Min Jung Park, MD, PhD 2 Young Kon Kim, MD, PhD Sanghyeok Lim, MD Hyunchul Rhim, MD, PhD Won Jae Lee, MD, PhD. Purpose: Materials and Methods:

Transcription

1 Note: This copy is for your personal non-commercial use only. To order presentation-ready copies for distribution to your colleagues or clients, contact us at Original Research n Gastrointestinal Imaging Min Jung Park, MD, PhD 2 Young Kon Kim, MD, PhD Sanghyeok Lim, MD Hyunchul Rhim, MD, PhD Won Jae Lee, MD, PhD Hilar Cholangiocarcinoma: Value of Adding DW Imaging to Gadoxetic Acid enhanced MR Imaging with MR Cholangiopancreatography for Preoperative Evaluation 1 Purpose: Materials and Methods: To assess the benefit of adding diffusion-weighted (DW) imaging to gadoxetic acid enhanced magnetic resonance (MR) imaging and MR cholangiopancreatography in the preoperative evaluation of hilar cholangiocarcinoma. The institutional review board approved this retrospective study and waived the requirement for informed consent. The study included 52 patients (36 men, 16 women; mean age, 63.4 years) with surgically confirmed hilar cholangiocarcinoma who underwent gadoxetic acid enhanced MR imaging and DW imaging at 3.0 T between August 2010 and December Two observers independently reviewed two image sets a gadoxetic acid set, including images from MR cholangiopancreatography, and a combined gadoxetic acid set and DW imaging set to evaluate the tumor involvement of each biliary confluence and vascular and liver invasion by using receiver operating characteristic (ROC) curve analysis. 1 From the Department of Radiology and Center for Imaging Science, Samsung Medical Center, Sungkyunkwan University School of Medicine, 50 Ilwon-dong, Gangnam-gu, Seoul , Republic of Korea. Received January 2, 2013; revision requested February 1; revision received April 3; accepted April 16; final version accepted August 13. Address correspondence to Y.K.K. ( jmyr@dreamwiz.com). 2 Current address: Department of Radiology, Yonsei University College of Medicine, Seoul, Korea. q RSNA, 2013 Results: Conclusion: For each observer, area under the ROC curve (A z ) values for tumor involvement of the biliary confluence were and for the gadoxetic acid set and and for the combined set, respectively (P..05). For detecting 105 biliary confluences with tumor involvement, the sensitivities with the combined set (97.1% [102 of 105] and 98.1% [103 of 105]) were higher than those with the gadoxetic acid set (91.4% [96 of 105] for both observers) (P =.029 and P =.016), although the specificities were similar with both image sets (P..05). For the detection of liver invasion, the combined set (75.0% [15 of 20] for both observers) yielded better sensitivity than the gadoxetic acid set (50.0% [10 of 20] and 45.0% [nine of 20]) (P =.016 and P =.031). For evaluation of vascular invasion, the two image sets showed similar diagnostic performance. In the preoperative evaluation of hilar cholangiocarcinoma, the addition of DW imaging to gadoxetic acid enhanced MR imaging could improve sensitivity in the evaluation of tumor extent along the bile duct and liver invasion. q RSNA, 2013 Online supplemental material is available for this article. 768 radiology.rsna.org n Radiology: Volume 270: Number 3 March 2014

2 Hilar cholangiocarcinoma is an adenocarcinoma that occurs at the confluence of the right and left hepatic ducts at the liver hilum. It can be cured only with surgical resection. However, because of the locally advanced nature of this disease at presentation and its association with critical hepatobiliary structures, surgical resectability rates had traditionally been low (,30%) (1), and surgery rendered limited benefit (2 4). In recent decades, with the improved resolution of hepatic imaging, partial Advances in Knowledge nn In the preoperative evaluation of the biliary confluence in patients with hilar cholangiocarcinoma, gadoxetic acid enhanced MR imaging yielded diagnostic performance, with mean area under the receiver operating characteristic curve of (standard deviation) for observer 1 and for observer 2. nn The combination of gadoxetic acid enhanced MR imaging and diffusion-weighted (DW) imaging yields better sensitivity in the detection of tumor involvement in the biliary confluence than gadoxetic acid enhanced MR imaging alone (97.1% [102 of 105] vs 91.4% [96 of 105] for observer 1 [P =.029] and 98.1% [103 of 105] vs 91.4% [96 of 105] for observer 2 [P =.016]), although specificities were similar for both image sets (P..99). nn In the detection of liver invasion by hilar cholangiocarcinoma, the combination of gadoxetic acid enhanced MR imaging and DW imaging yielded better sensitivity than gadoxetic acid enhanced MR imaging alone (75.0% [15 of 20] vs 50.0% [10 of 20] for observer 1 [P =.016] and 75.0% [15 of 20] vs 45.0% [nine of 20] for observer 2 [P =.031]). hepatectomy has been added to the traditional surgical strategy, which consisted of resection of the bile duct with biliary-enteric anastomosis of the intrahepatic