This study was approved by our institutional review board and the requirement for informed patient consent was waived due to the retrospective nature of the study.

A search of our histopathology and radiology records for the period June 2013-December 2015 revealed 66 patients with resected PanNEN, whose data were considered for inclusion. Inclusion criteria were: histopathological diagnosis of PanNEN after surgical resection and availability of good-quality preoperative MR examinations. Exclusion criteria were: lack of preoperative MR examinations (8 patients) and poor-quality MR examinations (3 patients).

Thirty subjects included in this study have been reported in a previous paper[14]; this prior article dealt with identification and characterization of PanNENs using MR and DW imaging, whereas in this manuscript we report on the correlation between MR and DW imaging findings, tumor grade and biological behavior expressed as the stage at diagnosis.

The final study population comprised 55 patients with a mean age of 57.6 years (range, 21-83 years): 29 men (mean age, 58.2 years; range, 35-83 years) and 26 women (mean age, 57 years; range, 21-77 years).

At histopathological examination of surgical specimens, 31 tumors (56.4%) were classified as G1, 20 (36.4%) as G2, and 4 (7.2%) as G3, according to the 2010 WHO classification[1].

Four patients underwent tumor enucleation or middle pancreatectomy without lymphadenectomy, therefore the TNM classification could not be applied; accordingly, the remaining 51 lesions were staged as follows, according to the ENETS TNM classification[15]: stage I, 15 lesions (29.4%); stage IIA or B, 10 lesions (19.6%); stage IIIA or B, 19 lesions (37.2%); stage IV, 7 lesions (13.7%).

All patients signed an informed consent before MR examination. MR imaging was performed in a single institution using a 1.5 T unit (Magnetom Avanto or Aera; Siemens, Erlangen, Germany) using phased-array body coils. The patients were asked to fast for 4-6 h before MR. One hundred mL of pineapple juice were administered 10-20 min before MR to prevent signal overlapping from fluid-containing organs on T2-weighted images. The MR imaging protocol is reported in Table 1. For DWI, b-values of 50, 400 and 800 s/mm² were used. DW images were acquired using a free-breathing echo-planar sequence; apparent diffusion coefficient (ADC) maps were obtained with a monoexponential decay model. Contrast-enhanced MR imaging involved a quadriphasic study; 0.1 mmol of a gadolinium chelate (gadobenate dimeglumine, MultiHance; Bracco, Milan, Italy) per kilogram of body weight were administered at 2.0-2.5 mL/s by using a power injector (Medrad, Pittsburgh, Pa., United States). Images were obtained before contrast material administration and during the late arterial/pancreatic, portal/venous and delayed phases (35-45, 75-80 and > 180 s after the start of contrast material administration).

Qualitative and quantitative MR features of each tumor were retrospectively analyzed on a workstation by two radiologists (RDR and PTM, with 6 and 26 years of experience in abdominal radiology); discrepancies were solved by consensus. Aside from knowing that the patients had PanNENs, both readers were blinded to all other histopathological features. The qualitative image analysis included: (1) location of the lesion; (2) margins; (3) presence of intratumoral cystic/necrotic areas; (4) main pancreatic duct (MPD) and/or common bile duct (CBD) dilation upstream to the lesion; (5) atrophy of the upstream parenchyma; (6) involvement of peripancreatic vessels; (7) liver metastases; (8) signal intensity of the lesion related to the adjacent parenchyma (hypo-, iso- or hyperintense); and (9) type of enhancement (homogeneous or heterogeneous). Quantitative image analysis included: (1) maximum diameter of the lesion; (2) caliber of the MPD/CBD upstream to the lesion; and (3) mean ADC values of the tumor. Pancreatic tail tumors were excluded from the evaluation of MPD features and pancreatic atrophy; pancreatic body/tail lesions were excluded from the evaluation of CBD dilation and caliber. ADC quantification was performed drawing multiple free-hand regions of interests on several contiguous axial slices to include nearly the whole tumor volume; the results were averaged to obtain the mean ADC value. Care was taken during ROI placement to avoid inhomogeneities; each ROI was positioned slightly inside the outer edge of each lesion, thereby preventing partial volume artifacts. The positioning of the ROI was performed in correlation with the other sequences.

