Presurgical Evaluation of Pancreatic Cancer: A Comprehensive Imaging Comparison of CT Versus MRI

Transcription

1 Gastrointestinal Imaging Original Research Chen et al. CT Versus MRI for Presurgical Evaluation of Pancreatic Cancer Gastrointestinal Imaging Original Research Fang-Ming Chen 1 Jian-Ming Ni Zhui-Yang Zhang Lei Zhang Bin Li Chun-Juan Jiang Chen FM, Ni JM, Zhang ZY, Zhang L, Li B, Jiang CJ Keywords: CT, MRI, pancreatic cancer, presurgical evaluation DOI: /AJR Received July 1, 2015; accepted after revision October 26, All authors: Department of Radiology, Wuxi Second People s Hospital Affiliated to Nanjing Medical University, 68 Zhongshan Rd, Jiangsu , PR China. Address correspondence to Z. Y. Zhang (zhangzhuiyang@163.com). AJR 2016; 206: X/16/ American Roentgen Ray Society Presurgical Evaluation of Pancreatic Cancer: A Comprehensive Imaging Comparison of CT Versus MRI OBJECTIVE. The purpose of this study was to compare comprehensive CT and MRI in the presurgical evaluation of pancreatic cancer. MATERIALS AND METHODS. Thirty-eight patients with pathologically proven pancreatic cancer were included in a retrospective study. CT with negative-contrast CT cholangiopancreatography and CT angiography (CTA) (CT image set) versus MRI with MRCP and MR angiography (MRI image set) were analyzed independently by two reviewers for tumor detection, extension, metastasis, vascular invasion, and resectability. These results were compared with the surgical and pathologic findings. RESULTS. The rate of detection of tumors was higher with MRI than with CT but not significantly so (reviewer 1, p = 1.000; reviewer 2, p = 0.500). In the evaluation of vessel involvement, nodal status, and resectability, although CT had higher ROC AUC values than did MRI (reviewer 1, vs 0.858, vs 0.503, and vs 0.774; reviewer 2, vs 0.849, vs 0.583, and vs 0.815), the differences were not statistically significant (p = vs 0.494, vs 0.244, and vs for reviewers 1 and 2). In the evaluation of tumor extension and organ metastases in the 38 patients, correct diagnosis of one of two liver metastases was achieved with both image sets, one case of omental and one case of peritoneal seeding were underestimated, and one case of stomach invasion was overestimated. CONCLUSION. MRI and CT had similar performance in the presurgical evaluation of pancreatic cancer. P ancreatic cancer (PC) is among the leading causes of cancer-related deaths around the world. According to a 2012 review [1], the relative 5-year overall survival rate is 5 18%. Complete surgical resection with chemotherapy offers the best outcome in this particular cancer [2]. However, early presurgical diagnosis of PC remains difficult with noninvasive imaging modalities [3, 4]. The major difficulties involve detection of small PC tumors [1] and accurate TNM staging, particularly in assessing vascular structure and microspread (i.e., nodal involvement and peritoneal seeding) [5]. Some studies have shown that endoscopic ultrasound has higher sensitivity and specificity than CT and MRI in the diagnosis of PC, especially for smaller tumors ( 2 cm) [1]. However, endoscopic ultrasound is an invasive technique, and successful procedures are operator dependent [5]. With the advent of MDCT, which affords submillimeter scanning and reconstruction coupled with a variety of postprocessing techniques, such as high-quality multiplanar reformation, CT angiography (CTA), and negativecontrast CT cholangiopancreatography, images can be acquired simultaneously [6 11]. The additional diagnostic value of negative-contrast CT cholangiopancreatography with minimum intensity projection allows better definition of sites of biliary obstruction [6] and plays a key complementary role in identifying each of the periampullary tumors of four origins [10]. Thus, it is necessary to reestimate their role in patients with suspected PC. Although MRI has comparable performance to CT because multiparametric images are generated [12 15], to our knowledge, there have been no reports regarding CT with negative-contrast CT cholangiopancreatography and CTA versus MRI with MRCP and MR angiography (MRA) in the preoperative evaluation of PC. The purpose of this study was to conduct a comprehensive presurgical imaging comparison of CT and MRI in patients with PC. 526 AJR:206, March 2016

2 CT Versus MRI for Presurgical Evaluation of Pancreatic Cancer Materials and Methods Patient Selection Our institutional review board approved this retrospective study, and informed consent was not required. From January 2008 to November 2014, patients with clinically suspected PC were identified from the database of our hospital information system. The inclusion criteria were as follows: PC treated surgically (including curative and palliative operations) and had detailed operative records; no chemoradiotherapy before either CT (unenhanced and dual-phase contrast-enhanced CT with negative-contrast CT cholangiopancreatography) or MRI (MRCP and dynamic contrastenhanced examinations); interval between CT or MRI examination and surgery less than 1 month; and final pathologic diagnosis based on biopsy of surgical specimen obtained at laparotomy. Of the 137 patients initially included, 99 patients were excluded from the study for the following reasons: CT without MRI (n = 59) or MRI without CT (n = 9); no surgical exploration instead of endoscopic nasobiliary drainage before CT or MRI (n = 8); histologic proof of PC obtained only by means of endoscopic biopsy (n = 19); and interval greater than 1 month between CT or MRI examination and surgery (n = 4). The other 38 patients (20 men, 18 women; age range, years; mean, 64.5 years) composed the study population. The mean interval between CT and MRI examinations was 3.2 days (range, 1 10 days). The mean interval between imaging and surgery was 8.8 days (range, 1 24 days) for CT and 6.2 days (1 21 days) for MRI. At many TABLE 1: Sequences and Parameters for 1.5- and 3-T MRI Sequence Unenhanced Axial T1-weighted in and out of phase TR/TE (ms) institutions [16 18], as it is at ours, CT is the preferred method for initial imaging of patients with suspected PC because it is fast and simple to perform and is well