Gonda, Tamas A.

Abstract Despite growing interest in AI for pancreatic cyst management, no prior study has systematically investigated how lesion characteristics influence model behavior or provided a comprehensive explainability analysis in IPMN malignancy risk stratification. We present the first multi-center study integrating explainable AI with uncertainty quantification to evaluate how cyst type, size, and location affect deep learning performance in IPMN malignancy prediction. Our retrospective study analyzed 170 IPMNs from seven centers using a radiomics-deep learning fusion model. Cases were stratified by dysplasia grade, IPMN type, size, and location; with model interpretability assessed using SHAP, LIME, Grad-CAM, and uncertainty quantification. Overall accuracy was 67.1%, with model uncertainty lower for correct versus incorrect predictions (0.72 vs 0.78, p < 0.001). On subgroup analysis: Low-Risk lesion accuracy exceeded High-Risk (81.3% vs. 42.9%, p < 0.001), BD accuracy was inversely correlated with cyst size while MD remained stable, and body and whole-pancreas cysts stratified more accurately than tail cysts (77.8%, 76.6% vs. 47.8%). AI models face significant, previously uncharacterized performance drops when evaluating complex high-risk IPMNs. Our study shows that integrating uncertainty quantification can successfully flag unreliable predictions, enabling a selective-prediction framework where clinicians can confidently rely on AI for low-risk, low-uncertainty lesions while deferring high-uncertainty cases to expert review.

Abstract Pancreatic cancer, projected to become the second-deadliest malignancy in Western societies by 2030, requires urgent innovations in early detection and risk stratification. Intraductal papillary mucinous neoplasms (IPMNs) represent critical precursor lesions, but current clinical guidelines demonstrate suboptimal accuracy in malignancy prediction, leading to either unnecessary surgeries or missed opportunities for early intervention. In this paper, we present Cyst-X, an artificial intelligence (AI) framework that accurately predicts IPMN malignant transformation using multicenter magnetic resonance imaging (MRI) data. Unlike most previous approaches that rely on computed tomography (CT), our method capitalizes on the superior soft tissue contrast of MRI, allowing for more precise identification of subtle imaging biomarkers. We developed and validated deep learning models on 723 T1-weighted and 738 T2-weighted MRI scans from 764 patients among seven international institutions, demonstrating significantly superior performance (AUC=0.82) compared to current clinical (Kyoto) guidelines (AUC=0.75) and expert radiologists. AI-based imaging features correlate strongly with clinically recognized malignancy markers, providing potential biologically relevant insights. This approach holds promise to significantly refine clinical decision-making, reduce unnecessary surgeries, and better identify high-risk IPMN patients for timely intervention. Our approach integrates a novel pancreas segmentation network with robust classification models that identify subtle imaging biomarkers associated with malignancy risk. Importantly, we demonstrate that these models retain high performance in a privacy-preserving federated learning setting, where institutions collaboratively train AI models without exchanging patient data to address key regulatory and ethical barriers. We publicly release the Cyst-X dataset--the first large-scale, multi-center pancreatic cyst MRI collection--to accelerate research in this field. This study addresses a critical clinical need while establishing technical foundations for privacy-preserving AI in radiology that could transform pancreatic cancer management through earlier intervention and reduced unnecessary procedures. The dataset can be accessed at https://osf.io/74vfs/, and the source code for our deep learning segmentation and classification models is available at https://github.com/NUBagciLab/Cyst-X.