For the purpose of statistical analysis, G2 and G3 tumors, as well as stage III and stage IV PanNENs were grouped and considered together.

Categorical variables derived from the qualitative analysis were compared between groups using the χ² or Fisher’s exact tests; the results are presented as numbers of cases and relative frequencies. Continuous variables derived from the quantitative analysis were compared between groups using Student’s t test or univariate ANOVA; the results are reported as mean ± SD. Student’s t tests and bivariate correlation tests were conducted to assess whether qualitative and quantitative variables were correlated with tumor size. For each significant qualitative feature, the diagnostic performance in the identification of G2-3 and stage III-IV lesions was tested by calculating sensitivity (Se, %), specificity (Sp, %), positive and negative predictive values (PPV/NPV, %). A receiving operator characteristic (ROC) analysis evaluated the diagnostic performance of significant quantitative features, and optimal cut-off values, along with Se, Sp, and PPV/NPV, were thus ascertained. Diagnosis obtained by pathological examination was the external gold standard. The sensitivity and specificity of combinations of the statistically significant MR criteria were also calculated. Statistical analysis was performed with SPSS software version 21.2 (IBM, Chicago, Ill., United States). All P values were considered statistically significant if ≤ 0.05.

Details on the qualitative features are reported in Tables 2 and 3. Most lesions were located in the body or in the head of the pancreas, without significant differences in terms of location between tumor grades and stages (P = 0.094 and 0.059, respectively). Most tumors presented well defined margins; among tumors presenting ill-defined margins, most were found to be G2-3 (13/16; Figure 1) or stage III-IV (15/16) at histopathological analysis, with significantly higher frequency when compared to G1 and stage I-II tumors (81.2% vs 18.8% and 93.8% vs 6.2%, P < 0.001). A significantly higher proportion of G2-3 tumors presented infiltration of peri-pancreatic vessels and liver metastases compared to G1 tumors (25% vs 3.2% and 29.2% vs 0%, P = 0.022 and 0.002, respectively; Figure 2). Atrophy of upstream pancreatic parenchyma and dilation of the MPD/CBD were more common among G2-3 tumors and stage III-IV lesions as compared to G1 and stage I-II tumors, although without significant differences between groups (P = 0.446/0.151, 0.215/0.151, and 0.264/0.275).

Open in New Tab   Full Size Figure   Download Figure

Figure 1 Small hyperfunctioning pancreatic neuroendocrine neoplasm (insulinoma) classified as G1, stage 1 tumor at histology (Ki67 < 1%, T1, N0, M0) in a 44 years-old woman. A: Axial volumetric interpolated breath-hold examination (VIBE) gradient echo image (repetition time msec/echo time msec, 4.3/1.4) with fat suppression shows a homogeneously hypointense lesion with well defined margins (arrow) in the pancreatic body; B: On axial T2-weighted half-Fourier single-shot turbo spin echo (HASTE) image (TR/TE, ∞/90), the tumor appears slightly hyperintense (arrow) compared to adjacent pancreatic parenchyma; C: At ADC quantification the tumor had a relatively high mean ADC value; D: The macroscopic pathological specimen (distal pancreatectomy, sagittal cut) shows a small, well-delimitated lesion bulging the contour of the pancreatic body.