tolerated [13]. Although MRI has been found to be equally sensitive and specific in the staging of PC [12, 13, 17], it is not as widely used as CT as the primary imaging modality because of cost and lack of availability [17, 18]. CT All CT scans were obtained with an MDCT scanner (Aquilion 64, Toshiba Medical Systems). The standard scanning protocol at our hospital consists of unenhanced and biphasic contrastenhanced scans. An automatic trigger scanning mode with a 5-second delay was used when an ROI was set on the descending aorta, for which the preset CT number was 200 HU. For this mode, the arterial and portal venous phases were set at delays of and seconds after the start of the injection of ml of nonionic contrast agent (ioversol, Optiray 320, Tyco Healthcare) at a rate of 3 4 ml/s through an antecubital vein. The parameters were as follows: 120 kvp; 0.5-second scanning time per rotation; detector collimation, 1.0 mm 32 or 0.5 mm 64; pitch factor, or 0.828; tube current with automatic dose modulation ranging, ma; FOV, cm 2. For the arterial and portal venous phase raw data, volume data were automatically reconstructed with both a slice thickness and a reconstruction interval of 1.0 mm. The portal venous phase raw data were reconstructed for negative-contrast CT Flip Angle ( ) cholangiopancreatography with a low-spatial-resolution algorithm for reducing image noise. The other parameters were as follows: slice thickness, mm; reconstruction interval, mm; matrix, ; FOV, cm 2. All reconstructed 2D source images (volume data) were transferred to the workstation (Advantage Workstation version 3.1, GE Healthcare). MRI MRI was performed with a 1.5-T (Signa Excite, GE Healthcare) (n = 24) or 3-T (Magnetom Skyra, Siemens Healthcare) (n = 14) MRI system with an eight-channel phased-array body coil. The examination protocol included the following sequences at both 1.5 and 3 T: axial unenhanced T1-weighted imaging with in- and out-of-phase fast spoiled gradient-recalled echo or volume-interpolated breathhold examination; axial T2-weighted imaging with fast recovery fast spin-echo (FSE) or periodically rotated overlapping parallel lines with enhanced reconstruction technique (Blade, a Siemens Healthcare implementation of this technique); coronal image with fast imaging employing steady-state acquisition or HASTE; coronal oblique thick-slab MRCP with single-shot FSE or HASTE with a breath-hold and acquisition of 6 12 images at a rotation for each; and coronal oblique 3D MRCP with fast recovery FSE or spatial and chemical-shift encoded excitation combined with a respiratory-gated sequence. For dynamic contrast-enhanced MRI, a single breath-hold was used for each of the following dynamic sequences. One sequence included axial un- Section Intersection Thickness (mm) Gap (mm) Coverage (mm) FOV (cm 2 ) Matrix 1.5 T 3 T 1.5 T 3 T 1.5 T 3 T 1.5 T 3 T 1.5 T 3 T 1.5 T 3 T 1.5 T 3 T 185/4.88, 185/2.33 Axial T2-weighted / /2.46, 4.1/ ,349/ (179) (180) Coronal 3.4/ / (138) MRCP (195) (204) (170) Thick slab 6000/ / D / / (77) (109) Contrast-enhanced MRI Axial T1-weighted 3.7/ / (185) Coronal T1-weighted 4.2/ / (168) Note Values in parentheses are means (195) (183) AJR:206, March

3 Chen et al. enhanced, arterial, and axial or coronal portal venous phases; coronal equilibrium phase; and axial delayed phase imaging with liver acceleration volume acquisition technique. The other sequence included axial unenhanced, early and late arterial, and portal venous phases; coronal equilibrium phase; and axial delayed phase imaging with volume-interpolated breath-hold examination technique. The arterial, portal venous, equilibrium, and delayed phase images were obtained seconds (36 40 seconds for the late arterial phase on the 3-T system), 60 seconds, seconds, and seconds after the start of the injection of ml of gadodiamide (Omniscan, GE Healthcare) through an antecubital vein at a rate of 2 ml/s. Fat suppression was regularly used for T2-weighted imaging, thick-slab MRCP, volume acquisition for 3D MRCP, and contrast-enhanced MRI sequences. The acquisition parameters and sequences for MRI are listed in Table 1. The unenhanced and contrast-enhanced T1- and T2-weighted images, thick-slab MRCP, and the source images for 3D MRCP were transferred to the workstations (Advantage Workstation version 4.3, GE Healthcare; Syngo Multimodality Workplace version D13, Siemens Healthcare). Image Postprocessing A radiologist with 8 years of experience in abdominal CT and MRI undertook the CT and MR image postprocessing at each of the workstations. The radiologist first created 2D CT images, including multiplanar reformations and curved planar reconstructions with the portal venous phase volume data and then generated negative-contrast CT cholangiopancreatograms with 3D tools as described in previous reports [10, 19]. For 3D MRCP, a 3D maximum-intensity-projection (MIP) tool was used to create 3D MRCP and MIP images saved as negative-contrast CT cholangiopancreatograms. For CTA and MRA, both thick-slab (thickness, mm) and 3D CT and MR arteriograms and portovenograms were created with MIP and volume-rendering techniques and the arterial and portal venous phase data. All of the CT and MR images were transferred to a PACS (MultiView with Synapse software, Fujifilm Medical Systems). Image Analysis Two radiologists (one with 14 years of experience in abdominal CT, the other with 8 years of experience in MRI) independently evaluated the two image sets (both postprocessing and source images) on the PACS screen. The CT set consisted of 2D CT images including negative-contrast CT cholangiopancreatograms and CTA images. The MRI set consisted of 2D MR images including MRCP and MRA images. To minimize the readers learning bias, a 2-week interval was observed between the two readings, and the images were provided randomly from the two image sets. The reviewers were asked to evaluate detection of the tumor as visible or invisible; to assess the presence or absence of peripancreatic major organs, vessel involvement, or metastases; and to determine resectability according to the evaluation criteria (described later) on CT or MR images. The assessment results were compared