhanced 10-minute or 20-minute delayed hepatobiliary phase images to precontrast and dynamic MR images. There were no significant differences in Az values between 10- and 20-minute delayed sets for all readers, although quantitative analysis showed that the liver-to-lesion SR was significantly higher with 20-minute delayed images.
With overall sensitivity of 97.2% and PPV of 96.3% for the detection of hepatic metastases, the results of our study on gadoxetate disodium-enhanced MRI including hepatobiliary phase images were better than those of prior studies using superparamagnetic iron oxides as a contrast material for the detection of hepatic metastases (23–26). As the subgroup analysis according to lesion size showed, the improvement in diagnostic accuracy could be attributed to improved detection of small (≤1 cm) hepatic metastases that can be frequently missed on conventional MRI (4). The diagnosis of small or early hepatic metastasis is important to improve outcome in patients treated by chemotherapy, and is also associated with better survival in surgically treated patients (27, 28). In our study, the sensitivities for the detection of small metastases with 10- and 20-minute delayed hepatobiliary phase images were 88.2% and 91.6%, respectively. Lesions that were identified only on hepatobiliary phase images were mostly small and showed some enhancement on dynamic images. However, those lesions showed marked hypointensity in the strongly enhanced background liver parenchyma (29). The improvement in lesion conspicuity on hepatobiliary phase imaging is supported by the significant increase in liver-to-lesion SR on both 10-minute and 20-minute delayed hepatobiliary phase images.
To achieve appropriate hepatic parenchymal enhancement, it is important to wait until the contrast agent is taken up by hepatocytes. For gadobenate dimeglumine-enhanced MRI, the reported imaging window for the acquisition of hepatobiliary phase images is 40–180 minutes after injection of the contrast material (30–32). In contrast, gadoxetate disodium-enhanced hepatobiliary phase images can be obtained after a shorter delay time of 20 minutes (13, 17, 29, 33–35), and some studies have implied that a scan delay for the acquisition of gadoxetate disodium-enhanced hepatobiliary phase images can be even further reduced (20, 21). In a study including 23 patients with hepatic metastases, Reimer et al (20) showed that lesion-to-liver contrast-to-noise ratios of hepatic metastases are comparable between gadoxetate disodium-enhanced 10-minute and 20-minute delayed images but did not compare the sensitivity for the two image sets. Motosugi et al (21) also showed that detection of various focal hepatic lesions on 10-minute delayed images was comparable to that of 20-minute delayed images, when the latter images were used as reference standard. Our study with a larger series showed that the accuracy of the detection of hepatic metastases was comparable between 10-minute and 20-minute delayed images, although the quantitative analysis showed significantly higher lesion-to-liver SR on 20-minute delayed images. Based on our results, we currently use a scan delay of ≈15 minutes to allow the acquisition of T2-weighted, heavily T2-weighted, and diffusion-weighted images between the dynamic and hepatobiliary phase imaging (36, 37). With this scan delay, the interval between dynamic imaging and acquisition of hepatobiliary phase image could be effectively utilized to shorten the total scan time without lowering the image quality (38, 39).
In our study the sensitivity for the dynamic set (MR set 1) was relatively lower in the chemotherapy group (77.9%) than in the nonchemotherapy group (82.6%). This might be attributed to the reduction of typical perilesional enhancement in standard dynamic images caused by reduction of tumor growth, inflammatory changes, and angiogenesis for liver metastases from colorectal cancer (40). In contrast to the dynamic set, our results showed that the sensitivities for both 10-minute and 20-minute delayed hepatobiliary phase images were comparable between the chemotherapy and nonchemotherapy groups. Although the liver-to-lesion SR for 10-minute delayed phase images was relatively lower in the chemotherapy group than the nonchemotherapy group, the SR was comparable between the two groups on 20-minute delayed images. These results suggest that chemotherapy-induced changes may slightly decrease the lesion conspicuity on 10-minute delayed images without significantly reducing lesion detection. Although the effect of chemotherapy on hepatobiliary MR images is not well understood, chemotherapy is known to cause variable hepatic parenchymal damages such as hepatic steatosis, steatohepatitis, and sinusoidal obstructive syndrome (41, 42). Therefore, the effect of chemotherapy may affect the uptake of gadoxetate disodium through organic anion transporters to diminish the hepatic enhancement of hepatobiliary phase (43). Despite this, the results of our study suggest that use of a scan delay of 10–15 minutes might be sufficient, even in patients with a history of chemotherapy.
There are some potential limitations in our study. First, our study was performed retrospectively, and the study population mostly consisted of patients with hepatic metastasis. Although five patients without hepatic metastasis were included, many other patients believed not to have hepatic metastasis were not included. However, because our purpose was to compare the relative diagnostic accuracy between three different image sets, we believe that the study design is acceptable. Second, we did not include patients who were not eligible for hepatic resection to establish a standard reference, and thus, we could not estimate the sensitivity and specificity in patients who did not undergo surgery. Third, the study population was limited to colorectal (41 patients) and gastric cancer (five patients). It might be a critical issue for applying our results to metastases of other origin, especially in hepatic metastasis from breast cancer. However, it is thought that the study population limited to colorectal and gastric cancer will be clinically relevant, because hepatic metastasis from malignancies other than gastrointestinal malignancy can rarely be treated by surgery. Unlike colorectal metastasis, the significance of hepatic resection for gastric metastasis has been controversial. However, recent studies showed that surgical treatment for hepatic metastases from gastric cancer could be a beneficial option (44, 45). Finally, the diagnostic accuracy was not assessed by delay time, such as 15 minutes after injection of the contrast material. Actually, we currently obtain hepatobiliary phase images with a scan delay of ≈15 minutes after the injection; during this delay period we obtain moderate and heavily T2-weighted images as well as diffusion-weighted images, which require approximately the same amount of time.
In conclusion, gadoxetate disodium-enhanced hepatobiliary phase images obtained 10- and 20-minutes after injection significantly improve the detection of hepatic metastases compared with precontrast and dynamic MRI. The 10-minute delayed phase images are comparable to 20-minute delayed images in terms of diagnostic accuracy, although significantly higher liver-to-lesion SR can be obtained by using 20-minute delayed images, especially in patients with a history of chemotherapy.