ducts. Accordingly, currently approximately 40% of patients with hilar cholangiocarcinoma may undergo surgical resection with curative intent, which has achieved 5-year survival rates of 11% 40% (1,5 9). The pathologic margin status is a strong risk factor influencing recurrence-free survival after resection of hilar cholangiocarcinoma (9 12). Therefore, the precise localization and determination of tumor extent along the bile duct, as well as in adjacent organs and vessels, is of paramount importance for surgical planning to achieve microscopically complete resection of hilar cholangiocarcinoma. Because of recent advances in magnetic resonance (MR) imaging technology, particularly the introduction of multichannel surface receiver coils and parallel imaging, diffusion-weighted (DW) imaging has been increasingly applied to liver MR imaging with improved image quality. To our knowledge, only one study (13) has documented the usefulness of DW imaging in the diagnosis of extrahepatic bile duct cancer. Because it offers high contrast between tumor and surrounding normal tissue, we hypothesized that DW imaging would add value to conventional liver MR imaging sequences, including MR cholangiopancreatography, in assessing the extent of hilar cholangiocarcinoma. Therefore, we performed this study to assess the incremental benefit of adding DW imaging to gadoxetic acid enhanced MR imaging and MR cholangiopancreatography in the preoperative Implication for Patient Care nn For the evaluation of hilar cholangiocarcinoma, DW imaging should be routinely included in the gadoxetic acid enhanced MR imaging protocol for detection of tumor extent along the bile duct and for detection of liver invasion. evaluation of hilar cholangiocarcinoma, with emphasis on tumor extent. Materials and Methods Study Population This retrospective study was approved by the institutional review board of Samsung Medical Center (Seoul, Korea), and the requirement for informed consent was waived. We searched our hospital s surgical database for the time period between August 2010 and December 2011 using the search phrase hilar cholangiocarcinoma and found 67 patients. The inclusion criteria were as follows: (a) preoperative gadoxetic acid enhanced liver MR imaging, including DW imaging and MR cholangiopancreatography; (b) surgery at our institution within 4 weeks after MR imaging; and (c) histologically confirmed hilar cholangiocarcinoma. Patients were excluded because of an absence of preoperative liver MR imaging (n = 9) and a time interval of more than 4 weeks between MR imaging and surgery (n = 6). Finally, 52 patients (mean age, 63.4 years; range, years; 36 men [mean age, 64 years; range, years] and 16 women [mean age, 62 years; range, years]) formed our study population. Among these patients, two had undergone percutaneous transhepatic biliary decompression Published online before print /radiol Content codes: Radiology 2014; 270: Abbreviations: ADC = apparent diffusion coefficient A z = area under the receiver operating characteristic curve CI = confidence interval DW = diffusion weighted HBP = hepatobiliary phase Author contributions: Guarantors of integrity of entire study, M.J.P., Y.K.K., S.L.; study concepts/study design or data acquisition or data analysis/interpretation, all authors; manuscript drafting or manuscript revision for important intellectual content, all authors; manuscript final version approval, all authors; literature research, M.J.P., Y.K.K., S.L.; clinical studies, all authors; statistical analysis, M.J.P., S.L.; and manuscript editing, all authors Conflicts of interest are listed at the end of this article. Radiology: Volume 270: Number 3 March 2014 n radiology.rsna.org 769

3 Table 1 MR Imaging Sequences and Parameters Imaging Sequence Repetition Time (msec)/echo Time (msec) Flip Angle (degrees) Section Thickness (mm) Matrix Size Bandwidth (Hz/pixel) Field of View (cm) Acquisition Time (sec) No. of Signals Acquired T1-weighted 3D dual GRE 3.5/ / Breath-hold multishot T2 weighted 1623/ / / Respiration-triggered single-shot 1156/ / heavily T2 weighted DW imaging 1600/ / T1-weighted 3D GRE 3.1/ / Breath-hold 2D MR 6417/ / cholangiopancreatography Navigator-triggered 3D MR cholangiopancreatography 1673/ / Note. GRE = gradient echo, 3D = three-dimensional, 2D = two-dimensional. procedures before MR imaging. The mean interval between the MR imaging examination and surgery was 12.4 days (range, 5 21 days). MR Imaging Examination All MR images were acquired by using a 3.0-T MR imaging system (Intera Achieva; Philips Healthcare, Best, the Netherlands) with a 16-channel phased-array receiver coil. MR imaging sequences are summarized in Table 1. MR cholangiopancreatography was performed by using two different methods a breath-hold two-dimensional single-projection turbo spin-echo MR cholangiopancreatography sequence performed at various angles (625, 15, 0 ) to the coronal plane and a navigator-triggered three-dimensional turbo spin-echo MR cholangiopancreatography