Differentiation of pancreatic neuroendocrine carcinoma from pancreatic ductal adenocarcinoma using magnetic resonance imaging: The value of contrast-enhanced and diffusion weighted imaging

Pancreatic neuroendocrine carcinoma (PNEC) is often misdiagnosed as pancreatic ductal adenocarcinoma (PDAC). This retrospective study differentiated PNEC from PDAC using magnetic resonance imaging (MRI), including contrast-enhanced (CE) and diffusion-weighted imaging (DWI). Clinical data and MRI findings, including the T1/T2 signal, tumor boundary, size, enhancement degree, and apparent diffusion coefficient (ADC), were compared between 37 PDACs and 13 PNECs. Boundaries were more poorly defined in PDAC than PNEC (97.3% vs. 61.5%, p<0.01). Hyper-/isointensity was more common in PNEC than PDAC at the arterial (38.5% vs. 0.0), portal (46.2% vs. 2.7%) and delayed phases (46.2% vs. 5.4%) (all p<0.01). Lymph node metastasis (97.3% vs. 61.5%, p<0.01) and local invasion/distant metastasis (86.5% vs. 46.2%, p<0.01) were more common in PDAC than PNEC. Enhancement degree via CE-MRI was higher in PNEC than PDAC at the arterial and portal phases (p<0.01). PNEC ADC values were lower than those of normal pancreatic parenchyma (p<0.01) and PDAC (p<0.01). Arterial and portal phase signal intensity ratios and ADC values showed the largest areas under the receiver operating characteristic curve and good sensitivities (92.1%–97.2%) and specificities (76.9%–92.3%) for differentiating PNEC from PDAC. Thus the enhancement degree at the arterial and portal phases and the ADC values may be useful for differentiating PNEC from PDAC using MRI.


INTRODUCTION
Pancreatic ductal adenocarcinoma (PDAC) is the most common malignant tumor of the pancreas. It is highly aggressive and rapidly fatal, with a five-year survival rate <5% [1]. While resection can be curative, resection rates remain low at 10-15% [2] due to local invasion or distant metastases. The PDAC characteristic vascularization pattern can be visualized using computed tomography (CT) or magnetic resonance imaging (MRI) [3,4], and PDACs are often hypovascularized as compared to adjacent normal tissue [5].
PDAC and PNEC treatment strategies and prognoses differ. For PNEC, surgery is indicated if curative resection is possible, even in those cases with limited metastases, for example to liver [19,20]. In addition, targeted therapy with sunitinib or everolimus [21] and somatostatin analogues (octrecotide) [22], or radionuclide-labeled somatostatin (DOTATATE) [23] may be valuable for PNEC, along with cytotoxic chemotherapy (e.g., cisplatin with etoposide). PNEC are generally less aggressive and have better outcomes than PDAC. The PNEC five-year survival rate is approximately 27.7% [24], which is higher than that of PDAC (<5%). Pretreatment differentiation of PNEC from PDAC is important in determining therapeutic strategies. To the best of our knowledge, no study has explored differences in imaging features between PDAC and PNEC. MRI, particularly in diffusion-weight imaging (DWI), has been used to differentiate pancreatitis and pancreatic cancer [25][26][27][28], and MRI has a similar or better performance in PDAC evaluation [29,30]. The present study assessed the value of MRI, including DWI and dynamic contrast-enhanced imaging, for differentiating PNEC and PDAC.

Patient and tumor characteristics
Thirty-seven PDAC and 13 PNEC patients were analyzed ( Figure 1, Table 1) in this retrospective study. Thirty-one PDAC patients underwent surgery and six underwent biopsy. Twelve PNEC patients underwent surgery and one underwent biopsy. We compared demographic data and clinical symptoms between PDAC and PNEC patients. No differences were found for age, gender, or clinical symptoms between those two groups. However, yellow urine or icterus was more common in PDAC compared with PNEC patients (27.0% vs. 7.7%, p>0.05). PNEC tended to occur in men compared with PDAC (76.9% vs. 62.2%, p>0.05). Carbohydrate antigen 19-9 (CA19-9) and CA125 levels were higher in PDAC than in PENC patients. Abnormal CA19-9 level was more common in PDAC than PNEC (81.8% vs 30.8%, p<0.05). In our series, 89.2% of PDAC patients were correctly diagnosed via MRI, while eight (61.5%) PNEC patients were misdiagnosed as PDAC via MRI.
Additionally, PNEC boundaries ( Figure 3) were relatively well defined compared with those of PDAC ( Figure 2).