with the surgical and the pathologic records. PC was defined at contrast-enhanced CT and MRI as a low- or high-attenuation or hypointense or hyperintense mass compared with the surrounding pancreatic parenchyma [3, 14, 15]. If no mass was discerned, indirect signs, including focal contour changes in the pancreas with abrupt termination of the bile duct, pancreatic duct, or both (double-duct sign) or the presence of the four-segment sign were suggestive of PC with both modalities [20, 21]. Once a mass was found, the reviewers further measured the maximal diameter of the low- or high-attenuation or hypointense or hyperintense region. Peripancreatic major organ (i.e., stomach or colon) invasion was considered present if a suspected low-attenuation or hypointense lesion directly invaded or reached the surface of the structures [8]. Regional lymph node evaluation followed the Japanese Pancreas Society classification [22]. Nodal metastasis was considered present when the shortaxis diameter was longer than 10 mm or when central necrosis was present at any size or its attenuation or signal intensity was greater than that of liver parenchyma in the portal venous phase [13, 23]. For the diagnostic criteria for vascular invasion in this study, we used the three-criterion image grading system suggested by Valls et al. [24] and modified in more recent literature [18, 25]. We graded the tumor-to-vessel contact for the five major peripancreatic vessels (celiac artery, hepatic artery, superior mesenteric artery [SMA] and vein, and portal vein) as follows: imaging grade 0, no contiguity between tumor and vessel; 1, tumor-to-vessel contiguity 50%; and 2, tumor-to-vessel contiguity > 50%, including complete vessel occlusion. This evaluation required approximately 5 10 minutes for each image set per patient. Criteria for Unresectability at CT and MRI According to the defined criteria, a tumor was considered unresectable if one or more of the following findings was present on CT or MR images. First, the tumor-to-vessel contiguity reached imaging grade 2, including the superior mesenteric to portal vein confluence [18, 26]. Second, peripancreatic organ involvement, such as of the stomach or colon [1, 13], other than choledochal or duodenal invasion was present, because the invaded choledochal or duodenal organs can be resected en bloc during a radical operation in patients with PC. Third, distant metastasis, such as to the liver, consisted of solid low-attenuation or hypointense masses with poor margin and rim enhancement on contrast-enhanced images [18], peritoneum and omentum presented with nodular thickening and contrast enhancement, or ascites presented without an underlying systemic condition [18, 27]. Fourth, positive lymph nodes were present around the celiac artery (lymph node station 9) or along the left lateral SMA (station 14) or abdominal aorta (station 16), as described in the literature [22, 28]. Criteria for Unresectability at Surgery In this study, peripancreatic vessel involvement was also categorized into three surgical grades according to exploration records: 0, separable peripancreatic vessel without adherence to tumor; 1, peripancreatic vessel adherent to the tumor but separable; and 2, peripancreatic vessel inseparable from or encased within the tumor. Patients were considered to have surgically unresectable tumors on the basis of the following criteria. First, a peripancreatic vessel was inseparable from or encased within the tumor (surgical grade 2) [12]. Although limited invasion of the portal or superior mesenteric vein (< 2 cm segment) was not considered an absolute contraindication to surgery [13, 14], the criterion for unresectability of major vessels at our hospital was inseparable tumor-to-vessel contact at the time of the study. The other criteria were inseparable peripancreatic organ (stomach or colon) with palpable mass [8]; palpable or visible nodule on the surface of the liver [13]; peritoneal or omental seeding, including ascites, in which malignant cells were found with fast cytologic analysis [7, 18]; and enlarged nodes at stations 9, 14, and 16 [22, 28]. Statistical Analysis Comparison of diagnostic accuracy in assessing tumor detection with both image sets was assessed by McNemar test (MedCalc Software for Microsoft Windows version , MedCalc). An ROC curve was generated to analyze sensitivity, specificity, and positive and negative predictive values. The comparison of diagnostic accuracy was estimated with the same software by calculation of the ROC AUC (A z ) and z score for each reviewer for the two image sets in the preoperative evaluation of vascular invasion, metastasis, and tumor resectability. A value of p < 0.05 indicated a statistically significant difference. To evaluate interobserver agreement in evaluating tumor detection and resectability, weighted kappa values were calculated with the same software as for the other tests. A kappa value < 0.20 indicated poor 528 AJR:206, March 2016

4 CT Versus MRI for Presurgical Evaluation of Pancreatic Cancer TABLE 2: Comparison of Surgical Grade With CT and MRI Grades in Cases of Major Vascular Invasion in Patients With Resectable Pancreatic Tumors (n = 24) Patient No. Surgical Grade A Fig year-old man with resectable cancer of pancreatic head. A, Axial portal venous phase CT image shows high-attenuation mass (arrow) at pancreatic head. Neither portal vein nor superior mesenteric artery (SMA) is contiguous with tumor (imaging grade 0). Finding was confirmed at surgery (surgical grade 0). Resected specimen contained 1.8-cm tumor. B, Curved coronal reformation of portal venous phase CT image clearly shows high-attenuation mass (arrow) at pancreatic head lying to right side of superior mesenteric vein (arrowhead) with discernible fat plane and upstream dilated biliary ducts. C, Axial portal venous phase MR image shows hyperintense mass (arrow) at pancreatic head but clearer tumor border, and both portal vein and SMA are intact to tumor (imaging grade 0). Reviewer 1 Reviewer 2 CT Grade MRI Grade CT Grade MRI Grade 1 SMV 1 SMV 1 SMV 1 SMA/SMV 1/1 SMA/SMV 1/1 2 SMPV 1 SMPV 1 SMPV 1 SMPV 1 SMPV SMV 1 SMV 1 SMV 1 SMV 1 SMV SMPV 1 SMPV 1 SMPV 1 SMPV 1 SMPV 1 7 SMPV 1 SMPV 1 SMPV 1 SMPV 1 SMPV SMV 1 SMV 1 SMV 1 SMV 1 SMV 1 11 SMV 1 SMV 1 SMV 1 SMV 1 SMV 1 12 SMV 1 SMV 1 SMV 2 SMV 1 SMV 1 13 CA 2 CA 2 CA 2 CA 2 CA 2 14 SMV 1 SMPV 1 SMPV 1 SMPV 1 SMPV SMV SMV 1 SMV 1 SMV 1 SMV 1 SMV 1 17 SMPV 1 SMPV 2 HA/SMPV 1/2 SMV 2 SMV 1 18 SMPV 1 SMPV 1 SMPV 1 SMPV 1 SMPV SMPV 1 SMPV 1 SMPV 1 SMV 1 SMV 1 21 SMPV 1 SMV 1 SMV 1 SMV 1 SMPV 1 22 SMPV 1 SMPV 2 SMPV 2 SMPV 2 SMV Note SMV = superior mesenteric vein, SMA = superior mesenteric artery, SMPV = superior mesenteric portal vein confluence, CA = celiac artery, HA = hepatic artery. B C AJR:206, March