sequence. DW imaging was performed by using respiratory-triggered single-shot echo planar imaging with b values of 0, 100, and 800 sec/mm 2. A spectral attenuated inversion-recovery technique was used for fat suppression at DW imaging. The apparent diffusion coefficient (ADC) was calculated with a monoexponential function by using b values of 100 and 800 sec/mm 2. For gadoxetic acid enhanced imaging, unenhanced, arterial phase, portal phase, 3-minute late phase, and 20-minute hepatobiliary phase (HBP) imaging was performed by using a T1-weighted three-dimensional turbo-field-echo sequence with a spectral attenuated inversion-recovery fatsuppression technique. The time for the arterial phase imaging was determined by using the MR fluoroscopic bolus detection technique. By using a power injector, mmol of the gadoxetic acid (Primovist; Bayer Healthcare, Berlin, Germany) per kilogram of body weight was administered intravenously at 1 ml/ sec, followed by a 20-mL saline flush. Image Analysis Two gastrointestinal radiologists (S.L. [observer 1] and Y.K.K. [observer 2], with 5 and 12 years of experience in liver MR imaging study interpretation, respectively), independently reviewed MR images for tumor involvement of each bile duct confluence and vascular and liver parenchymal invasion. They were blinded to the surgical and pathologic findings but were informed that the study population had hilar cholangiocarcinoma. The review consisted of separate sessions for the two image sets, with a 4-week interval. At the first session, the observers were randomly shown either (a) the gadoxetic acid set (precontrast T1- and T2-weighted images and gadoxetic acid enhanced arterial, portal, 3-minute delay, and 20-minute HBP images with MR cholangiopancreatography images) or (b) the combined gadoxetic acid set and DW images (DW images of b = 100 and 800 sec/mm 2 with ADC maps). In the second session, observers reviewed the image set they had not previously reviewed. To evaluate bile duct tumor involvement, the observers assigned a confidence level for each confluence (primary confluence and right and left secondary confluences) on the basis of a five-point scale as follows: A score of 1 indicated probably no tumor involvement; a score of 2, possibly no tumor involvement; a score of 3, indeterminate; a score of 4, probable tumor involvement; and a score of 5, definite tumor involvement. In the evaluation of vascular and liver parenchymal invasion, the observers assigned a classification of presence or absence of invasion. The MR imaging interpretation criteria for tumor involvement (14 18) in each image set are described in Appendix E1 (online). Because the spatial resolution of the DW images was inferior to that of the gadoxetic acid images, assessment of lesion localization was performed primarily with the gadoxetic acid set, and the DW imaging set was used for lesion characterization. When there was a discrepancy in the interpretations between the gadoxetic acid set and the DW imaging set, the observers were asked to give priority to the findings that drew higher observer confidence levels. Reference Standard The final diagnosis of all tumors was based on histopathologic examination of 770 radiology.rsna.org n Radiology: Volume 270: Number 3 March 2014

4 Table 2 Correlation of Bismuth- Corlette Classification of Hilar Cholangiocarcinoma at MR Imaging Study Interpretation and Pathologic Findings Observer and Bismuth-Corlette Classification Pathologic Tumor Classification II IIIa IIIb IV Observer 1 II 5/4 3/0 2/1 0 IIIa 2/3 21/25 0 3/2 IIIb 0 0 9/10 1/0 IV 0 2/1 0 4/6 Observer 2 II 5/5 3/0 2/1 0 IIIa 2/2 20/24 0 3/1 IIIb 0 0 9/10 1/0 IV 0 3/2 0 4/7 Total Table 3 Diagnostic Performance with Gadoxetic Acid MR Image Set and That with Combined DW and Gadoxetic Acid MR Image Set for Predicting Biliary Tumor Involvement on Per Bile Duct Confluence Basis Observer and Image Set A z Value* Sensitivity Specificity Observer 1 Gadoxetic acid set (0.922, 0.988) 91.4 [96/105] (85.9, 96.9) 92.2 [47/51] (83.7, 97.5) Combined set (0.953, 0.996) 97.1 [102/105] (93.2, 99.3) 92.2 [47/51] (84.1, 97.5) P value Observer 2 Gadoxetic acid set (0.905, 0.996) 91.4 [96/105] (85.9, 96.9) 90.2 [46/51] (77.8, 96.3) Combined set (0.935, 0.993) 98.1 [103/105] (94.9, 99.7) 92.2 [47/51] (83.7, 97.5) P value Note. P values were estimated by using nonparametric analysis of clustered receiver operating characteristic curve data or the generalized estimating equations method and were corrected with the Bonferroni method because of multiple testing. * A z = area under the receiver operating characteristic curve. Data are means 6 standard deviations, with 95% CIs in parentheses. Data in brackets are numbers of bile duct confluences; data in parentheses are 95% CIs Note. Data are numbers of tumors for the gadoxetic acid image set/for the combined image set. the surgical specimen and resection margin assessment. Six of 52 patients were noted as having microscopically positive resection margins