Abstract Distinguishing high-risk intraductal papillary mucinous neoplasms (IPMNs), pancreatic cysts requiring surgery, from low-risk lesions remains a clinical challenge, often resulting in unnecessary procedures due to limited specificity of current methods. While radiomics and deep learning (DL) have been explored for pancreatic cancer, cyst-level malignancy risk stratification of IPMNs remains untapped. We conducted a multi-institutional study (seven centers, 359 T2W MRI images) to assess the feasibility of AI for predicting IPMN dysplasia grade using cyst-level image features. We developed and compared 2D and 3D radiomics-only, deep learning (DL)-only, and radiomics-DL fusion models, using expert radiologist scoring as a baseline reference. Model performance was evaluated using held-out test data. The radiomics-DL fusion model showed the highest discriminatory ability on the test set (AUC 0.692), outperforming the radiomics-only model (AUC 0.665). Expert accuracy varied widely (37.4%-66.7%). The fusion model integrating deep learning and radiomics features from routine T2W MRI (AUC: 0.692) demonstrates potential for objective, cyst-level risk stratification of IPMNs in a multi-center cohort, outperforming both radiomics-only models and expert radiologists. While performance requires improvement for standalone clinical use, this approach offers a scalable, non-invasive method to potentially improve diagnostic accuracy and reduce unnecessary surgical interventions.

AbstractEUS-guided gallbladder drainage (EUS-GBD) using lumen apposing metal stents (LAMS) has excellent technical and short-term clinical success for acute cholecystitis (AC). The goals of this study were to determine the long-term clinical outcomes and adverse events (AEs) of EUS-GBD with LAMS.A multicenter, retrospective study was conducted at 18 US tertiary care institutions. Inclusion criteria: any AC patient with attempted EUS-GBD with LAMS and minimum 30-day post-procedure follow-up. Long-term clinical success was defined as absence of recurrent acute cholecystitis (RAC) > 30 days and long-term AE was defined as occurring > 30 days from the index procedure.A total of 109 patients were included. Technical success was achieved in 108 of 109 (99.1%) and initial clinical success in 106 of 109 (97.2%). Long-term clinical success was achieved in 98 of 109 (89.9%) over a median follow-up of 140 days (range 30–1188). On multivariable analysis (MVA), acalculous cholecystitis (odds ratio [OR] 15.93, 95% confidence interval [CI] 1.22–208.52, P = 0.04) and the occurrence of a LAMS-specific AE (OR 63.60, 95% CI 5.08–799.29, P <0.01) were associated with RAC. AEs occurred in 38 of 109 patients (34.9%) at any time, and in 10 of 109 (9.17%) > 30 days from the index procedure. Most long-term AEs (7 of 109; 6.42%) were LAMS-specific. No technical or clinical factors were associated with occurrence of AEs. LAMS were removed in 24 of 109 patients (22%). There was no difference in RAC or AEs whether LAMS was removed or not.EUS-GBD with LAMS has a high rate of long-term clinical success and modest AE rates in patients with AC and is a reasonable destination therapy for high-risk surgical candidates.

AbstractBackgroundIdentifying branch‐duct intraductal papillary mucinous neoplasms (BD‐IPMNs) at lowest risk of progression may allow for a reduced intensity of surveillance.ObjectiveWe aimed to externally validate the previously developed Dutch‐American Risk stratification Tool (DART‐1; https://rtools.mayo.edu/DART/), which identifies cysts at low risk of developing worrisome features (WFs) or high‐risk stigmata (HRS).MethodsThree prospective cohorts of individuals under surveillance for BD‐IPMNs were combined, independent from the original development cohort. We assessed the performance (discrimination and calibration) of DART‐1, a multivariable Cox‐proportional logistic regression model with five predictors for the development of WFs or HRS.ResultsOf 832 individuals (mean age 77 years, SD 11.5) under surveillance for a median of 40 months (IQR 44), 163 (20%) developed WFs or HRS. DART‐1's discriminative ability (C‐statistic 0.68) was similar to that in the development cohort (0.64–0.72) and showed moderate calibration. DART‐1 adequately estimated the risk for patients in the middle risk quintile, and slightly underestimated it in the lowest quintiles. Their range of predicted versus observed 3‐year risk was 0%–0% versus 0%–3.7% for Q1; 0.3%–0.4% versus 3%–11% for Q2; and 2.6%–3% versus 2.4%–9.8% for Q3. The development of WFs or HRS was associated with pancreatic cancer (p < 0.001). Vice versa, in absence of WFs or HRS, the risk of malignancy was low (0.3%).ConclusionsThe performance of DART‐1 to predict the development of WFs or HRS in BD‐IPMN was validated in an external international cohort, with a discriminative ability equal as in the development cohort. Risk estimations were most accurate for patients with BD‐IPMNs in the middle risk quintile and slightly underestimated in the lowest quintiles.

