Ralph H. Hruban

Pancreatic cancer

Substantial progress has been made in our understanding of the biology of pancreatic cancer, and advances in patients' management have also taken place. Evidence is beginning to show that screening first-degree relatives of individuals with several family members affected by pancreatic cancer can identify non-invasive precursors of this malignant disease. The incidence of and number of deaths caused by pancreatic tumours have been gradually rising, even as incidence and mortality of other common cancers have been declining. Despite developments in detection and management of pancreatic cancer, only about 4% of patients will live 5 years after diagnosis. Survival is better for those with malignant disease localised to the pancreas, because surgical resection at present offers the only chance of cure. Unfortunately, 80-85% of patients present with advanced unresectable disease. Furthermore, pancreatic cancer responds poorly to most chemotherapeutic agents. Hence, we need to understand the biological mechanisms that contribute to development and progression of pancreatic tumours. In this Seminar we will discuss the most common and deadly form of pancreatic cancer, pancreatic ductal adenocarcinoma.

Increased Prevalence of Pancreatic Cysts in Nonsecretor (Fut2 Null) Individuals Under Pancreatic Surveillance

Background: Pancreatic surveillance commonly detects pancreatic cysts, most of which are presumed intraductal papillary mucinous neoplasms (IPMN). FUT2, the gene responsible for secretor status, has been implicated in IPMN development, although not with pancreatic cancer risk. Methods: We evaluated the association between inactivating variants in FUT2, FUT3 and ABO and pancreatic cysts in 1,865 patients in the multicenter Cancer of the Pancreas Screening (CAPS) cohort. Multivariable logistic regression was performed to estimate adjusted odds for presumed IPMN prevalence. A Johns Hopkins surgical cohort of 1582 patients with IPMN, pancreatic ductal adenocarcinoma (PDAC), or carcinoma arising from IPMN was analyzed to examine trend in variant proportion by malignant progression. Results: In the CAPS cohort, older age (OR per year; 1.06 [1.05-1.07]; p < 0.001), FUT2-null (non-secretors) (OR: 1.66 [1.30-2.12]; p<0.001), non-O blood group (OR: 1.31 [1.06-1.62]; p=0.012), having familial-only vs. genetic/familial risk (OR: 1.49 [1.20-1.85] p<0.001), and participating site (OR; 2.20 [1.78-2.73], p<0.001) were associated with having pancreatic cysts. There were no significant FUT2/ABO presumed IPMN prevalence interactions. In the combined cohort, the proportion of FUT2-null patients with IPMN without carcinoma (by imaging or pathology) was 319 (27.4%) of 1162; 57 (24.3% of 234) with IPMN and carcinoma, and 225 (22.5%) of 998 with PDAC alone (p=0.009 for trend). Conclusions: IPMN is more prevalent in FUT2-null patients under pancreas surveillance, but the increased FUT2-null prevalence is not observed in patients with IPMN with carcinoma. Impact: Pancreatic cancer risk conferred by having IPMN may be lower in FUT2-null compared to FUT2-intact individuals.

An Iterative, Pathologist-in-the-Loop Workflow for Generation of Clinical-Grade Synthetic Pathology Images in a Diverse Cohort of Pancreatic Tumors

Background/Objectives: The training of diagnostic pancreatic pathologists is largely limited by the diversity of available pathology images, especially those of rare diseases or conditions. Methods: Using a cohort of seven pancreatic neoplasms, we developed an iterative, pathologist-in-the-loop workflow integrating scalable tile pruning, generative-model optimization, and postprocessing truncation. Two pathologists evaluated synthetic-image quality and subtype representation on a 0–3 scale; the independent pathologist was blinded to image source and truncation condition and also rated curated real training tiles. Results: Untruncated images had lower class-balanced FID than per-class-truncated images (6.64 versus 29.08) and higher recall and coverage, whereas truncation increased precision. The independent pathologist rated per-class-truncated images higher than matched non-truncated images (mean difference 0.80, bootstrap 95% CI 0.57–1.03). Quadratic-weighted Cohen’s κ was 0.599 (95% CI 0.489–0.691); after grouping ratings as 0–1 versus 2–3, raw agreement was 80.7% (95% CI 75.0–86.4%). Conclusions: Per-class truncation improved independently rated image quality and subtype representation. Successful synthetic histology generation requires careful data curation, domain-specific oversight, and independent validation; clinical utility requires separate task-based evaluation.