in the final pathology reports (R1 resection), even though results of intraoperative frozen-section biopsies had indicated negative resection margins. In two other patients, palliative segmental bile duct resection with a positive resection margin (R2 resection) was performed. In these patients, intraoperative evaluation of the extent of bile duct tumor and vascular and liver involvement were also performed. Statistical Analysis For diagnostic performance in terms of bile duct tumor involvement, receiver operating characteristic curve analysis was performed and a pairwise comparison was made between the two image sets by using nonparametric analysis of clustered data (19). Only grades of 4 or 5 were considered positive for tumor involvement. For bile duct tumor involvement and vascular invasion, sensitivity, specificity, and accuracy were calculated and compared between the two image sets by using the clustered data method (20) and the generalized estimating equations method, respectively. For analyses of clustered data, 95% confidence intervals (CIs) and P values were corrected by using the Bonferroni method. For liver invasion, sensitivity, specificity, and accuracy were calculated, and a comparison was made between the image sets by using the McNemar test. k Scores were calculated to determine the degree of interobserver agreement. Statistical analyses were performed by using the statistics softwares MedCalc, version 11.4 (MedCalc Software, Mariakerke, Belgium) and SPSS, version 18.0 (SPSS, an IBM company, Chicago, Ill). Null hypotheses of no difference were rejected if P values were less than.05. Results According to the Bismuth-Corlette classification of hilar cholangiocarcinoma (Table E1 [online]), 26 of the 52 tumors were type IIIa, 11 were type IIIb, eight were type IV, and seven were type II (Table 2). Morphologically, the majority of the tumors (n = 44) were the periductal infiltrative type, and the remaining tumors were the periductal infiltrative type with mass formation in the liver (n = 2) or the intraductal growing type (n = 6). Bile Duct Involvement Among 156 bile duct confluences in 52 patients, 105 were found to have tumor involvement at histopathologic examination. All 52 tumors were hyperintense compared with liver parenchyma at both b = 100 sec/mm 2 and b = 800 sec/mm 2 DW imaging. As summarized in Table 3, for identifying tumor involvement of each bile duct confluence, both observers showed higher A z values in interpreting the combined set than in interpreting the gadoxetic acid set, although the difference was not statistically significant (0.980 vs for observer 1; vs for observer 2; P..05 for both observers) (Fig 1). Diagnostic accuracies were slightly higher for the combined set than for the gadoxetic acid set (95.5% [149 of 156] vs 91.7% [143 of 156] for observer 1; 96.2% [150 of 156] vs 91.0% [142 of 156] for observer 2), although there was no significant difference (P..05). In addition, sensitivity with the combined set was significantly higher than that with the gadoxetic acid set for both observers (97.1% [102 of 105] vs 91.4% [96 of 105], P =.029 for observer 1; 98.1% [103 of 105] vs 91.4% [96 of 105], P =.016 for observer 2), although there were no significant differences in specificity between the two image sets (P..05). Radiology: Volume 270: Number 3 March 2014 n radiology.rsna.org 771

5 Figure 1 Figure 1: Images in 54-year-old man with Bismuth-Corlette type IIIa hilar cholangiocarcinoma. (a) Three-dimensional MR cholangiopancreatography image shows dilatation of the intrahepatic bile duct bilaterally and separation of the right anterior and posterior and left hepatic ducts (arrows). (b) Unenhanced and (c) gadoxetic acid enhanced three-dimensional gradient-echo 3-minute delayed phase images show thickening of the bile duct at the hepatic hilum (arrow) with gradual hyperenhancement. (d) On single-shot echo-planar DW image obtained at b = 800 sec/ mm 2, the thickened hilar duct shows hyperintensity. Both observers correctly interpreted this tumor as type IIIa both before and after DW images were added to the gadoxetic acid enhanced MR image set. In the per-patient analysis of data in the 52 patients, use of the combined image set resulted in accurate diagnosis of the Bismuth-Corlette classification of 45 tumors (86.5%) and 46 tumors (88.5%) for observer 1 and observer 2, respectively, whereas use of the gadoxetic acid set resulted in correct diagnosis of 39 tumors (75.0%) and 38 tumors (73.1%) (Table 2). The addition of DW images led to correction of the Bismuth-Corlette classification, which was misclassified at the gadoxetic acid set interpretation, for eight and nine tumors for observers 1 and 2, respectively. Among tumors that were misclassified, five were misclassified with the gadoxetic acid set but were correctly classified with the combined image set by both observers (classification changed from type II to IIIa, n = 2; from type IIIa to IV, n = 1; from type IIIb to IV, n = 1; and from type