5 Chen et al. agreement; , fair agreement; , moderate agreement; , good agreement; and 0.80, excellent agreement. Results Standard of Reference Surgical exploration was performed according to protocol for all patients included in the study. The operation included the liver, extrahepatic bile duct, peritoneum, omentum, stomach, colon, and vital vessels in addition to the pancreas. Thirty-one of the 38 patients had a tumor in the head or uncinate process of the pancreas, five had tumors in the body, and two had tumors in the tail. The histologic diagnoses were 32 pancreatic ductal adenocarcinomas and six pancreatic mucinous adenocarcinomas. Thirty patients had moderately differentiated adenocarcinoma, two had poorly differentiated adenocarcinoma, and six had well-differentiated adenocarcinoma. Twenty-four patients underwent standard curative resection. Negative margins were found in 23 patients. In the one patient with positive margins, tumor was found in the pancreatic body during microscopic analysis. The operations were Whipple procedure (n = 3), Child operation (n = 13), pylorus-preserving pancreaticoduodenectomy (n = 1), and distal pancreatectomy with splenectomy (n = 7). At surgery or pathologic examination, 10 patients were found to have A Fig year-old woman with resectable cancer of pancreatic head. A, Axial portal venous phase CT image shows neither low- nor high-attenuation mass at pancreatic head (arrow). Tumor to superior mesenteric vein (SMV) contiguity (arrowhead) is less than 50% (imaging grade 1). At surgery, SMV was adherent to tumor but separable (surgical grade 1). Resected specimen contained 2.2-cm tumor. B, Negative-contrast CT cholangiopancreatogram shows dilatation of biliary tree and main pancreatic duct. Distal segments of bile duct (arrowhead) and pancreatic duct (arrow) form typical four-segment duct sign, which strongly suggests presence of pancreatic head cancer. C, Axial portal venous phase MR image shows neither hypointense nor hyperintense mass at pancreatic head (arrow), and tumor-to-smv contiguity (arrowhead) is less than 50% (imaging grade 1). D, Thick-slab HASTE MRCP image shows same findings as B. Arrow indicates distal segments of pancreatic duct; arrowhead, distal segments of bile duct. lymph node involvement at stations 11 (n = 1), 12 (n = 1), 13 (n = 3), and 17 (n = 5). The mean tumor size in the resected specimens was 2.58 ± 1.23 (SD) cm (range, cm), specifically, 2 cm or less in 10 patients and greater than 2 cm in 14 patients. B Fourteen patients with pancreatic head cancers underwent palliative procedures, including choledochojejunostomy (n = 6), gastrojejunostomy (n = 4), and external biliary drainage (n = 4) because of a single liver metastatic lesion (n = 1), involvement of one or TABLE 3: Comparison of Surgical Grade With CT and MRI Grades in Cases of Major Vascular Invasion in Patients With Unresectable Pancreatic Tumors (n = 14) Patient No. Surgical Grade Reviewer 1 Reviewer 2 CT Grade MRI Grade CT Grade MRI Grade 1 All 5 vessels 2 All 5 vessels 2 All 5 vessels 2 All 5 vessels 2 All 5 vessels 2 2 SMPV 2 SMV 2 SMV. 2 SMA/SMV 1/2 SMV 2 3 SMPV 2 SMPV 2 SMPV 2 SMPV 2 SMPV 2 4 SMPV 2 SMPV 1 SMV 1 PV 1 PV 1 5 SMV 2 SMV 1 SMV 2 SMV 1 SMV/SMA 2/1 6 SMPV 2 SMPV 2 SMPV 2 SMPV 2 SMPV 2 7 NA SMPV SMV 1 8 SMPV 2 SMPV 2 SMPV 2 SMPV 2 SMV 2 9 SMPV 2 SMPV 2 SMPV 2 SMV 2 SMV 2 10 SMPV 2 SMPV 2 SMV 1 SMPV 2 SMPV 1 11 SMA 2 SMA 2 0 SMA HA/SMPV 2/1 HA/SMPV 2/1 SMV 1 SMV 1 SMV 1 13 SMPV 2 SMPV 2 SMA/SMPV 1/2 SMA/SMPV 1/2 SMA/SMPV 1/2 14 SMPV 2 SMPV 2 HA/SMPV 1/2 HA/SMPV 1/2 HA/SMPV 1/2 Note SMPV = superior mesenteric portal vein confluence, SMV = superior mesenteric vein, SMA = superior mesenteric artery, PV = portal vein, NA = not available, HA = hepatic artery. C D 530 AJR:206, March 2016

6 CT Versus MRI for Presurgical Evaluation of Pancreatic Cancer more major vessels (surgical grade 2) (n = 7), vascular invasion with enlarged nodes at station 9 (n = 2), vascular invasion with liver metastasis (n = 1), or vascular invasion with omental milletlike seeding (n = 1) or peritoneal seeding (n = 2). A D Tumor Detection In the CT set, the tumor detection rates were 94.7% (36/38 patients) with a mean tumor size of 2.61 ± 1.32 cm (range, cm) for reviewer 1 and 92.1% (35/38 patients) with a mean tumor size of 2.81 ± 1.34 cm (range, cm) for reviewer 2. Tumors had low attenuation compared with the surrounding pancreatic parenchyma in 35 of 36 patients for reviewer 1 and 34 of 36 patients for reviewer 2. One patient had a high-attenuation mass (Fig. 1). Patients in whom no visible mass was found on CT images had the double-duct sign or four-segment sign, as observed by both reviewers. In the MRI set, the tumor detection rates were 97.4% (37/38 patients) for both reviewers. The tumors were hypointense in 36 of 37 patients and hyperintense in one patient. Mean tumor size was 2.42 ± 1.36 cm (range, cm) for reviewer 1 and 2.59 ± 1.34 cm (range, cm) for reviewer 2. The other patient had no visible mass but had thickening of the bile duct wall and the four-segment sign as visualized on MR images (Fig. 2). There was no statistical difference in the rate of detection of tumors with the two image sets (reviewer 1, p = 1.000; reviewer 2, p = 0.500). Vascular Evaluation A total of 185 major vessels were evaluated during the surgical exploration of the 38 patients: 120 vessels in 24 patients with resectable tumors and 65 vessels in 13 patients with unresectable tumors. One patient with an unresectable tumor underwent no further vascular evaluation because both preoperative CT and MRI findings suggested