Discordant pathology following pancreatoduodenectomy for suspected intraductal papillary mucinous neoplasm: reassessment of cyst fluid genetic analysis

A molecular analysis of pancreatic cyst fluid is revolutionizing the clinical management of pancreatic cysts. Yet, there remains need for ongoing correlation of molecular testing with surgical pathology. An ~70-year old male was found to have focal main pancreatic duct dilation up to 7 mm at the pancreatic head/neck on imaging without any discrete mass lesions. Duct fluid analysis found elevated carcinoembryonic antigen and decreased glucose level suggestive of intraductal papillary mucinous neoplasm (IPMN). Yet, molecular testing revealed no mutations. In the absence of a mass lesion, he was recommended surveillance. A year later, he had stable imaging and underwent repeat molecular testing, which revealed KRAS and GNAS mutations. He underwent pancreatoduodenectomy for suspected main duct IPMN. Final pathology revealed low-grade, multifocal pancreatic intraepithelial neoplasia, not IPMN. Ongoing assessment of concordance between molecular testing and postoperative pathology after pancreatectomy is needed to refine the multimodal diagnostic evaluation of pancreatic cysts.

Integrated driver gene-based survival prediction in resected pancreatic cancer

BACKGROUND: The aim of this study was to develop an overall survival (OS) prediction model integrating key somatic driver gene mutations in patients with pancreatic ductal adenocarcinoma (PDAC) undergoing curative-intent pancreatectomy. METHODS: Patients with resected PDAC from a prospectively maintained pancreatectomy database from 2018 to 2023 were identified. Somatic mutations in KRAS, TP53, SMAD4, and CDKN2A were assessed using in-house or commercial sequencing panels. Restricted cubic spline analyses were used to assess non-linear effects. An integrated Cox proportional hazards model incorporating genomic and clinicopathological variables was developed to predict OS and validated using 1000 bootstrap resamples. RESULTS: In total, 885 patients were identified, of whom 663 met the inclusion criteria. After a median follow-up of 40.8 months, 387 deaths occurred (median OS of 28.4 months). The 1-, 2-, and 3-year OS rates were 80.0%, 56.2%, and 41.5% respectively. In univariable analysis, mutations in KRAS, TP53, SMAD4, and CDKN2A were each associated with inferior OS. Mortality risk increased with the cumulative number of mutated driver genes and remained independently significant in multivariable analysis (0-4; HR per mutation 1.27; P < 0.001). Other independent OS predictors included advanced age, resectability status, elevated pretreatment carbohydrate antigen 19-9 (CA19-9), larger tumour size, positive nodal stage, lymphovascular invasion, poor differentiation, neoadjuvant response classification, and receipt of adjuvant chemotherapy. The integrated model showed moderate discrimination (C-index 0.690 (95% c.i. 0.669 to 0.723); optimism-corrected C-index 0.676), which was significantly higher than AJCC eighth edition pathological staging alone (C-index 0.568 (95% c.i. 0.545 to 0.599); ΔC = 0.122; P < 0.001), with reasonable calibration. CONCLUSION: The cumulative burden of driver gene mutations provided independent and incremental prognostic information after PDAC resection. External validation is required before clinical application.

Brunner’s gland hamartoma in the setting of a duodenal diverticulum: A case report

FelkerJournal of the American College of Cardiology

Fulminant myocarditis linked to ~11% higher fractional shortening at six months vs acute myocarditis.

Insulinomatosis: A case report of a multifocal insulin-secreting entity

Virtual multiplex staining of the pancreatic islets across type 1 diabetes progression using a Schrödinger bridge

Classical hematoxylin and eosin (H&E) staining enables review of tissue morphology but lacks information regarding the molecular state of cells. Immunohistochemical (IHC) techniques label specific proteins in tissue, allowing differentiation of relevant structures that may go undetectable in H&E. However, the IHC process is complex, expensive, and time-consuming, especially for multiplex IHC (mIHC) limiting its use in large cohorts. Stain conversion of H&E to IHC using generative artificial intelligence models such as generative adversarial networks (GANs) represent one solution to this problem. However, GANs are unstable during out of distribution sampling and are prone to hallucinations or mode collapse, limiting their accuracy in challenging image conversion tasks. To address this, the field has recently turned to diffusion models. Here, we introduce Schrödinger-bridge for Multiplex ImmunoLabel Estimation (SMILE). Unlike conventional diffusion models that map from source to target through an intermediate Gaussian noise, Schrödinger-bridge diffusion models skip this step and have been shown to better preserve structures during image translation. To test the performance of SMILE, we generated a large cohort of high-fidelity H&E-mIHC image pairs from pancreatic organ donors, targeting insulin, glucagon, and CD3. Our dataset well-sampled across type-1 diabetes status, pancreas anatomical location, age, and sex. Using this cohort, we demonstrate the superiority of SMILE compared to GANs via a comprehensive evaluation framework incorporating texture, distribution, and antibody-specific metrics, as well as blinded pathologist reviews. We further confirmed the ability of SMILE to generate accurate mIHC images from H&Es generated at an external site, to perform whole slide image conversion, and to generate realistic three-dimensional maps of the pancreatic islets in non-diabetic, auto-antibody positive, and type-1 diabetic donor tissue. Finally, we performed stain conversion of paired H&E to HER2 and Ki67 images in breast cancer, confirming the superiority of SMILE in diverse stain conversion applications. Collectively, this framework provides a scalable pipeline for high-throughput proteomic inference from archival H&Es, providing transformative potential for pancreatic research and digital pathology.