IV to IIIa, n = 1). In two patients with type IIIa tumors, bile duct wall thickening or enhancement at the right secondary biliary confluence was not clearly seen on the gadoxetic acid set. However, diffusion restriction was clearly identified on DW images (Fig 2). Two type IV tumors were mutually underestimated by both observers as type IIIa and type IIIb tumors on the gadoxetic acid set. On review, in one of the patients, a small intraductal tumor Figure 2 Figure 2: Images in 70-year-old woman with Bismuth-Corlette type IIIa hilar cholangiocarcinoma. (a) Three-dimensional MR cholangiopancreatography image shows separation of the right and left hepatic ducts (arrow). (b) Gadoxetic acid enhanced portal venous phase image shows thickening of the bile duct at the hepatic hilum (arrow). Wall thickening or separation of the bilateral secondary confluence was not definite, and both observers interpreted this tumor as type II on the basis of conventional MR images. However, on (c) a DW image obtained at b = 800 sec/mm 2, the tumor is clearly seen in the hilar duct (arrow) and in the right secondary confluence (arrowheads) as areas of hyperintensity. The tumor was also seen as areas of hypointensity on the ADC map (not shown). Both observers interpreted this tumor as type IIIa after the addition of DW images to the conventional MR image set. 772 radiology.rsna.org n Radiology: Volume 270: Number 3 March 2014

6 Figure 3 Figure 3: Images in 76-year-old man with Bismuth-Corlette type IIIa hilar cholangiocarcinoma who had undergone a left percutaneous transhepatic biliary decompression procedure before MR imaging. (a, b), Gadoxetic acid enhanced portal venous phase images show thickening of the bile duct with separation of the right secondary confluence (arrow in a) and irregular narrowing of the left main duct extending to B4 (arrow in b). (c) On a T2-weighted MR image, the left secondary confluence also shows thickening and moderate hyperintensity, suggesting the possibility of tumor involvement (arrow). Both observers classified this tumor as a type IV tumor. However, (d) a DW image obtained at b = 800 sec/mm 2, as well as an ADC map (not shown), clearly displayed diffusion restriction confined to the hilar duct and the right secondary biliary confluence (arrow), so the tumor was classified as a type IIIa tumor. At pathologic examination, the thickened left secondary biliary confluence proved to be a manifestation of cholangitis. growth on the segment IV duct was clearly visible with DW images but was overlooked on the gadoxetic acid set. In the other patient, only mild and even concentric wall thickening without luminal narrowing was seen on the right secondary biliary confluence on the gadoxetic acid set. However, diffusion restriction was definite, and so the lesion was correctly classified as a type IV tumor. On the other hand, in one patient with a type IIIa tumor, the wall of the left secondary biliary confluence appeared to be thickened with signal intensity change on the gadoxetic acid set, and hence the tumor was classified as a type IV tumor. However, DW imaging clearly displayed diffusion restriction confined to the primary and right secondary biliary confluence. At pathologic examination, the wall thickening of the left secondary biliary confluence proved to be a manifestation of cholangitis (Fig 3). Both observers overestimated two type II tumors as type IIIa tumors with both image sets because wall thickening with prominent enhancement was visible at the right secondary biliary confluence on the gadoxetic acid set and diffusion restriction was present on the DW images. However, pathologic examination revealed the thickened wall of the right secondary biliary confluence to have been caused by inflammation without tumor infiltration. On the other hand, one type IV tumor and one type IIIb tumor were underestimated by both observers as type IIIa and type II tumors, respectively, with both image sets. At review, in each patient, mild and even concentric wall thickening and enhancement of the bile duct without luminal obstruction was seen at the left secondary biliary confluence on the gadoxetic acid set, but diffusion restriction was not definite. However, pathologic examination revealed submucosal spread of tumor in the bile duct. Among the two patients who had undergone percutaneous transhepatic biliary decompression procedures before MR imaging, the gadoxetic acid set in one patient (Fig 3) and both image sets in the other patient led to overestimation of the bile duct tumor extent. In addition, in nine patients, diffuse signal intensity changes in the unilateral hepatic lobe were demonstrated on T2-weighted images and gadoxetic acid enhanced images, particularly on HBP and DW images. These signal intensity changes were probably due to cholangiohepatitis and bile stasis resulting from complete ipsilateral ductal obstruction. In three of these nine patients, DW images more clearly displayed hyperintensity in the unilateral hepatic lobe than other images (Fig 4). Vascular and Liver Parenchymal Invasion Surgical and pathologic results revealed arterial invasion in 13 (25.0%) of the 52 patients (right hepatic artery, n = 7; left hepatic artery, n = 6), portal venous invasion in 10 (19.2%) patients (right portal vein, n = 3; right and main portal veins, n = 1; left portal vein, n = 6), and liver parenchymal invasion in 20 (38.5%) patients (segment IV, n = 9; segment V, n = 5; segment VIII, n = 2; segment I, n = 4). As summarized in Table 4, for assessment of vascular invasion, the combined set and the gadoxetic acid set showed similar diagnostic performances Radiology: Volume 270: Number 3 March 2014 n radiology.rsna.org 773