Open in New Tab   Full Size Figure   Download Figure

Figure 2 Non-hyperfunctioning pancreatic neuroendocrine neoplasm of the pancreatic head classified as G2, stage 3b tumor at histology (Ki67 15%, T4, N1, M0) in a 55 years-old man. A: Axial volumetric interpolated breath-hold examination (VIBE) gradient echo image (repetition time msec/echo time msec, 4.3/1.4) with fat suppression shows a hypointense tumor with ill-defined margins (arrow); B: On coronal fat-saturated T1-weighted volumetric interpolated breath-hold examination (VIBE) gradient echo image (TR/TE 3.5/1.3 ms) acquired during the delayed phase of the dynamic study the tumor shows heterogeneous contrast enhancement, with infiltration of the superior mesenteric vein (arrow); C: The tumor presents intermediate mean ADC value; D: The surgical specimen (pancreaticoduodenectomy with en bloc resection of the superior mesenteric vein, transverse cut) shows a large lesion of the pancreatic head with irregular margins infiltrating the superior mesenteric vein (arrow).

Most lesions were hypointense on T1- and T1fs-weighted images, hyper- or isointense on T2w images and hyperintense on fsT2w images. On post-contrast images, most tumors were hyperenhancing on late arterial/pancreatic and portal phase images; most tumors of this series were iso- or hyperintense on delayed phase images. No significant differences between tumor grades and stages were found regarding the signal intensity on both unenhanced and post-contrast images. Intratumoral fluid areas were more frequently found within G2-3 PanNENs, albeit without significant differences between tumor grades and stages (P = 0.279 and 0.464). Heterogeneous enhancement was significantly more common among G2-3 tumors as compared to G1 lesions (P = 0.006); nevertheless, lesions with heterogeneous enhancement were significantly larger than those presenting homogeneous enhancement, independently from tumor grade (39.3 ± 23.5 mm vs 19.8 ± 12.8 mm, P < 0.001).

Details on the results of the qualitative image analysis and the comparison between grades and ENETS TNM stages are reported in Table 4. The lesions had a mean diameter of 28.7 ± 20.8 mm; G2-3 tumors were significantly larger than G1 PanNENs (P = 0.017), while no significant differences were found between tumor stages in terms of size (P = 0.080). The mean caliber of the MPD/CBD was 6 ± 2 and 13.4 ± 3.9 mm, respectively; no significant differences between tumor grades were found regarding the extent of MPD/CBD dilation (P = 0.256 and 0.151, respectively), while the MPD/CBD diameter was significantly larger in stage I-II PanNENs as compared to stage III-IV tumors (P = 0.038 and 0.034, respectively). G2-3 and stage III-IV PanNENs presented significantly lower mean ADC values compared to G1 and stage I-II tumors (1.09 ± 0.28 × 10^-3 mm²/s vs 1.45 ± 0.53 × 10^-3 mm²/s and 1.10 ± 0.25 × 10^-3 mm²/s vs 1.53 ± 0.55 × 10^-3 mm²/s; P = 0.003 and 0.001, respectively; Figure 3). ADC values were not correlated to tumor size (P = 0.236).

Open in New Tab   Full Size Figure   Download Figure

Figure 3 Non-hyperfunctioning pancreatic neuroendocrine neoplasm of the pancreatic head classified as G3, stage 3b tumor at histology (Ki67 60%, T4, N1, M0) in a 76 years-old woman. A: Axial T2-weighted half-Fourier single-shot turbo spin echo (HASTE) image (TR/TE, ∞/90) shows a large, slightly hyperintense tumor in the pancreatic head (arrow); B: On coronal fat-saturated T1-weighted volumetric interpolated breath-hold examination (VIBE) gradient echo image (TR/TE 3.5/1.3 ms) acquired during the delayed phase of the dynamic study the tumor shows heterogeneous contrast enhancement (arrow); C: At quantitative analysis of ADC map the tumor shows low mean ADC value; D: The surgical specimen (pancreaticoduodenectomy, sagittal cut) shows a large lesion of the pancreatic head with heterogeneous aspect.