PancreaSeq Genomic Classifier (PancreaSeq GC) Improves Pancreatic Cyst Classification and Detection of Advanced Neoplasia: A Multi-institutional Validation Study

TP53/KRAS variant allele frequency ratio predicts advanced disease in mucin-producing cystic neoplasms of the pancreas

OBJECTIVE: To identify molecular markers that predict advanced disease in mucin-producing cystic neoplasms of the pancreas. METHODS: In this single-institution study, 454 pancreatic cystic fluid specimens were examined by a clinical next-generation sequencing assay. RESULTS: TP53 mutations were detected in 25 (7.9%) of 318 specimens harboring KRAS (BRAF or GNAS) mutation(s). Of the 25, 12 had advanced cytology/histology (positive group), and 5 did not have advanced cytology/histology (negative group). Eight cases were classified as indeterminate. Since KRAS and TP53 variant allele frequencies (VAFs) can be as low as 1% to 5%, we analyzed the TP53/KRAS VAF ratio and demonstrated a significantly higher VAF ratio in the positive group (P = .005). A VAF ratio of 1 or higher, interpreted as an indicator of a dominant TP53 clone, and concurrent TP53, CDKN2A, and SMAD4 mutations were seen only in the positive group. Lower TP53 ratios, interpreted as a minor TP53 clone, were seen in all 5 negative group specimens. CONCLUSIONS: These results suggest that VAF ratios and concurrent mutations can be incorporated into multimodality assessments of pancreatic cysts. Longitudinal studies in patients with a lower initial TP53 VAF ratio are warranted to elucidate whether serial VAF ratios are more accurate markers than a single VAF measurement.

Benchmarking robustness of automated CT pancreas segmentation: achieving human-level reliability through human-in-the-loop optimization

Background Deep learning–based pancreas segmentation in CT has advanced rapidly yet remains evaluated primarily with mean overlap metrics that fail to capture robustness—defined as the proportion of cases reaching human-level performance. Models performing well on mean Dice or surface metrics can still fail unpredictably across scanners or anatomies. Because early detection and quantitative biomarkers rely on consistent segmentation, robustness is critical for clinical deployment. Purpose To systematically evaluate the robustness of deep learning models for pancreas segmentation relative to human readers and to investigate an active learning strategy to improve reliability. Materials and Methods We retrospectively assembled 903 venous-phase CT scans from patients with presumed normal-appearing pancreases and without known pancreatic disease (2005-2023), split into 803 for training/validation and 100 healthy test cases. Each test case had 4 independent human segmentations. Inter-reader variability on this healthy-only test set defined the empirical human distribution, providing an upper-bound estimate of robustness. We introduced a Fractional Threshold (FT) metric, measuring the proportion of model predictions exceeding the minimum human performance. Robustness was assessed across models trained from scratch, fine-tuned, or pretrained, including both normal and abnormal cases. An active learning approach identified high-uncertainty predictions for human revision. Statistical comparisons were performed using the Wilcoxon signed-rank and proportions Z-tests. Results The best model, a 3-dimensional U-Net trained from scratch, achieved a Dice Similarity Coefficient (DSC) of 0.88 ± 0.04 and Normalized Surface Dice (NSD) of 0.77 ± 0.09, approaching human-level segmentation (DSC = 0.89 ± 0.03; NSD = 0.75 ± 0.07). However, FT for DSC and NSD remained lower than human performance in most cases, indicating persistent model variability. Human-in-the-loop revision of acquisition-flagged outliers increased FT to 0.99, with an average time of 1.54 minutes per case, corresponding to a 23-fold workload reduction. Conclusion Automated pancreas segmentation reduces workload but remains constrained by tail-case failures. Active learning enhances model reliability, bridging the gap between artificial intelligence and human-level performance.