liver metastasis, which was confirmed with follow-up CT. In this case, the surgical procedure was mainly for implanting 5-fluorouracil particles on the surface of the tumor and establishing external biliary drainage. According to the surgical grading system, 27 vessels in 14 patients were grade 2, including the celiac artery (n = 2), hepatic artery (n = 2), SMA (n = 2), superior mesenteric vein (n = 1), and both superior mesenteric vein and portal vein (n = 10), in which the confluence of the two veins was invaded by the tumor simultaneously. One patient with resectable pancreatic body cancer was found to have celiac artery involvement. However, vascular variations were found in this patient, that is, the left gastric artery directly originated from the abdominal aorta, B Fig year-old man with cancer of pancreatic head found unresectable at surgery. A, Axial arterial phase CT image shows both celiac artery (arrow) and hepatic artery (arrowhead) encased by tumor (imaging grade 2). B, CT portovenogram shows confluence of superior mesenteric vein (arrowhead) and portal vein (arrow) occluded by tumor (imaging grade 2). C, Arterial phase MR image shows findings similar to CT findings in A. Arrow indicates celiac artery; arrowhead, hepatic artery. D, MR portovenogram shows findings similar to CT findings in B. Arrow indicates portal vein; arrowhead, superior mesenteric vein. and the common hepatic artery arose from the SMA. Thus, radical resection was performed. The correlation of surgical grade with CT and MRI grade for vascular invasion in 185 major vessels is listed in Tables 2 and 3, and representative images are shown in Figures 1 4. For reviewer 1, the sensitivity and specificity in assessing vascular involvement (imaging grade 2) were 85.2% and 97.5% with CT and 74.1% and 97.5% with MRI (Table 4). For reviewer 2, the sensitivity and specificity for assessing vascular involvement were 77.8% and 98.1% with CT and 70.4% and 99.4% with MRI. The statistical difference in the diagnosis of vascular invasion with both image sets was not significant for either reviewer (reviewer 1, p = 0.189; reviewer 2, p = 0.494). Tumor Extension and Metastasis In the two image sets, both reviewers correctly identified small liver metastasis (9 mm) in one patient but did not detect a metastatic nodule on the liver surface (5 mm) in another patient. Also in both image sets, both reviewers missed one peritoneal and one omental seeding; they did find, however, that one patient with peritoneal implants had ascites; thus a peritoneal seeding was judged correctly. In the 38 patients, two pancreatic tail cancers and one pancreatic body cancer had adhesions C AJR:206, March

7 Chen et al. TABLE 4: Comparison of CT With MRI Findings in Prediction of Vascular Invasion, Nodal Metastasis, and Tumor Resectability in Patients With Pancreatic Cancer (n = 38) Performance Measure Vascular invasion Reviewer 1 Reviewer 2 CT MRI p CT MRI Sensitivity (%) 85.2 (23/27) 74.1 (20/27) 77.8 (21/27) 70.4 (19/27) Specificity (%) 97.5 (154/158) 97.5 (154/158) 98.1 (155/158) 99.4 (157/158) PPV (%) 85.2 (23/27) 83.3 (20/24) 87.5 (21/24) 95.0 (19/20) NPV (%) 97.5 (154/158) 95.7 (154/161) 96.3 (155/161) 95.2 (157/165) Accuracy (%) 95.6 (177/185) 94.1 (174/185) 95.1 (176/185) 95.1 (176/185) A z Nodal metastasis Sensitivity (%) 64.3 (9/14) 28.6 (4/14) 57.1 (8/14) 50.0 (7/14) Specificity (%) 58.3 (14/24) 70.8 (17/24) 70.8 (17/24) 66.7 (16/24) PPV (%) 47.4 (9/19) 36.4 (4/11) 53.3 (8/15) 46.7 (7/15) NPV (%) 73.7 (14/19) 63.0 (17/27) 73.9 (17/23) 69.6 (16/23) Accuracy (%) 60.5 (23/38) 55.3 (21/38) 65.8 (25/38) 60.5 (23/38) A z Tumor resectability Sensitivity (%) 87.5 (21/24) 83.3 (20/24) 87.5 (21/24) 91.7 (22/24) Specificity (%) 85.7 (12/14) 71.4 (10/14) 78.6 (11/14) 71.4 (10/14) PPV (%) 91.3 (21/23) 83.3 (20/24) 87.5 (21/24) 84.6 (22/26) NPV (%) 80.0 (12/15) 71.4(10/14) 78.6 (11/14) 83.3 (10/12) Accuracy (%) 86.8 (33/38) 78.9 (30/38) 84.2 (32/28) 84.2 (32/38) A z Note PPV = positive predictive value, NPV = negative predictive value, A z = ROC AUC. A Fig year-old man with resectable cancer of pancreatic body. A, Arterial phase CT image depicts left gastric (arrowhead) and celiac arteries originating from abdominal aorta; celiac artery is encased by tumor (arrow). B, Volume-rendered CT arteriogram shows superior mesenteric artery and hepatic artery (curved white arrow) originating as common trunk and left gastric artery (arrowhead) directly originating from abdominal aortic artery. Celiac artery (straight white arrow) arising from abdominal aorta has communication with hepatic artery but is narrowed by tumor (black arrow). Thus, invaded celiac artery could be resected with tumor, which was confirmed at surgery. Resected specimen contained 6.2-cm tumor. C, Arterial phase MR image shows same findings as A. Arrowhead indicates left gastric artery; arrow, celiac artery. D, Volume-rendered MR arteriogram shows same vascular variations as B, but celiac artery (straight white arrow) appears to terminate between invaded segment (black arrow) and hepatic artery (curved white arrow) because of limited spatial resolution. Arrowhead indicates left gastric artery. B p C D 532 AJR:206, March 2016

8 CT Versus MRI for Presurgical Evaluation of Pancreatic Cancer to the stomach, and another pancreatic body cancer had adhesions to both the stomach and the colon. All of these were separable at surgical exploration. Both reviewers overestimated pancreatic body cancer in one patient as having stomach invasion, because obliteration of the fat plane between the tumor and stomach was observed in both image sets. Nodal evaluation with CT had sensitivity and specificity of 64.3% and 58.3% for reviewer 1 and 57.1% and 70.8% for reviewer 2 (Table 4). With MRI, the sensitivity and specificity were 28.6% and 70.8% for reviewer 1 and 50% and 66.7% for reviewer 2. In addition, both reviewers found enlarged nodes at station 9 in two patients in both image sets. The differences in accuracy were not statistically significant in either image set (reviewer 1, p = 0.328; reviewer 2, p = 0.244). Tumor Resectability Compared with the surgical and pathologic results, the CT findings on tumor resectability