The diagnostic performance of MR imaging parameters in identifying G2-3 and stage III-IV PanNENs is presented in Table 5. Among qualitative features, vascular invasion and liver metastases had the highest specificity in identifying G2-3 tumors (96.7%, 95%CI: 81.5-99.8). Ill-defined margins had high specificity for the identification of G2-3 and stage III-IV PanNENs (90.3% and 96%, 95%CI: 73.1-97.5 and 77.7-99.8, respectively; Figure 2). The ROC analysis of lesions’ size provided an area under the curve (AUC) for the identification of G2-3 tumors of 0.757; the best cut-off value was found to be 17.5 mm, which provided a sensitivity of 91.7% (95%CI: 71.5-98.5) and a specificity of 61.3 (95%CI: 42.3-77.6).

By excluding 8 hyperfunctioning tumors and increasing the size threshold to 20 mm, a previously identified cutoff for the choice between surgical resection and follow-up[5], the AUC was found to be 0.736, sensitivity dropped to 79.2% (95%CI: 57.3-92.1), while specificity remained substantially unchanged (60.9%, 95%CI: 38.8-79.5). The ROC analysis of ADC values provided an AUC value of 0.740 for the identification of G2-3 tumors; the best cut-off value for the identification of G2-3 tumors was found to be 1.21 × 10^-3 mm²/s, that yielded a sensitivity of 70.8% (95%CI: 48.7-86.6) and a specificity of 64.5% (95%CI: 45.4-80.2). The ROC analysis of ADC values for the identification of stage III-IV tumors provided an AUC of 0.773; the best cut-off value for the identification of stage III-IV tumors was found to be 1.28 × 10^-3 mm²/s (sensitivity 80.7%, 95%CI: 60-92.7; specificity 64%, 95%CI: 42.6-81.3).

Details regarding the diagnostic accuracy of different combinations of MR features in diagnosing G2-3 tumors are reported in Table 6. The highest sensitivity and specificity were reached using at least two of these five criteria in combination (79.2%, 95%CI: 57.3-92.1; and 80.6%, 95%CI: 61.9-91.9); in particular, the highest specificity values (100%) in diagnosing G2-3 tumors were reached using a combination of ill-defined margins and vascular involvement and/or liver metastases and ill-defined margins.

Pretreatment prediction of the biological behavior of pancreatic neuroendocrine neoplasms (PanNENs) is crucial to determine an efficient treatment strategy for these tumors, especially when unresectable. Nevertheless, even invasive methods, as fine-needle aspiration (FNA), have a limited accuracy in determining the true proliferative activity of these tumors. Imaging methods could estimate the malignancy of PanNENs in a non-invasive way. Nevertheless, these results are based on small study populations and the diagnostic accuracy of imaging features in predicting the biological behavior of these tumors is still undefined.

The aims of this study are to describe magnetic resonance (MR) and diffusion-weighted imaging features of PanNENs according to the 2010 World Health Organization grade classification and European Neuroendocrine Tumor Society tumor-nodes-metastases (TNM) staging system by comparing them to histopathology and to determine the accuracy of MR imaging features in predicting the biological behavior of these tumors.

One of the major strenghts of our study is that histopathological analysis was based on surgical specimens, while in most previous study FNA was considered as the external gold standard. This study demonstrated that MR imaging features of PanNENs vary according to the grade of differentiation and the European Neuroendocrine Society TNM stage and can predict the biological behavior of these tumors. In particular, besides other well-known features as liver metastases and vascular infiltration, irregular margins and low apparent diffusion coefficient (ADC) values are good predictors of less-differentiated tumors and of those with higher biological malignancy. MR features may be helpful, in addiction to tissue sampling, to define the most appropriate treatment strategy for PanNENs.

MR features, especially the mean ADC value calculated by the placement of regions of interest within the tumor, may be helpful, in addiction to tissue sampling, to define the most appropriate treatment strategy for PanNENs.

The paper is well written. The idea of using MR imaging to correlate with the grade/stage of tumors is not novel but in the context of PanNen may provide useful information.