RPS 2401 - Pancreas cancers and high-risk pancreatic lesions

Author Block: A. I. Baeva, A. Ustalov, S. A. Shmeleva, I. Gruzdev, V. Struchkov, E. V. Kondratyev, A. Pokshubina; Moscow/RU
Purpose: The five-year survival rate for patients with solid pseudopapillary epithelial neoplasm (SPEN) after surgery exceeds 98%, whereas pancreatic ductal carcinoma (PDAC) is highly aggressive, characterized by invasive growth and early metastasis. Differential diagnosis of small SPEN and PDAC is challenging. The aim of the present study was to develop and evaluate a ML-radiomic model for distinguishing SPEN from PDAC to improve diagnosis in pancreatic cancer.
Methods or Background: A cohort of 97 patients was retrospectively analyzed, including 39 with SPEN under 30 mm in diameter without local invasive growth and 58 with PDAC, all undergoing contrast-enhanced CT scans and histopathological examination of tumor tissues between 2022 and 2025. Tumor segmentation was manually performed on arterial and venous images using a 3D ROI. Radiomic features were extracted with the "Radiomics" extension in 3D-Slicer. Seven machine learning (ML) models were developed based on these features, and their performance was evaluated using ROC AUC to identify the best classifier.
Results or Findings: The AutoML_Stacking model achieved the highest ROC-AUC of 0.971, demonstrating excellent discriminatory ability. Similar results were obtained for the LGBM and LR models (ROC-AUC of 0.965). All models had ROC-AUC values between 0.920 and 0.971, indicating robustness of the radiomic features. Sensitivity ranged from 0.972 to 1.0, reflecting good SPEN detection.
Conclusion: ML-radiomics diagnostic models for distinguishing SPEN from PDAC were developed. The AutoML_Stacking and LightGBM models exhibit high diagnostic efficiency, highlighting the potential of radiomics for differentiating small SPEN from PDAC, aiding early diagnosis in pancreatic cancer.
Limitations: Not applicable.
Funding for this study: No funding was provided for this study.
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information:

Author Block: Z. N. Doghramachi, T. Cankurtaran, D. KALEMCİ, M. ALIYEV, Z. N. DAYICAN, O. Kayık, B. Yavuz Sarsam, A. Isik; Ankara/TR
Purpose: To investigate the independent effects of fat infiltration (Proton Density Fat Fraction, PDFF) and iron overload (R2∗) on quantitative diffusion-weighted imaging (DWI) metrics in the liver and pancreas, using the spleen as a parenchymal reference tissue.
Methods or Background: This cross-sectional study included 102 adults (55.8 ± 11.3 years). Abdominal quantitative MRI comprised multi-echo gradient-echo acquisitions for PDFF and R2∗ mapping, and multi-b-value DWI (b=0–2500 s/mm2) for diffusion analysis (monoexponential ADC, logarithmic ADC, and biexponential fast/slow components). Statistical analyses included partial correlation (adjusted for R2∗) and multivariable linear regression (adjusted for age, sex, and BMI) to assess associations between PDFF, R2∗, and diffusion metrics in the liver and pancreas, as well as their ratio to spleen values.
Results or Findings: In the liver, Fat Fraction (FF) showed positive associations with the biexponential fast component (r=0.280, p=0.004) and the ratio of fast/slow components (r=0.293, p=0.003). No independent associations were found for monoexponential or logarithmic ADC. In the pancreas, R2∗ was independently and positively related to monoexponential ADC (β=0.380, p<0.001) and negatively to logarithmic ADC (β=−0.293, p=0.003). The association between pancreatic FF and logarithmic ADC lost significance after R2∗ adjustment. Ratio-based analysis (Pancreas/Spleen) further demonstrated an independent effect of pancreatic FF on the biexponential slow ADC component.
Conclusion: Hepatopancreatic fat infiltration (PDFF) and R2∗ showed independent, albeit weak, associations with specific DWI metrics. These findings suggest that the effects of both fat and iron need to be considered when interpreting quantitative diffusion measurements in the liver and pancreas.
Limitations: This study was limited by the relatively small sample size, the absence of participants with high-grade steatosis, and the lack of voxel by voxel analysis, as all measurements were ROI-based.
Funding for this study: Funding was provided by Siemens Healthineers, Erlangen, Germany:
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: The study was approved by [Baskent University]