7 Figure 4 for both reviewers (P..05), although there were trends toward slightly improved sensitivity for the combined set compared with the gadoxetic acid set. In the evaluation of liver parenchymal invasion, the combined set (75.0% for both observers) showed significantly higher sensitivity than the gadoxetic acid set (observer 1, 50.0%; observer 2, 45.0%) (observer 1, P =.016; observer 2, P =.031) (Fig 5). Interobserver Agreement For evaluation of bile duct involvement, the two observers showed excellent interobserver agreement for both the gadoxetic acid set (weighted k = 0.897) and the combined set (weighted k = 0.896). For evaluation of vascular invasion, interobserver agreement was excellent for both the gadoxetic acid set (weighted k = for hepatic artery; weighted k = for portal vein) and the combined set (weighted k = for hepatic artery; weighted k = for portal vein). For assessment of liver Figure 4: Images in 49-year-old man with Bismuth-Corlette type II hilar cholangiocarcinoma. (a) Three-dimensional MR cholangiopancreatography image shows obstruction of the hilar duct (arrow) and dilatation of the right intrahepatic duct. (b) DW image obtained at b = 800 sec/mm 2 clearly displays hyperintensity of the right portion of the liver, indicating bile stasis associated with complete ipsilateral ductal obstruction, which was not clearly demonstrated on gadoxetic acid enhanced portal venous phase T2-weighted images (not shown) or on (c) a 20-minute HBP image. invasion, interobserver agreement was good for both the gadoxetic acid set (weighted k = 0.750) and the combined set (weighted k = 0.785). Discussion In our study, the combined sets yielded diagnostic accuracies of 95.5% (149 of 156) and 96.2% (150 of 156) for the observers for the determination of tumor extent accuracies that are slightly superior to the range of 71% 94% in previous studies (21 24). This might also indicate the benefit of adding DW imaging. Because of the excellent softtissue contrast and three-dimensional data acquisition with respiratory motion correction, the combined use of contrast material enhanced three-dimensional MR imaging and MR cholangiopancreatography has been reported to improve the diagnostic accuracy of preoperative staging of hilar cholangiocarcinoma (17,18,22,23). In addition, in our study, the high sensitivity of the combined image set for the evaluation of biliary tumor extent might be attributable to the information on the cellularity and architectural distortion of the lesion obtained from DW imaging in addition to the information obtained from conventional MR imaging. Therefore, for detecting 105 biliary confluences with tumor involvement, the sensitivity with the combined set was higher than that with the gadoxetic acid set for both observers (P,.05). Addition of DW imaging to the gadoxetic acid set led to correction of the Bismuth-Corlette classification in eight and nine patients for the two observers; these corrections were mostly of tumors that were underestimated with the gadoxetic acid set. At high b value DW imaging, most tumors (except three) were clearly demonstrated as hyperintense lesions while the background signal intensity, including that of the liver parenchyma, was suppressed. These lesions might not be clearly delineated at conventional MR imaging, particularly when fatty tissue in the hepatic hilum is not abundant or when tumor involvement approximates to the secondary biliary confluence. In addition, it might be difficult to differentiate between intraductal tumor and stones or sludge with conventional MR imaging. In that case, additional reading of DW images could be beneficial because stones and sludge showed signal suppression on high b value images. Meanwhile, we found a tendency toward overestimation of tumor extent, as well as higher rates of tumor detection, after the addition of DW images, which could explain the relatively low specificity in the evaluation of tumor extent. Given that cholangiocarcinoma develops in a stepwise carcinogenetic pathway from inflammation and biliary intraepithelial neoplasia (25), it should be acknowledged that it is difficult to distinguish chronic inflammation or precancerous lesions from superficial early cancers at both imaging and pathologic examination. In consideration of advances in surgical technique and the introduction of frozen-section biopsy during surgery, an imaging modality that provides high sensitivity for tumor detection may be more acceptable than 774 radiology.rsna.org n Radiology: Volume 270: Number 3 March 2014