had a higher A z value than the MRI findings (reviewer 1, vs 0.774; for reviewer 2, vs 0.815), but the statistical differences were not significant (reviewer 1, p = 0.193; reviewer 2, p = 0.813) (Table 4). The main cause of misdiagnosis was vascular involvement. Using CT, reviewer 1 overestimated major vascular involvement in two of 38 patients and underestimated it in two other patients; this reviewer had three overestimations and four underestimations with MRI. Using CT, reviewer 2 overestimated major vascular involvement in two patients and underestimated it in three patients; this reviewer had one overestimation and four underestimations with MRI. For both reviewers, although peritoneal or omental seeding and liver metastases were misdiagnosed in three of five patients, vascular invasion was identified in two of them. Therefore, the condition of only one of those patients was underestimated. In addition, vascular variations in one patient who had celiac artery involvement were correctly identified in both image sets (Fig. 4). Interobserver Agreement Interobserver agreement on tumor detection was for CT versus for MRI; vascular invasion, versus 0.897; nodal metastasis, versus 0.422; and tumor resectability, versus Discussion With the development of surgical techniques, the most important aspects of presurgical evaluation of PC may be the detection of tumor size, prediction of vascular involvement, and identification of microspread with CT or MRI [5, 7, 13]. Correct visualization of peripancreatic major vessels and metastases no doubt affects surgical strategies and patient prognosis. Unfortunately, small liver and peritoneal metastases and vascular involvement remain important pitfalls in preoperative evaluation [8]. Accordingly, many reports have described the necessity for and efficiency of comprehensive imaging techniques in the assessment of PC [7, 13 15, 27, 29, 30]. Erturk et al. [7] reported that the addition of multiplanar reformation images to the CT technique significantly improved the accuracy of tumor detection. The high specificity of MRI may have been the result of using MRCP images in their MRI examination protocol. Park et al. [13] also found that dynamic contrast-enhanced MRI with MRCP had superior tumor conspicuity and comparable diagnostic performance to those of MDCT in the preoperative staging of PC. Our results showed that detection of tumors with both image sets was not significantly different between reviewers. These results were similar to those in a previous report [15] that CT and MRI had equivalent detection rates in classifying pancreatic malignant lesions (including metastases, ampullary carcinomas, and adenocarcinomas) or even in differentiating pancreatic head carcinoma from other types of periampullary carcinomas (e.g., ampullary carcinoma, distal bile duct carcinoma, and duodenal carcinoma) [11]. This may be attributed to the use of near-isotropic volume data, which allows excellent 3D reformatting in any plane [16, 18] and improves the detection of PC at CT [11, 31]. Nevertheless, higher conspicuity of tumors and interobserver agreement were observed for MRI than for CT for both reviewers in our study. Because of the greater soft-tissue contrast of MRI compared with that of CT, a PC that is small or isoattenuating on CT images may be identified more easily and accurately with MRI [9, 13]. This may explain the smaller difference in mean tumor size on MR images than on CT images for both reviewers in our study. It has been hypothesized that identification of tumor size, CT attenuation, and MRI signal intensity will benefit not only the determination of tumor extent but also the assessment of prognosis in patients with PC because isoattenuating and hyperattenuating or isointense and hyperintense tumors have relatively well-differentiated pathologic characteristics and better survival rates than the usual hypoattenuating and hypointense tumors [3, 4, 13]. The histopathologic findings of isoattenuating PC include welldifferentiated tumor cells, remaining atrophic acini, and a predominant proportion of fibrous stroma consisting mainly of loose fibrosis [4]. Takeshita et al. [3] also reported that isoattenuating and hyperattenuating enhancement on CT images may be considered to reflect an abundant vascular structure in smaller PCs (< 2 cm). In our study, both reviewers found one hyperattenuating-hyperintense tumor was found in both image sets and another isointense tumor on MR images. On CT images, reviewer 1 found two and reviewer 2 found three isoattenuating tumors. For those isoattenuating-isointense tumors, both negative-contrast CT cholangiopancreatography and MRCP were helpful for locating the mass in the pancreas on the basis of the dilated upstream pancreatic duct. For the evaluation of major vessel involvement, earlier studies showed that the sensitivity of CT in revealing venous invasion was 75% [24] and the accuracy was 100% [16] in patients with pancreatic carcinoma when the criterion of 50% tumor-to-vessel contact was used. One study [15] showed that sensitivity and specificity in assessing vessel infiltration were 70 90% and 98 98% for CT and 50 80% and 96 98% for MRI when they used the criteria proposed by Lu et al. [26]. Lu et al. proposed a 5-point grading system based on the degree of contact between the tumor and a major vessel. In their study, tumor contiguity with more than 50% of the perimeter of a vessel was found to be the optimal threshold for predicting vascular invasion, having sensitivity of 84% and specificity of 98%. Lee et al. [14] used separate criteria for arterial and venous invasion, particularly with encasement, vessel margin irregularity, or tumor incursion into the periarterial fat plane with the tumor lying in juxtaposition to the vessel for arterial invasion. They found that the sensitivity and specificity were 61 64% and 94 96% for CT and 57 57% and 98 99% for MRI. In this study, we used a three-criterion imaging grading system based on the literature [18, 25] to assess tumor-to-vessel contiguity compared with surgical exploration records. Although overestimation and underestimation were observed with both modalities for both reviewers (Tables 2 and 3), our outcomes showed that the sensitivity, specificity, and AJR:206, March