SummaryBackgroundBecause most pancreatic intraductal papillary mucinous neoplasms (IPMNs) will never become malignant, currently advocated long‐term surveillance is low‐yield for most individuals.AimTo develop a score chart identifying IPMNs at lowest risk of developing worrisome features or high‐risk stigmata.MethodsWe combined prospectively maintained pancreatic cyst surveillance databases of three academic institutions. Patients were included if they had a presumed side‐branch IPMN, without worrisome features or high‐risk stigmata at baseline (as defined by the 2012 international Fukuoka guidelines), and were followed ≥ 12 months. The endpoint was development of one or more worrisome features or high‐risk stigmata during follow‐up. We created a multivariable prediction model using Cox‐proportional logistic regression analysis and performed an internal‐external validation.Results875 patients were included. After a mean follow‐up of 50 months (range 12‐157), 116 (13%) patients developed worrisome features or high‐risk stigmata. The final model included cyst size (HR 1.12, 95% CI 1.09‐1.15), cyst multifocality (HR 1.49, 95% CI 1.01‐2.18), ever having smoked (HR 1.40, 95% CI 0.95‐2.04), history of acute pancreatitis (HR 2.07, 95% CI 1.21‐3.55), and history of extrapancreatic malignancy (HR 1.34, 95% CI 0.91‐1.97). After validation, the model had good discriminative ability (C‐statistic 0.72 in the Mayo cohort, 0.71 in the Columbia cohort, 0.64 in the Erasmus cohort).ConclusionIn presumed side branch IPMNs without worrisome features or high‐risk stigmata at baseline, the Dutch‐American Risk stratification Tool (DART‐1) successfully identifies pancreatic lesions at low risk of developing worrisome features or high‐risk stigmata.

SUMMARY Volumetric laser endomicroscopy (VLE) uses optical coherence tomography (OCT) for real-time, microscopic cross-sectional imaging. A US-based multi-center registry was constructed to prospectively collect data on patients undergoing upper endoscopy during which a VLE scan was performed. The objective of this registry was to determine usage patterns of VLE in clinical practice and to estimate quantitative and qualitative performance metrics as they are applied to Barrett's esophagus (BE) management. All procedures utilized the NvisionVLE Imaging System (NinePoint Medical, Bedford, MA) which was used by investigators to identify the tissue types present, along with focal areas of concern. Following the VLE procedure, investigators were asked to answer six key questions regarding how VLE impacted each case. Statistical analyses including neoplasia diagnostic yield improvement using VLE was performed. One thousand patients were enrolled across 18 US trial sites from August 2014 through April 2016. In patients with previously diagnosed or suspected BE (894/1000), investigators used VLE and identified areas of concern not seen on white light endoscopy (WLE) in 59% of the procedures. VLE imaging also guided tissue acquisition and treatment in 71% and 54% of procedures, respectively. VLE as an adjunct modality improved the neoplasia diagnostic yield by 55% beyond the standard of care practice. In patients with no prior history of therapy, and without visual findings from other technologies, VLE-guided tissue acquisition increased neoplasia detection over random biopsies by 700%. Registry investigators reported that VLE improved the BE management process when used as an adjunct tissue acquisition and treatment guidance tool. The ability of VLE to image large segments of the esophagus with microscopic cross-sectional detail may provide additional benefits including higher yield biopsies and more efficient tissue acquisition. Clinicaltrials.gov NCT02215291