8 Table 4 Sensitivity and Specificity of Gadoxetic Acid MR Image Set and Combined DW and Gadoxetic Acid MR Image Set for Predicting Vascular and Liver Parenchymal Invasion Observer and Image Set Hepatic Artery Portal Vein Liver Parenchyma Accuracy (%) Sensitivity (%) Specificity (%) Accuracy (%) Sensitivity (%) Specificity (%) Accuracy (%) Sensitivity (%) Specificity (%) Observer 1 Gadoxetic acid set 96.6 (146/156) 61.5 (8/13) [29.9, 93.1] 96.5 (138/143) [93.3, 99.7] 97.4 (152/156) 72.7 (8/11) [34.0, 100] 99.3 (144/145) [99.3, 99.3] 73.1 (38/52) 50.0 (10/20) [27.9, 72.1] 87.5 (28/32) [70.1, 95.9] Combined set 95.5 (149/156) 84.6 (11/13) [61.3, 100] 96.5 (138/143) [93.3, 99.7] 98.1 (154/156) 90.9 (11/11) [72.4, 100] 98.6 (143/145) [96.4, 100] 86.5 (45/52) 75.0 (15/20) [50.6, 90.4] 93.8 (30/32) [77.8, 98.9] P value Observer 2 Gadoxetic acid set 95.5 (149/156) 76.9 (10/13) [49.6, 100] 97.2 (139/143) [94.0, 100] 96.8 (154/156) 63.6 (7/11) [24.2, 100] 99.3 (144/145) [99.3, 99.3] 71.2 (37/52) 45.0 (9/20) [23.8, 68.0] 87.5 (28/32) [70.1, 95.9] Combined set 94.2 (147/156) 84.6 (11/13) [61.3, 100] 95.1 (136/143) [91.2, 98.9] 98.7 (155/156) 90.9 (11/11) [72.4, 100] 99.3 (144/145) [99.3, 99.3] 84.6 (44/52) 75.0 (15/20) [50.6, 90.4] 90.6 (29/32) [73.8, 97.5] P value Note. Data in parentheses are numbers of vascular segments for hepatic artery and portal vein and numbers of patients for liver parenchyma; data in brackets are 95% CIs. P values were estimated by using the generalized estimating equations method and were corrected with the Bonferroni method because of multiple testing for the hepatic artery and portal vein. The P value for specificity in the evaluation of portal vein invasion could not be assessed owing to the structure of the data. Figure 5 Figure 5: Images in 68-year-old man with Bismuth-Corlette type IV hilar cholangiocarcinoma. (a) Gadoxetic acid enhanced 3-minute delayed phase and (b) 20-minute HBP images show thickening of the hilar duct (arrow) and dilatation of the intrahepatic ducts bilaterally. (c) DW image obtained at b = 800 sec/mm 2 clearly shows a small hyperintense area (arrowhead) in segment IV adjacent to the thickened hilar duct (arrow), indicating liver invasion. This lesion (arrowhead) is also visible in b, but was missed during image interpretation. During surgery, invasion of segment IV of the liver was confirmed. one that provides high specificity in clinical practice. The accuracy of the combined image set in the determination of liver invasion was 86.5% and 84.6% for the two observers, which was higher than the 73.1% and 71.2% accuracies they achieved with the gadoxetic acid set, as well as the 80% accuracy achieved with gadolinium-enhanced 1.5-T MR imaging with MR cholangiopancreatography in a previous study (23). Particularly in patients with obstructive cholangitis, the addition of DW imaging may be helpful in detecting liver invasion, because the hepatocyte uptake of gadoxetic acid is often insufficient to delineate liver parenchymal lesions. Conversely, in our study, tumors in two patients were mutually underestimated in the two image sets by both observers, yet pathologic examination revealed submucosal tumor spread along each bile duct. The identification of submucosal tumor spread is difficult with any imaging modality, because submucosal spread cannot cause reduction of the bile duct lumen (26). In addition, the low spatial resolution of DW imaging could help explain its limited contribution in the determination of vascular invasion in the hepatic hilum, where the anatomy of the bile duct and hepatic vessels is complex. There were limitations to our study. First, because our study population Radiology: Volume 270: Number 3 March 2014 n radiology.rsna.org 775

9 included patients who underwent surgery, tumors that were considered unresectable, which would be more easily diagnosed correctly, were not included. Therefore, the overall diagnostic performance might be underestimated compared with that in daily clinical practice. On the other hand, the lack of a control population with benign biliary obstruction or inflammation might have led to overestimation of the diagnostic sensitivity. Second, because DW imaging has been recently introduced into the routine MR imaging protocol, we had no established criteria for distinguishing between bile duct cancer and cholangitis; as a result, the decisions by the observers were made subjectively. Nevertheless, the observers showed excellent agreement regarding biliary tumor involvement. Third, it is uncertain whether our results could be applicable to conventional gadolinium-enhanced MR imaging owing to different characteristics of the gadoxetic acid, especially in assessment of liver invasion. In conclusion, in