Minimizing and quantifying uncertainty in AI-informed decisions: Applications in medicine

Recurrent BEND2 Fusion Genes Identified by Whole Transcriptome Sequencing of Non-Functional Pancreatic Neuroendocrine Tumors Correlate with Poor Patient Prognosis

The Brescia International Multidisciplinary Consensus Guidelines on the Optimal Pathology Assessment and Multidisciplinary Pathways of Non‐Pancreatic Neoplasms in and Around the Ampulla of Vater (PERIPAN)

Three-dimensional assessments are necessary to determine the true, spatially resolved composition of tissues

Methods for spatially resolved cellular profiling of tissue sections enable in-depth study of inter- and intra-sample heterogeneity but often profile small regions, requiring evaluation of many samples to compensate for limited assessment. Recent advances in three-dimensional (3D) tissue mapping offer deeper insights; however, attempts to quantify the information gained in transitioning to 3D remains limited. Here, to compare inter- and intra-sample tissue heterogeneity, we analyze >100 pancreas samples as cores, whole-slide images (WSIs), and cm 3 -sized 3D samples. We show that tens of WSIs and hundreds of tissue microarrays are needed to approximate the compositional tissue heterogeneity of tumors. Additionally, spatial correlations of pancreatic structures decay significantly within microns, demonstrating that isolated two-dimensional (2D) sections poorly represent their surroundings. Through 3D simulations, we determined the number of slides necessary to accurately measure tumor burden. These results quantify the power of 3D mapping and establish sampling methods for biological studies prioritizing composition or incidence.

CODAvision: best practices and a user-friendly interface for rapid, customizable segmentation of medical images

Image-based machine learning tools have emerged as powerful resources for analyzing medical images, with deep learning-based semantic segmentation commonly utilized to enable spatial quantification of structures in images. However, customization and training of segmentation algorithms requires advanced programming skills and intricate workflows, limiting their accessibility to many investigators. Here, we present a protocol and software for automatic segmentation of medical images guided by a graphical user interface (GUI) using the CODAvision algorithm. This workflow simplifies the process of semantic segmentation of microanatomical structures by enabling users to train highly customizable deep learning models without extensive coding expertise. The protocol outlines best practices for creating robust training datasets, configuring model parameters, and optimizing performance across diverse biomedical image modalities. CODAvision enhances the usability of the CODA algorithm ( Nature Methods , 2022) by streamlining parameter configuration, model training, and performance evaluation, automatically generating quantitative results and comprehensive reports. We expand beyond the original implementation of CODA to serial histology by demonstrating robust performance across numerous medical image modalities and diverse biological questions. We provide sample results in data types including histology, magnetic resonance imaging (MRI), and computed tomography (CT). We demonstrate the diverse use of this tool in applications including quantification of metastatic burden in in vivo models and deconvolution of spot-based spatial transcriptomics datasets. This protocol is designed for researchers with interest in rapid design of highly customizable semantic segmentation algorithms and a basic understanding of programming and anatomy.

Prognostic Features in Surgically Resected Well-Differentiated Pancreatic Neuroendocrine Tumors: An Analysis of 904 Patients with 7882 Person-Years of Follow-Up

Importance The clinical behavior of well-differentiated pancreatic neuroendocrine tumors (PanNETs) is difficult to predict. Objective To define more accurately prognosticators for patients with a surgically resected PanNET. Design The pathology and Ki-67 immunolabeling index of PanNETs resected from 904 patients was correlated with patient outcome. Setting Academic tertiary care hospital. Participants Consecutive patients who had a PanNET resected between 1985 and 2025. Results The mean patient age at surgery was 56.6 years (SD 14.0), 477 were male (52.8%), and 7882 person-years of follow-up were obtained (mean 8.8 years, SD 6.5). The 10-year survival was 81% (95% CI: 77,86%) for patients with G1 PanNETs (Ki-67 <3%), 68% (95% CI: 61,76%) for patients with G2a PanNETs (Ki-67 3-<10%), 44% (95% CI: 29,66%) for patients with G2b PanNETs (Ki-67 of 10%-≤20%), and 23% (95% CI: 8,61%) for patients with G3 PanNETs. Metastases (HR 4.7, p <0.0001), vascular invasion (HR 3.0, p <0.0001), tumor size ≥ 2 cm (HR 2.88, p <0.0001), perineural invasion (HR 2.42, p<0.0001), and positive margins (HR 2.18, p <0.0001) were associated with worse overall survival. Insulinoma (HR 0.34, p=3e-04), sclerosing variant (HR 0.47, p=0.05), and cystic variant (HR 0.61, p=0.05) were associated with improved overall survival. T stage and N stage were all statistically significant classifiers of overall survival. Similar associations were found with respect to disease relapse. There was a significant (P<0.001) increase in the proportion of patients diagnosed with stage I vs stage IV disease over time. Conclusions and relevance This study supports the classification of PanNETs into four grades (G1, G2a, G2b, and G3) based on Ki-67 labeling, which allows a more accurate prognostic assessments of patients. Key Points Question Does subdividing grade 2 well-differentiated pancreatic neuroendocrine tumors (PanNETs) into grade 2a (Ki-67 3-<10%), and grade 2b (Ki-67 10%-≤20%) improve patient prognostication after surgery? Findings In this single-institution cohort study of 904 adults, patients with grade 2a and patients with grade 2b PanNETs had distinct outcomes. Furthermore, the magnitude of the risk associated with metastases, perineural invasion, positive margins, tumor size, vascular invasion, sclerosing variant and cystic variant was refined. Meaning This study supports classifying PanNETs into four grades (G1, G2a, G2b, and G3), and provides a basis for accurate prognostic assessments of patients.

