This study was approved by our institutional review board, and patient informed consent was waived considering the retrospective nature of the study. From January 2015 to December 2020, 19 patients with pathologically confirmed pancreatic MiNEN were identified by searching medical database in our institution. Among these patients, 12 patients (male 8, female 4; mean age ± standard deviation [SD], 55.8 ± 11.1 years; range 37–73 years) were eventually enrolled in this study based on the following inclusion criteria: (1) CT and MR images included contrast enhancement (arterial phase and portal venous phase) before surgery, (2) no history of major abdominal surgery, and (3) primary pancreatic tumor. To form a 1:2 matching with the pancreatic MiNEN group, we selected 35 PDAC and 31 NET based on the above criteria. A total of 24 patients with PDAC (male 16, female 8; mean age ± SD, 58.9 ± 13.8 years; range 36–83 years) and 24 patients with pancreatic NET (male 16, female 8; mean age ± SD, 50.6 ± 14.5 years; range 31–72 years) were later selected based on gender, age, location, and treatment by using the Matchlt package of the R software (version 3.4.4, R Core Team 2017). All lesions were pathologically proven by pancreatectomy upon resection (n = 60).
CT and MRI examination
CT examinations were performed on 16 -or 64-multidetector CT scanners (Lightspeed 16 or Lightspeed 64; GE Medical Systems, Milwaukee, WI, USA) by using a dual-phase scan (arterial and portal phases). Arterial and portal phase scans were obtained with a delay of 15–20 s by using a bolus-tracking technique and a delay of 60–70 s after injection of 2.5–3 mL/kg of non-ionic iodinated contrast medium intravenously (Omnipaque 300 mg/mL, GE Healthcare) at a rate of 2.5–3.5 mL/s by using a power injector. The detector collimation values of GE 16- and 64-multidetector scanners were 0.75 and 0.6 mm, respectively, the pitch was 1.5, the rotation time was 0.6 s, the tube voltage was 120 kV, the tube current was 250 mA, the reconstructed slice thickness was 4.0 mm, and the reconstruction interval was 4.0 mm.
MRI examinations were performed using a 1.5-T MRI scanner (GE Signa HD, GE Healthcare Systems, Milwaukee, WI, USA) with 8- and 12-channel phased-array torso coil. The applied image sequences with scan parameters were shown as follows: repetition time (TR)/echo time (TE) of 520/14 ms, section thickness of 3 mm, field of view (FOV) of 400 × 280 mm2, and matrix of 256 × 256 for the fast-spin-echo-based pre-contrast T1-weighted images with fat suppression; (TR/TE) of 3,500/95 ms, section thickness of 3 mm, FOV of 400 × 280 mm2, and matrix of 320 × 256 for the respiratory-triggered fast-spin echo-based T2-weighted images with fat suppression; TR/TE of 4.9/1.0 ms, section thickness of 3 mm, FOV of 380 × 380 mm2, and matrix of 320 × 288 for the 2D-fast imaging employing steady-state acquisition (2D-FIESTA) in axial and coronal views; TR/TE of 3,600/70 ms, section thickness of 5 mm, FOV of 360 × 360 mm2, and matrix of 128 × 128 for the free-breathing single-shot echo-planar diffusion weighted image with a and b values of 0 and 800 s/mm2, respectively; and TR/TE of 4.1/1.5 ms, section thickness of 3 mm, FOV of 350 × 280 mm2, and matrix of 320 × 256 for the contrast-enhanced T1-weighted images with contrast-enhanced phases including arterial and portal phases. These parameters were applied at 15 s (arterial phase), 60 s (portal phase) after injection of Gadodiamide intravenously (Omniscan, GE; 0.1 mmol/kg body weight) at a rate of 1.5 mL/s by using an autoinjector.
CT and MRI images were independently reviewed by two senior radiologists with 9 and 10 years of experience in abdominal imaging on the hospital image archiving and communication system. Interobserver agreement for imaging features was assessed after initial image analysis. Discrepancies between the two readers were resolved by a consensus after joint image re-evaluation. The paired CT and MRI data of 60 cases were randomly reviewed by two independent readers twice and the time interval between two assessments for each reader was least 1 month. Both readers were blinded to all clinical and pathological data.
The following lesion relevant characteristics were evaluated: largest transverse diameter of the tumor, tumor location, tumor contours and margins, heterogeneity, tumor calcification, presence of visible lymph nodes (short axis diameter > 10 mm), bile duct dilatation, presence of main pancreatic duct (MPD) enlargement (diameter > 3 mm), adjacent organ involvement, and vascular encasement by tumor. According to the internal components, tumors were classified as purely solid lesion, solid lesion with minor cystic components (cystic component < 10% of the tumor), mixed solid and cystic lesion (cystic component > 10% of the tumor) . Areas with density and/or intensity similar to that of cerebrospinal fluid on both CT and MR images were considered as cystic components of tumor.
The enhancement characteristics on the arterial (pancreatic) phase include diffuse and rim type of increased enhancement. The enhancement characteristics on the portal venous phase include washout and progressive enhancement. The degree of CT enhancement was divided into no enhancement (+ 0–10 HU), mild (+ 10–20 HU), moderate (+ 20–50 HU), or marked (+ > 50 HU) enhancement compared with pre-contrast phase .
Available records of clinical and pathological data were retrieved for each patient. The following clinical data were extracted from medical records: sex, age, reasons for admission (abdominal pain or discomfort or jaundice), and surgical options. Tests for the serum tumor markers, including serum carcinoembryonic antigen (CEA), serum amylase, and serum carbohydrate antigen 19-9 (CA 19-9), were available in the clinical information system. These serological exams were mainly performed within 30 days before or after CT and MRI.
All sections were retrospectively reviewed by a pathologist with 16 years of experience, who was blinded to the imaging manifestations. The pathological diagnosis was based on hematoxylin–eosin and immunohistochemical staining results.
Descriptive statistical values were calculated for all variables evaluated on CT and MRI. One-way ANOVA or Student’s t test was used for continuous variables, and chi-square test or Fisher’s exact test was used for categorical variables. These variables were then entered into univariate analysis with a conditional logistic regression model to determine the independent predictors of pancreatic MiNEN on CT and MRI. Multivariable logistic regression analyses were performed by adverse selection of significant variables in the univariable analyses combined with clinical significance, and intermediate factors were excluded. The sensitivity, specificity, accuracy, positive predictive value, negative predictive value (NPV), positive likelihood ratio (LR (+)), and negative likelihood ratio (LR (−)) of each significant imaging feature and combinations of these features were calculated.
Interobserver agreement was performed for each variable by using kappa (k) statistics with the following scale: poor, < 0.20; fair, 0.20–0.39; moderate, 0.40–0.59; substantial, 0.60–0.79; and almost perfect, 0.80–1.00 . All statistical analyses were performed using SPSS Statistics 26.0 and R software (version 3.4.4, R Core Team 2017). Two-sided p values of < 0.05 were considered statistically significant.