9 Chen et al. accuracy were similar to those in a previous report [15]. Sensitivity, however, was higher than in the study by Lee and colleagues [14] for both image sets. In our study, the sensitivity and negative predictive value of predicting vascular involvement were higher with CT than MRI for both reviewers. Meanwhile, the specificity and positive predictive value were equal to or greater for CT than MRI for reviewer 1 but greater for MRI than for CT for reviewer 2. Although the accuracy differences were not statistically significant, the A z values were higher for CT than for MRI for both reviewers. In addition, CT arteriography had higher spatial resolution than did MR arteriography in outlining the major peripancreatic vessels (Fig. 4). In contrast to a 5-point grading system [26], our imaging grading system seemed simple when a threshold of greater than 50% tumor contiguity was used as a criterion of unresectability in predicting vascular involvement, which is also recommended by National Comprehensive Cancer Network guidelines [18]. The presence or absence of peripancreatic major organ invasion and of liver, nodal, and peritoneal metastasis is another important factor in predicting tumor resectability. It has been reported [12, 15] that MRI is superior to CT in the detection of hepatic, peritoneal, and omental metastases. In our study, however, both image sets correctly depicted liver metastasis in one patient but missed one small metastatic nodule (5 mm) in another patient. In addition, one peritoneal and one omental seeding were underestimated, and in both image sets both reviewers overestimated the condition of another patient with pancreatic body cancer as having stomach invasion. The results were similar to those in previous reports [13, 18] and suggested that microspread (i.e., small liver metastases and milletlike seeding on the peritoneum and omentum) is still difficult to detect with CT and MRI [8, 9, 27]. This may be the cause of limited spatial resolution with the current modalities. In addition, prediction of nodal metastasis on the basis of size, attenuation, and signal intensity had a lower accuracy and A z value. The accuracy was % and the A z value for CT, and the respective values were % and for MRI. The results are similar to those of other studies. In one study [32], the accuracy of CT was 63 81%. In another study [13], the A z value for lymph node metastasis was for CT and for MRI for two reviewers. This may reflect the fact that both the morphometric and characteristic evaluation criteria are not specific for discriminating normal and metastatic nodes [32]. However, the low rate of detection of metastatic nodes has limited importance because most of the nodes are resected at surgery [24]. Suspicion of extraregional lymph node metastases on the basis of CT findings alone should not be considered a contraindication to exploration [32], although nodal status is an important predictor of survival in patients with resectable PC [28]. In the tumor resectability evaluation, diagnostic accuracy was not statistically significant for either reviewer or either image set. Our results also showed higher sensitivity, specificity, and accuracy of CT and MRI than did the study by Park et al. [13]. Compared with the results of Lee et al. [14] and Koelblinger et al. [15], we found similar sensitivities but higher specificities for CT. Although small metastases in the liver and peritoneum-omentum were underestimated in three patients in our study, two of these metastatic lesions were evaluated correctly as unresectable because both reviewers identified vascular invasion in both image sets. Thus, the surgical treatment of those patients remained unchanged. Liver metastasis or peritonealomental seeding coexisting with major vascular involvements may not be rare, according to a previous report [15], and may indicate that careful analysis of major vessels will be helpful in avoiding pitfalls in presurgical evaluation of PC with the two modalities. Our study had several limitations. First, this retrospective study included only patients with surgically proven tumors; many patients with advanced tumors were excluded on the basis of presurgical CT or MRI findings. Therefore, the population was fairly small. Second, the enhanced CT protocol had only arterial and portal venous phases. The pancreatic phase, the use of which may increase the tumor detection rate [27, 29], was not included in this study. However, we performed the portal venous phase with a 60- to 68-second delay, which may be a reasonable alternative, as suggested in previous studies [3, 10]. In addition, portal venous phase images can be postprocessed into negative-contrast CT cholangiopancreatograms and CT portovenograms, which is valuable for visualizing the pancreaticobiliary system and evaluating vascular involvement. Third, the reviewers were aware that the purpose of the study was to evaluate PC without other periampullary tumors and the relation of PC to peripancreatic major structures. Therefore, confidential bias of the interpretation may have been present compared with daily practice [7]. A last limitation was that both 1.5-T and 3-T MRI with different protocols were used in this study, which may have affected the outcomes (e.g., in assessment of vascular invasion). Because of the higher contrast-tonoise and signal-to-noise ratios with 3-T than with 1.5-T MRI, contrast-enhanced MRA at 3-T has higher spatial resolution than at 1.5-T [33]. For example, with 1.5-T MRI, both reviewers underestimated the vascular invasion in 3 of 14 unresectable tumors as opposed to underestimating it in 1 of 14 unresectable tumors with 3-T MRI. This may be due to the use of thinner sections at 3 T than at 1.5 T in this study. Thus, further study may be required to evaluate the potential benefits to imaging the abdomen with 3-T MRI. Conclusion Our results showed that CT and MRI had similar performance in presurgical evaluation of PC. Although CT is the preferred method for initial imaging evaluation of patients with suspected PC [13, 16 18], in view of our