Proteogenomic characterization of non-functional pancreatic neuroendocrine tumors unravels clinically relevant subgroups

The majority of neuroendocrine neoplasms in pancreas are non-functional pancreatic neuroendocrine tumors (NF-PanNETs), which exhibit a high occurrence of distant metastases with limited therapeutic options. Here, we perform a comprehensive molecular characterization of 108 NF-PanNETs through integrative analysis of genomic, transcriptomic, proteomic, and phosphoproteomic profiles. Proteogenomic analysis provides functional insights into the genomic driver alterations of NF-PanNETs, revealing a potential mediator of MEN1 alterations using Men1-conditional knockout mice. Machine-learning-based modeling uncovers a three-protein signature as an independent prognostic factor, which is validated by an independent external cohort. Proteomic and phosphoproteomic-based stratification identifies four subtypes with distinct molecular characteristics, immune microenvironments, and clinicopathological features. Drug screening using patient-derived tumor organoids identifies cyclin-dependent kinase (CDK) 5 and Calcium Voltage-Gated Channel Subunit Alpha1 D (CACNA1D) as ubiquitous and subtype-specific targets, respectively, with in vivo validation using xenograft models. Together, our proteogenomic analyses illustrate a comprehensive molecular landscape of NF-PanNETs, revealing biological insights and therapeutic vulnerabilities.

Diagnostic Performance of Combined Conventional CT Imaging Features and Radiomics Signature in Differentiating Grade 1 Tumors from Higher-Grade Pancreatic Neuroendocrine Neoplasms

BACKGROUND/OBJECTIVES: Accurate identification of grade 1 (G1) pancreatic neuroendocrine tumors (PanNETs) is crucial due to their rising incidence and emerging nonsurgical management strategies. This study evaluated whether combining conventional CT imaging features, CT radiomics features, and clinical data improves differentiation of G1 PanNETs from higher-grade tumors (G2/G3 PanNETs and pancreatic neuroendocrine carcinomas [PanNECs]) compared to using these features individually. METHODS: A retrospective analysis included 133 patients with pathologically confirmed PanNETs or PanNECs (70 males, 63 females; mean age, 58.5 years) who underwent pancreas protocol CT. A total of 28 conventional imaging features, 4892 radiomics features, and clinical data (age, gender, and tumor location) were analyzed using a support vector machine (SVM) model. Data were divided into 70% training and 30% testing sets. RESULTS: The SVM model using the top 10 conventional imaging features (e.g., suspicious lymph nodes and hypoattenuating tumors) achieved 75% sensitivity, 81% specificity, and 79% accuracy for identifying higher-grade tumors (G2/G3 PanNETs and PanNECs). The top 10 radiomics features yielded 94% sensitivity, 46% specificity, and 69% accuracy. Combining all features (imaging, radiomics, and clinical data) improved performance, with 94% sensitivity, 69% specificity, 79% accuracy, and an F1-score of 0.77. The radiomics score demonstrated an AUC of 0.85 in the training and 0.83 in the testing set. CONCLUSIONS: Conventional imaging features provided higher specificity, while radiomics offered greater sensitivity for identifying higher-grade tumors. Integrating all three features improved diagnostic accuracy, highlighting their complementary roles. This combined model may serve as a valuable tool for distinguishing higher-grade tumors from G1 PanNETs and potentially guiding patient management.