MRI findings: quantitative analyses
We quantitatively analyzed tumor sizes, signal intensities in T1 weighted images, and apparent diffusion coefficient (ADC) values. Mean PNEC tumor size was greater than that of PDAC (5.2 cm vs. 3.3 cm, p=0.03) ( Table 2). Figure 4 shows the signal intensity ratio in unenhanced (pre-contrast image) and contrast-enhanced T1 weighted images measured by two readers. PDAC signal intensity ratios were lower than those of PNEC at arterial and portal phases (p≤0.01) (Figure 4).
Representative DWI and ADC value maps for PDAC and PNEC are shown in Figure 5. PDAC DWI signal intensity was lower than that of PNEC. Mean ADC values for pancreatic parenchyma, PDAC, and PNEC were 1.38, 1.04, and 0.87×10 -3 mm 2 /s (pooled data), respectively ( Figure 6). In both readers, mean PDAC and PNEC ADC values were lower than in normal pancreas parenchyma (p<0.05 or 0.01). PNEC ADC values were also lower compared with those of PDAC (p< 0.01).

Imaging feature diagnostic performances
The sensitivity and specificity of the different imaging features for PNEC identification (vs. PDAC) ranged from 52.6%-97.2% and 38.5%-100%, respectively ( Table 3). The area under the curve (AUC) ranged from 0.667-0.954 (Table 3). Signal intensity ratio at arterial and portal phases, and ADC value had the largest AUC, indicating that these features can potentially differentiate PNEC from PDAC (Figure 7). Cutoff values were 0.768 for signal intensity ratio at arterial phase with 97.2% sensitivity and 92.3% specificity, 0.823 for signal intensity ratio at portal phase with 97.2% sensitivity and 76.9% specificity, and 1.0×10 -3 mm 2 /s for ADC values with 92.1% sensitivity and 91.7% specificity.

DISCUSSION
PNEC can mimic PDAC in CT or MR images due in part to a characteristic hypoenhanced pattern [8,18]. It is valuable to accurately diagnose patient tumor type (PNEC vs. PDAC) before surgery, because treatment strategies and prognoses differ between the two types. The present study compared PNEC MR imaging features with those of PDAC. We found that ill-defined tumor boundaries, hypointensity in contrast-enhanced images, lymph node metastases, and local invasion/distant metastases were more common in PDAC than PNEC. Quantitative data further indicated that signal intensity ratios at arterial and portal phases, and ADC values have potential for differentiating those two tumors.
PDAC is a fast-progressing malignancy with surrounding tissue invasion and metastases to distant organs. Tumors usually exhibit ill-defined margins [13]. Lymph nodes metastases and local invasion or distant metastases are also common in PDAC [31]. In our series, 97% of PDAC cases had ill-defined margins and 97% exhibited lymph node metastases, which was consistent with previous studies. PNECs also frequently exhibited ill-defined boundaries (61%), and often metastasized to lymph nodes (61%) or invaded the surrounding tissues. However, these features were still more common in PDAC than PNEC. To some degree, these qualitative features may be useful for differentiating PNEC from PDAC.
Contrast-enhanced CT and MRI are helpful for differential diagnosis of benign and malignant lesions based on characteristic vascularization patterns [13,32,33]. Several studies showed differences between pancreatic carcinoma and mass-forming focal pancreatitis using contrast-enhanced approaches [31,33]. PDACs are frequently hypovascular, and enhancement degree is lower than the surrounding pancreatic parenchyma, in particular at arterial and portal phases. However, most PNECs also showed hypoenhancement at arterial and portal phases. Thus, it is challenging to discriminate PNEC from PDAC using regularly qualitative features in contrast-enhanced images. We speculated that quantitative analysis of imaging features would provide more useful diagnostic information. Our data showed that PNEC enhancement degree is higher than that of PDAC at arterial and

Figure 4: The signal intensity ratio compared with parenchyma in pancreatic ductal adenocarcinoma (PDAC) and pancreatic neuroendocrine carcinoma (PNEC) calculated by two readers. Signal intensity ratios in PDAC were lower than
PNEC at arterial and portal phases. * p≤ 0.01, compared with PDAC. The bias between the two readers were -1.2%(-9.6%, 7.2%), -2.2% (-14.0%, 9.5%) and 1.4% (-6.8%, 9.5%) for signal intensity ratio at arterial, portal and delayed phases, respectively. portal phases. The signal intensity ratios at arterial and portal phases showed good sensitivity and specificity in differentiating PNEC from PDAC. Our results confirmed that quantitative assessment provides reliable information in differentiating PNEC and PDAC.
DWI has also been widely used to differentiate benign and malignant pancreatic lesions [27,33], and may be applicable for grading PNENs [12,34] and differentiating PNENs and PDAC [35,36]. Lee, et al. and Wang, et al. found no differences in ADC values between PNENs and PDAC [35,36]. However, G1 and G2 PNETs, which may have higher ADC values than PNECs [12,35], were included in these studies. Therefore, we speculate that it is important to consider PNEN grade when evaluating ADC values. As PNECs frequently exhibited high mitotic ability (>20 mitoses per 10 high powered fields (HPFs)) and proliferation (Ki-67 index >20%) in our study, PNECs should therefore exhibit high tissue cellularity and restricted water mobility. Our data supported this speculation. PNEC ADC values were lower than those of PDAC. Moreover, ADC showed very good sensitivity and specificity in differentiating PNEC from PDAC. The data from two readers were pooled together. AER: signal intensity ratio at arterial phase; PER: signal intensity ratio at portal phase; DER: signal intensity ratio at delayed phase; ADC: apparent diffusion coefficients; AUC: area under curve; CA19-9: Carbohydrate antigen 19-9; CI: confidential interval Our study had several limitations. First, retrospective studies may exhibit selection and verification bias. Some PNEC patients were excluded due to lack of MRI examinations, and the PNEC study population could not reflect the entire PNEC spectrum. Second, only CE-MR and routine DWI were studied, and some novel approaches, such as intravoxel incoherent motion (IVIM) DWI and texture analysis, would provide more information [37,38]. Finally, larger number of b values would be more sensible rather to use only two b values [37].
In conclusion, we compared PDAC and PNEC MR imaging features, and found that ill-defined margins, lymph node metastases, and local invasion are more common in PDAC. In addition, PNEC enhancement degree at arterial and portal phases are higher than those of PNEC. PNEC usually exhibits lower ADC values than PDAC. Enhancement degree at arterial and portal phases and ADC value may be useful for differentiating PNEC from PDAC.