the preoperative evaluation of hilar cholangiocarcinoma, the addition of DW imaging to gadoxetic acid enhanced MR imaging with MR cholangiopancreatography could improve sensitivity in the detection of tumor extent along the bile duct and the detection of liver invasion. Disclosures of Conflicts of Interest: M.J.P. No relevant conflicts of interest to disclose. Y.K.K. No relevant conflicts of interest to disclose. S.L. No relevant conflicts of interest to disclose. H.R. No relevant conflicts of interest to disclose. W.J.L. No relevant conflicts of interest to disclose. References 1. Ito F, Cho CS, Rikkers LF, Weber SM. Hilar cholangiocarcinoma: current management. Ann Surg 2009;250(2): Beazley RM, Hadjis N, Benjamin IS, Blumgart LH. Clinicopathological aspects of high bile duct cancer: experience with resection and bypass surgical treatments. Ann Surg 1984;199(6): Blumgart LH, Hadjis NS, Benjamin IS, Beazley R. Surgical approaches to cholangiocarcinoma at confluence of hepatic ducts. Lancet 1984;1(8368): Tompkins RK, Thomas D, Wile A, Longmire WP Jr. Prognostic factors in bile duct carcinoma: analysis of 96 cases. Ann Surg 1981;194(4): Hemming AW, Reed AI, Fujita S, Foley DP, Howard RJ. Surgical management of hilar cholangiocarcinoma. Ann Surg 2005; 241(5): ; discussion Dinant S, Gerhards MF, Rauws EA, Busch OR, Gouma DJ, van Gulik TM. Improved outcome of resection of hilar cholangiocarcinoma (Klatskin tumor). Ann Surg Oncol 2006;13(6): Seyama Y, Kubota K, Sano K, et al. Longterm outcome of extended hemihepatectomy for hilar bile duct cancer with no mortality and high survival rate. Ann Surg 2003;238(1): Nimura Y, Kamiya J, Kondo S, et al. Aggressive preoperative management and extended surgery for hilar cholangiocarcinoma: Nagoya experience. J Hepatobiliary Pancreat Surg 2000;7(2): Saxena A, Chua TC, Chu FC, Morris DL. Improved outcomes after aggressive surgical resection of hilar cholangiocarcinoma: a critical analysis of recurrence and survival. Am J Surg 2011;202(3): Klempnauer J, Ridder GJ, von Wasielewski R, Werner M, Weimann A, Pichlmayr R. Resectional surgery of hilar cholangiocarcinoma: a multivariate analysis of prognostic factors. J Clin Oncol 1997;15(3): Igami T, Nagino M, Oda K, et al. Clinicopathologic study of cholangiocarcinoma with superficial spread. Ann Surg 2009;249(2): Wakai T, Shirai Y, Moroda T, Yokoyama N, Hatakeyama K. Impact of ductal resection margin status on long-term survival in patients undergoing resection for extrahepatic cholangiocarcinoma. Cancer 2005; 103(6): Cui XY, Chen HW. Role of diffusion-weighted magnetic resonance imaging in the diagnosis of extrahepatic cholangiocarcinoma. World J Gastroenterol 2010;16(25): Kim MJ, Mitchell DG, Ito K, Outwater EK. Biliary dilatation: differentiation of benign from malignant causes value of adding conventional MR imaging to MR cholangiopancreatography. Radiology 2000;214(1): Taouli B, Koh DM. Diffusion-weighted MR imaging of the liver. Radiology 2010;254(1): Lee HY, Kim SH, Lee JM, et al. Preoperative assessment of resectability of hepatic hilar cholangiocarcinoma: combined CT and cholangiography with revised criteria. Radiology 2006;239(1): Park HS, Lee JM, Choi JY, et al. Preoperative evaluation of bile duct cancer: MRI combined with MR cholangiopancreatography versus MDCT with direct cholangiography. AJR Am J Roentgenol 2008;190(2): Ryoo I, Lee JM, Chung YE, et al. Gadobutrol-enhanced, three-dimensional, dynamic MR imaging with MR cholangiography for the preoperative evaluation of bile duct cancer. Invest Radiol 2010;45(4): Obuchowski NA. Nonparametric analysis of clustered ROC curve data. Biometrics 1997;53(2): Rao JN, Scott AJ. A simple method for the analysis of clustered binary data. Biometrics 1992;48(2): Manfredi R, Masselli G, Maresca G, Brizi MG, Vecchioli A, Marano P. MR imaging and MRCP of hilar cholangiocarcinoma. Abdom Imaging 2003;28(3): Cho ES, Park MS, Yu JS, Kim MJ, Kim KW. Biliary ductal involvement of hilar cholangiocarcinoma: multidetector computed tomography versus magnetic resonance cholangiography. J Comput Assist Tomogr 2007; 31(1): Masselli G, Manfredi R, Vecchioli A, Gualdi G. MR imaging and MR cholangiopancreatography in the preoperative evaluation of hilar cholangiocarcinoma: correlation with surgical and pathologic findings. Eur Radiol 2008;18(10): Kiryu S, Dodanuki K, Takao H, et al. Freebreathing diffusion-weighted imaging for the assessment of inflammatory activity in Crohn s disease. J Magn Reson Imaging 2009; 29(4): Zen Y, Sasaki M, Fujii T, et al. Different expression patterns of mucin core proteins and cytokeratins during intrahepatic cholangiocarcinogenesis from biliary intraepithelial neoplasia and intraductal papillary neoplasm of the bile duct: an immunohistochemical study of 110 cases of hepatolithiasis. J Hepatol 2006;44(2): Bhuiya MR, Nimura Y, Kamiya J, et al. Clinicopathologic studies on perineural invasion of bile duct carcinoma. Ann Surg 1992;215(4): radiology.rsna.org n Radiology: Volume 270: Number 3 March 2014