Author Block: F. Prato, D. Palumbo, G. Orsi, M. Macchini, C. Carconi, A. Chiti, M. Falconi, M. Reni, F. De Cobelli; Milan/IT
Purpose: Neoadjuvant chemotherapy (NAT) is becoming the treatment of choice for pancreatic ductal adenocarcinoma (PDAC). In this scenario, the traditional classification of borderline-resectable and anatomically resectable disease—based mainly on CT—appears progressively outdated, also due to inter-observer variability. Although upfront surgery is rare, accurate baseline staging remains crucial to define prognosis and guide therapy, while avoiding futile surgery in patients with occult systemic disease. Advanced imaging such as FDG-PET and MRI with liver-specific contrast media (EOB-MRI) may provide incremental value over CT.
Methods or Background: In this prospective study, patients with a resectable or borderline resectable (according to CT scan), were screened using both FDG-PET and 3T-EOB-MRI before the initiation of NAT.
Results or Findings: Of 154 screened patients, 124 were enrolled (upfront resectable = 71, borderline = 53); one patient was later reclassified as locally advanced. Additional imaging excluded 24 patients (18.8%) due to previously undetected metastases, while 5 (3.3%) were excluded for protocol deviations. After a median follow-up of 20 months, 18/154 (11.6%) were considered true positives (“screening failures”), while 117/154 (76%) were true negatives (correctly classified as resectable/borderline). False negatives occurred in 8/154 cases (5.2%) and false positives in 6/154 (3.9%), mostly related to PET overestimation. Retrospective analysis suggested that MRI alone, particularly when interpreted by expert radiologists, could further reduce both false positives and negatives. In upfront resectable patients, predictors of occult metastases (on baseline CT scan) included tumor necrosis (p=0.009), dilated Wirsung duct (p=0.04), and CA19.9 ≥200 U/mL (p<0.001).
Conclusion: EOB-MRI proved more effective than FDG-PET in identifying occult metastases, especially in the liver and peritoneum, thereby reducing futile surgeries. These findings support prioritizing EOB-MRI in baseline staging protocols for PDAC, particularly in high-risk patients, to optimize therapeutic strategies and patient selection.
Limitations: Monocentric
Funding for this study: Patient Associations
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: Ethics Committee of Ospedale San Raffaele (Milan, IT) (06/INT/2022)

Author Block: Q. DAI, M. Lu; Chengdu, China/CN
Purpose: To compare the diagnostic performance of pancreatic lesions using percutaneous ultrasound (US)-guided core needle biopsy (CNB) with and without contrast-enhanced ultrasound (CEUS).
Methods or Background: The patients were divided into two groups, US and CEUS group, based on whether CEUS was performed prior to biopsy. Before and after propensity score matching (PSM), the CNB-relevant characteristics of the two groups, including the first puncture success rate, the number of sampling, complication rate, type of complications, and degree of abdominal pain, were compared. The accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and area under the curve (AUC) of percutaneous US-guided CNB were compared between the groups.
Results or Findings: This study included 277 patients with pancreatic lesions who underwent percutaneous CNB before PSM. After controlling for potential biases using PSM, significant differences were observed in the first puncture success rate and the number of sampling (P<0.05). Importantly, before and after PSM, the CEUS group achieved a higher first-puncture success rate while obtaining a lower number of sampling (P<0.05). Furthermore, the significant difference was observed in the AUC for diagnostic performance between the two groups when compared using DeLong’s test (P=0.043).
Conclusion: Performing CEUS before percutaneous CNB for pancreatic lesions can help achieve better biopsy results, reduce the number of punctures samples, increase the success rate of biopsies, and avoid the need for repeat biopsies.
Limitations: First, the sample size was derived from a single center, and the percutaneous CNB technique for pancreatic lesions requires radiologists with extensive experience in US intervention, which may have affected the generalizability of our findings.
Second, as a retrospective study, there remains a potential for selection bias despite conducting a PSM analysis.
Funding for this study: No.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: Ethics Committee of Sichuan Cancer Hospital (approval number: SCCHEC-02-2024-127)