Patient selection
This study included patients treated for PDAC and PNEC at our institution between September 2015 and July 2016, and September 2012 and July 2016, respectively. A total of 69 patients with surgically or biopsy diagnosed PDAC were identified. Twenty-two patients were excluded due to lack of MRI examination. Ten patients were excluded because of only a singlephase scan (n=5) or lack of DWI (n=5). We identified 19 patients with surgically or biopsy diagnosed PNEC. Six were excluded due to metastatic NEC (n=2), or lack of MRI examination (n=3) or DWI (n=1). In total, 37 PDAC and 13 PNEC patients were included in this study ( Figure  1). The pathological diagnosis of PNEC was based on the WHO 2010 classification for NENs: NEC G3, >20 mitoses per 10 HPF, Ki-67 index >20%. In addition, demographic and clinical data were retrieved from medical records. This retrospective study was approved by our institutional review board with waiver of patient formal consent.

MR imaging analysis
All MR images were retrospectively reviewed by two abdominal radiologists with more than eight years of experience in abdominal MRI examination on a picture archiving communication system (PACS) workstation. The radiologists were blinded to the final histopathological results and MR diagnosis. The following imaging information was reviewed: tumor position (headneck or body-tail), tumor margin [well-defined: smooth or lobulating margin with few spiculations or infiltrations (<20%); ill-defined: the perimeter of the tumor showed spiculation or infiltration (>20%)] (11), size, presence of cystic components (solid, cystic components <25%; or mixed cystic-solid), T1 and T2 signal, signal on DWI, enhancement degree at arterial, portal, and delayed phases (hypo-, iso-, or hyperintense compared with normal pancreas). The presence of intrahepatic/extrahepatic bile duct dilatation, pancreatic duct dilatation, pancreatic parenchymal atrophy, lymph node metastases, and local invasion/distant metastases were also reviewed. Pancreatic duct dilation was defined as the main pancreatic duct diameter ≥4 mm. Intrahepatic and extrahepatic bile duct dilatation were confirmed if the duct diameter was >5 mm and >8 mm, respectively. Areas that were hypointense on precontrast T1-weighted images, markedly hyperintense on T2-weighted images, and with no enhancement were identified as cystic components.
Quantitative analyses were performed using ADC values and tumor signal intensities on unenhanced and enhanced T1 weighted images by two abdominal radiologists. The signal intensity ratio of tumor to pancreas [signal intensity ratio=signal intensities of tumors/ normal pancreatic parenchyma] was calculated. Cystic components were avoided during signal intensity analysis. On DWI and T1 weighted images, regions of interest (ROI) were centered on the solid tumor portions while avoiding necrotic or cystic components, and the most peripheral portions that might result in partial volume effects of adjacent extra-lesional tissues. Tumor and pancreatic parenchyma ADC values were measured. For the normal pancreas, signal intensities and ADC values were noted at a similar ROI avoiding the main duct. ADC values and signal intensities were measured at least three times by each radiologist. The means and the agreement between the two readers were analyzed.

Pathology analysis
Tumor tissue specimens were fixed with 10% formalin, embedded in paraffin, and sliced for hematoxylin-eosin (H&E) staining. PDAC was typically characterized by moderately to poorly differentiated glandular structures. PNEC was diagnosed based on the 2010 WHO classification for neuroendocrine neoplasms by counting the number of mitoses per 10 HPFs and assessing Ki-67 proliferation index (percentage of positive cells in areas of highest nuclear labeling). NEC G3 was regarded as >20 mitoses per 10 HPF, Ki-67 index >20%.