outcomes and National Comprehensive Cancer Network guidelines, either pancreas-specific CT or pancreas-specific MRI is the preferred technique for evaluating PC. Thus, it is probably best to use the strengths of a given institution in terms of equipment, experience, and skill when choosing which modality to use for presurgical evaluation. References 1. Shrikhande SV, Barreto SG, Goel M, Arya S. Multimodality imaging of pancreatic ductal adenocarcinoma: a review of the literature. HPB 2012; 14: Páez D, Labonte MJ, Lenz HJ. Pancreatic cancer: medical management (novel chemotherapeutics). Gastroenterol Clin North Am 2012; 41: Takeshita K, Kutomi K, Haruyama T, et al. Imaging of early pancreatic cancer on multidetector row helical computed tomography. Br J Radiol 2010; 83: Yoon SH, Lee JM, Cho JY, et al. Small ( 20 mm) pancreatic adenocarcinomas: analysis of enhancement patterns and secondary signs with multiphasic multidetector CT. Radiology 2011; 259: Tamm EP, Bhosale PR, Vikram R, de Almeida Marcal LP, Balachandran A. Imaging of pancreatic adenocarcinoma: state of the art. World J Radiol 2013; 5: Denecke T, Degutyte E, Stelter L, et al. Minimum 534 AJR:206, March 2016

10 CT Versus MRI for Presurgical Evaluation of Pancreatic Cancer intensity projections of the biliary system using 15. Koelblinger C, Ba-Ssalamah A, Goetzinger P, et ment of resectability before surgery. AJR 2002; 16-channel multidetector computed tomography al. Gadobenate dimeglumine enhanced 3.0-T 178: in patients with biliary obstruction: comparison MR imaging versus multiphasic 64 detector row 25. Tran Cao HS, Balachandran A, Wang H, et al. Ra- with MRCP. Eur Radiol 2006; 16: CT: prospective evaluation in patients suspected diographic tumor-vein interface as a predictor of 7. Erturk SM, Ichikawa T, Sou H, et al. Pancreatic of having pancreatic cancer. Radiology 2011; intraoperative, pathologic, and oncologic out- adenocarcinomas: MDCT versus MRI in the de- 259: comes in resectable and borderline resectable tection and assessment of locoregional extension. J Comput Assist Tomogr 2006; 30: Kaneko OF, Lee DM, Wong J, et al. Performance of multidetector computed tomographic angiography in determining surgical resectability of pancreatic head adenocarcinoma. J Comput Assist Tomogr 2010; 34: Lee ES, Lee JM. Imaging diagnosis of pancreatic cancer: a state-of-the-art review. World J Gastroenterol 2014; 20: Wang FB, Ni JM, Zhang ZY, et al. Differential diagnosis of periampullary carcinomas: comparison of CT with negative-contrast CT cholangiopancreatography versus MRI with MR cholangiopancreatography. Abdom Imaging 2014; 39: Li B, Zhang L, Zhang ZY, et al. Differentiation of noncalculous periampullary obstruction: comparison of CT with negative-contrast CT cholangiopancreatography versus MRI with MR cholangiopancreatography. Eur Radiol 2015; 25: Arslan A, Buanes T, Geitung JT. Pancreatic carcinoma: MR, MR angiography and dynamic helical CT in the evaluation of vascular invasion. Eur J Radiol 2001; 38: Park HS, Lee JM, Choi HK, Hong SH, Han JK, Choi BI. Preoperative evaluation of pancreatic cancer: comparison of gadolinium-enhanced dynamic MRI with MR cholangiopancreatography versus CT. J Magn Reson Imaging 2009; 30: Lee JK, Kim AY, Kim PY, Lee MG, Ha HK. Prediction of vascular involvement and resectability by multidetector-row CT versus MR imaging with MR angiography in patients who underwent surgery for resection of pancreatic ductal adenocarcinoma. Eur J Radiol 2010; 73: Catalano C, Laghi A, Fraioli F, et al. Pancreatic carcinoma: the role of high-resolution multislice spiral CT in the diagnosis and assessment of resectability. Eur Radiol 2003; 13: Grenacher L, Klauss M, Dukic L, et al. Diagnosis and staging of pancreatic carcinoma: MRI versus multislice-ct a prospective study. Rofo 2004; 176: Al-Hawary MM, Francis IR, Chari ST, et al. Pancreatic ductal adenocarcinoma radiology reporting template: consensus statement of the Society of Abdominal Radiology and the American Pancreatic Association. Gastroenterology 2014; 146: Zhang ZY, Li JP, Hu CH, et al. Improvement in imaging time and quality of 3D negative-contrast computed tomography cholangiography with minimum intensity projections: application of vari-slice manual cut and erosion functions. Clin Imaging 2009; 33: Kim TU, Kim S, Lee JW, et al. Ampulla of Vater: comprehensive anatomy, MR imaging of pathologic conditions, and correlation with endoscopy. Eur J Radiol 2008; 66: Kim JH, Kim MJ, Chung JJ, et al. Differential diagnosis of periampullary carcinoma at MR imaging. RadioGraphics 2002; 22: Japan Pancreas Society. Classification of pancreatic carcinoma. 2nd English ed. Tokyo, Japan: Kanehara, 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: Valls C, Andía E, Sanchez A, et al. Dual-phase helical CT of pancreatic adenocarcinoma: assess- pancreatic cancer. J Gastrointest Surg 2014; 18: Lu DS, Reber HA, Krasny RM, Kadell BM, Sayre J. Local staging of pancreatic cancer: criteria for unresectability of major vessels as revealed by pancreatic-phase, thin-section helical CT. AJR 1997; 168: Wong JC, Raman S. Surgical resectability of pancreatic adenocarcinoma: CTA. Abdom Imaging 2010; 35: Tol JA, Gouma DJ, Bassi C, et al. Definition of a standard lymphadenectomy in surgery for pancreatic ductal adenocarcinoma: a consensus statement by the International Study Group on Pancreatic Surgery (ISGPS). Surgery 2014; 156: Ichikawa T, Erturk SM, Sou H, et al. MDCT of pancreatic adenocarcinoma: optimal imaging phases and multiplanar reformatted imaging. AJR 2006; 187: Sandrasegaran K, Lin C, Akisik FM, Tann M. State-of-the-art pancreatic MRI. AJR 2010; 195: Fukushima H, Itoh S, Takada A, et al. Diagnostic value of curved multiplanar reformatted images in multislice CT for the detection of resectable pancreatic ductal adenocarcinoma. Eur Radiol 2006; 16: Tseng DS, van Santvoort HC, Fegrachi S, et al. Diagnostic accuracy of CT in assessing extra-regional lymphadenopathy in pancreatic and periampullary cancer: a systematic review and metaanalysis. Surg Oncol 2014; 23: Chang KJ, Kamel IR, Macura KJ, Bluemke DA. 3.0-T MR imaging of the abdomen: comparison with 1.5 T. RadioGraphics 2008; 28: AJR:206, March