The institutional review board approved this prospective single-centre cohort study and written informed consent was obtained. The study was registered on clinicaltrials.gov (NCT04311047) and conducted in accordance with the Declaration of Helsinki and the Dutch Medical Research Involving Human Subjects Act. Treatment-naïve patients with pancreatic, duodenal, or periampullary adenocarcinoma scheduled for surgical resection at the Radboudumc (tertiary hepatobiliary centre) between May 2017 and December 2020 were consecutively included. Eligibility was determined at the multidisciplinary pancreaticobiliary cancer meeting. The periampullary region was defined as the distal common bile duct, ampulla of Vater, and 2 cm of the duodenum surrounding the ampulla. Exclusion criteria were neoadjuvant treatment, distant metastases on preoperative imaging, other concomitant malignancies (prior malignancies: at least 5 years disease-free), or contraindications for MRI or USPIO (allergy, hemochromatosis, thalassemia, or sickle cell anaemia).

Preoperative USPIO-enhanced MRI was conducted using ferumoxtran-10 (Ferrotran, SPL Medical B.V., investigational product), administered intravenously (2.6 mg/kg body weight) 24–36 h before the scan, as previously described [16]. MRI was performed on a 3-T MRI system (Magnetom Prisma, Siemens Healthcare). Butylscopolamine and glucagon were administered to minimise peristaltic motion, unless contraindicated. Patients were positioned feet first supine, with body phased array coils around the upper abdomen. The scan range was from diaphragm to aortic bifurcation. Images were acquired with repeated breath-holds of maximally 20 s on expiration. MRI sequences included T2-weighted HASTE for anatomical reference, T1-weighted VIBE DIXON for LN localisation, and fat-suppressed T2*-weighted multi gradient echo (mGRE) sequence for USPIO visualisation, mGRE images were reconstructed to a single T2*-weighted computed echo time (TE) of 12 ms [17]. Table 1 describes the technical details of the MR sequences.

A radiologist (A.N., 18 years experience in abdominal imaging) analysed all USPIO-enhanced MRI images. A second radiologist (J.H., 23 years experience in pancreatic imaging) supervised the analysis. Disagreements were resolved in a consensus meeting. Both radiologists were blinded to the histopathology results. All visible LNs, regional and distant, were annotated and measured (short axis on axial orientation). LN locations were indicated using the classification of the Japan Pancreas Society [18]. LNs were scored on the iron-sensitive T2*-weighted mGRE (TE = 12 ms) sequence, using diagnostic guidelines adapted from Anzai et al (Table 2, [19]). Type 1–4 LNs were classified as suspicious for metastases and type 5–7 LNs as non-suspicious. The distinction between regional and distant LNs was based on the TNM classification (eighth edition) by the UICC [5]. The definition of regional nodes differs depending on the cancer type (pancreatic head or tail, cholangial, ampullary, or duodenal), the respective definition for each tumour type was applied.

Operative procedures, performed by experienced surgeons in pairs (mean experience nine years), contained pylorus resecting pancreatoduodenectomy (PRPD) with lymphadenectomy (stations no. 5, 6, 8a, 12b1/2, 12c, 13a/b, 14a/b, and 17a/b) or distal pancreatectomy (DP) with lymphadenectomy (station no. 10, 11, and 18) as indicated by the International Study Group on Pancreatic Surgery (ISGPS) [20]. Additional para-aortic or other distant LNs were resected if indicated as suspicious on MRI and resection was technically possible and safe. Resection specimens were pinned on an anatomical drawing for orientation.

A high-resolution ex vivo MRI of the fresh resection specimen and separately resected LNs was performed on a 7-T preclinical MRI system (Clinscan, Bruker® BioSpin). Ex vivo images had a resolution of 0.29 × 0.29 × 0.29 mm3 and consisted of a water and lipid-excited T1-weighted image according to a predefined workflow [16]. Subsequently, the tumour was fixated in formaldehyde for at least 48 h and then sliced into 4-mm slices following the axial slicing technique [21]. LNs from the peripancreatic fat were removed and enclosed separately, LNs larger than 5 mm were sectioned in 2–3 mm thick sections for paraffin embedding. Coupes were 4 µm with haematoxylin and eosin (H&E) staining. In contrast to measurement on MRI (short axis on axial orientation), at pathology, the largest diameter of the LNs was measured, as is clinical routine.

Aided by the ex vivo MRI, LNs from in vivo MRI were matched to pathology for node-to-node analysis. Clinical follow-up imaging was used to confirm if an LN was indeed resected or not. Matching criteria were size, shape, and location, including anatomical landmarks such as surrounding vessels and organs. Only matched LNs were included in the validation analysis. A pancreas pathologist (L.B., ten years of expertise) re-evaluated all matched LNs.

Statistical analysis was performed using SPSS (version 25). The sensitivity and specificity of USPIO-MRI were calculated with crosstabs, using pathology as the reference standard. Continuous variables were summarised using standard descriptive statistics (mean, standard deviation, median, and range), and categorical variables were summarised with frequencies.

After prospective analysis, a third radiologist (P.Z., eight years experience with USPIO-enhanced MRI) reviewed all false positive and false negative LNs to investigate potential reasons for misclassification.