Author Block: V. Koch¹, L. D. Grünewald¹, K. Eichler¹, R. Hammerstingl¹, O. Darwish², J. Gotta¹, T. Vogl¹, R. Sinkus²; ¹Frankfurt/DE, ²Paris/FR
Purpose: The aim of this study was to evaluate magnetic resonance elastography (MRE) using the acoustic (2D) and gravitational transducer (2D/3D) concept for detecting pancreatic cancer and assessing its potential as a marker of chemotherapy response.
Methods or Background: In this prospective study, 51 patients with confirmed pancreatic cancer undergoing upper abdominal MRI were enrolled between June 2023 and April 2024. An additional 15 healthy volunteers were included as a reference standard. All participants underwent two examinations at 40 Hz (Aera, 1.5 T, Siemens Healthineers, Germany): first with a commercial MRE system (Resoundant, 2D-MRE, SE-EPI sequence, 11 s breath-hold), and subsequently with the gravitational transducer approach (2D- and 3D-MRE, GRE sequence, TE = 9.2 ms [in-phase], fractional motion encoding at 30 mT/m, 14 s breath-hold). Data were independently analyzed by three experienced readers. To assess potential biases in 2D/3D stiffness reconstruction, analytic plane waves with known wavelength and absorption were superimposed at varying amplitudes and temporal noise levels, and used to derive quality indices for 2D and 3D MRE.
Results or Findings: Significant differences in stiffness, shear wave speed, and phase angle were observed between healthy volunteers and pancreatic cancer patients (p < .05). Subtype-specific signatures enabled differentiation between ductal adenocarcinoma and neuroendocrine tumors (p < .05). Stiffness changes tracked chemotherapy response (p < .05), while higher baseline values correlated with longer hospitalization. 2D-MRE systematically overestimated stiffness, whereas proposed 3D-MRE Quality Indices improved reliability.
Conclusion: Gravitational MRE enables accurate, noninvasive characterization of pancreatic tissue. Both approaches correlated well with the commercial standard, while 3D-MRE uniquely delivered functional imaging markers, opening new avenues in abdominal imaging.
Limitations: Limited study population. Prospective validation in larger patient cohorts needed.
Funding for this study: No funding.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: Approved by the institutional ethics committee.

Author Block: Y. Wang, Y. Li, H. Yu, Z. Huang, B. Song; Chengdu/CN
Purpose: To compare the prognostic performance of pathological (pN) and clinical N staging (cN) with Node-RADS for recurrence-free survival (RFS) in patients with pancreatic ductal adenocarcinoma (PDAC) following pancreatoduodenectomy.
Methods or Background: This retrospective study included patients with pathologically confirmed PDAC underwent pancreatoduodenectomy between June 2018 and January 2023 and had preoperative contrast-enhanced abdominal CT. Two observers independently assessed lymph nodes ≥0.4 cm in short-axis diameter on venous-phase images using conventional CT criteria and Node-RADS, with a 4-week interval. Node-RADS scores ≥3 were defined as positive. For each patient, total and station-based Node-RADS scores were calculated. Prognostic performance was evaluated using time-dependent ROC curves and Cox regression. Interobserver agreement was assessed with intraclass correlation coefficients (ICC).
Results or Findings: 220 patients (median age, 63 years; 138 men) were analyzed. Median RFS was 10 months. The AUCs of pN for 12-, 24-, and 36-month prediction were 0.546, 0.560, and 0.581, while cN yielded 0.541, 0.501, and 0.538. Node-RADS total scores showed AUCs of 0.581, 0.654, and 0.605. The number of positive stations achieved higher performance (0.611, 0.717, 0.695). Station 16a2 scores showed prognostic value (AUCs 0.533, 0.653, 0.631). Node-RADS total score (p=0.0368), positive station number (p=0.00163), and stations 16a2 (p=0.0312) and 17 (p<0.001) as adverse prognostic factors. Interobserver agreement was moderate (ICC = 0.528).
Conclusion: Node-RADS outperformed pN and cN in predicting RFS after pancreatoduodenectomy. Positive station number and stations 16a2 and 17 provided additional prognostic value, supporting Node-RADS as a practical tool for preoperative risk stratification in PDAC.
Limitations: This retrospective single-center study lacks external validation, which may restrict generalizability. Additionally, CT images obtained over four years may have been affected by inter-scanner and technical variability, potentially influencing Node-RADS assessment.
Funding for this study: This study received no external funding.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: West China Hospital, Sichuan University - approval number 2025 − 1152

Author Block: C. Wei, Y. Zhu, X. Zhao; Beijing/CN
Purpose: To investigate the value of habitat imaging employing baseline dual-layer spectral CT (DLCT) in predicting recurrence of pancreatic ductal adenocarcinoma (PDAC) after radical resection, and explore the relationship with pathological tumor-stroma ratio (TSR).
Methods or Background: we retrospectively enrolled 136 resected PDAC patients who underwent multiphase DLCT and randomly assigned them to training and validation cohorts. Extracellular volume (ECV) fraction and arterial enhancement fraction (AEF) were generated from spectral images. Tumor voxels were clustered into habitats via K-means, and quantitative habitat parameters were extracted. Cox regression was used to build clinical-radiological, habitat, and combined models for recurrence-free survival (RFS). Model performance was assessed by concordance index (c-index) and time-dependent ROC, and recurrence risk was compared by Kaplan–Meier analysis. Spearman correlation and multivariable linear regression evaluated the relationship between TSR and habitat metrics.
Results or Findings: The combined model integrating fraction of habitat-1 (f1), fraction of habitat-4 (f4), CA19-9 >180 U/mL, and rim-enhancement achieved c-indices of 0.912 and 0.899 in training and validation cohorts, respectively, outperforming the clinical-radiological model. Time-dependent ROC showed AUCs >0.85 for RFS prediction. Patients predicted as high-risk had significantly shorter RFS than low-risk (p < 0.001). f1 and f4 correlated with TSR and enabled quantitative prediction of TSR.
Conclusion: The combined model, integrating habitat quantitative parameters, CA19-9 and rim-enhancement, providing a noninvasive approach for predicting the risk of recurrence in PDAC preoperatively. Habitat quantitative parameters could be used to quantitative predict pathological TSR.
Limitations: This retrospective design might introduce selection bias, as the analysis was limited to resectable PDAC, which only accounts for a small part of all PDAC. Furthermore, although a validation cohort was used, external validation from multi-center with larger sample size was still needed to further confirm the stability of the results.
Funding for this study: None
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: This retrospective study was conducted in accordance with the Declaration of Helsinki. The study protocol was approved by the Institutional Review Board of the National Cancer Center/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College (Approval #21/241-4521; November 10, 2021).

Author Block: D. Palumbo, F. Prato, G. Belfiori, P. CAMISA, B. Ferrara, L. Piemonti, M. Falconi, S. Crippa, F. De Cobelli; Milan/IT
Purpose: Intraductal papillary mucinous neoplasms (IPMNs) are established precursors of pancreatic ductal adenocarcinoma (PDAC). Current guidelines recommend surgery based on high-risk stigmata (HRS) and worrisome features (WF), yet these criteria lack accuracy and no reliable non-invasive biomarker of malignancy is available. We aimed to identify preoperative radiological biomarkers to distinguish low-grade from high-grade/invasive IPMNs.
Methods or Background: Between June 2020 and January 2025, 206 patients undergoing surgical resection for suspected pancreatic cystic neoplasms were prospectively enrolled. Final pathology showed 49 low-grade dysplasias (LGD), 42 high-grade dysplasias (HGD), and 40 invasive carcinomas (IPMC). For the present analysis, IPMN were stratified into low-risk (LGD) and high-risk (HGD + IPMC) groups. Preoperative imaging included CE-MRI with dedicated MRCP incorporating mDixon Quant Fat Fraction sequences and contrast-enhanced CT. Inter-modality agreement between CT and MRI for WF/HRS was evaluated. A composite risk score was constructed by summing WF and HRS, weighted according to hazard ratios reported in the literature. Biological samples (peripheral/portal blood, cyst fluid) were systematically collected intraoperatively and analyzed for inflammatory and metabolic markers.
Results or Findings: Pancreatic fat fraction (assessed on non-atrophic parenchima) was significantly higher in high-risk vs low-risk IPMN (15.7% ± 13.8 vs 6.9% ± 5.0, p=0.023), but showed no correlation with molecular variables. Distal pancreatic atrophy showed a trend toward significance (p=0.062). Individual WF/HRS performed poorly, with wide inter-modality variability (K-Cohen 0.02 to 0.87). Conversely, their additive integration into a risk score significantly predicted high-risk IPMN and correlated with inflammatory markers on cyst-fluid samples, particularly TNF-α (p=0.005).
Conclusion: Pancreatic fat fraction on MRI emerges as a promising non-invasive biomarker for stratifying IPMN. WF/HRS are inconsistent as isolated criteria, but their cumulative integration into a risk score enhances predictive accuracy and correlates with local inflammatory mediators.
Limitations: Monocentric
Funding for this study: IRCCS Ospedale San Raffaele
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: Ethics Committee of Ospedale San Raffaele (Milan, IT) (06/INT/2022)