Introduction

Endoscopic ultrasound (EUS) elastography has emerged as a valuable tool for assessing pancreatic tissue stiffness, aiding in the differentiation of normal pancreas, chronic pancreatitis (CP), and solid pancreatic masses, including pancreatic adenocarcinoma (PDAC) and cholangiocarcinoma (CC). The newly developed two-dimensional shear wave elastography (2D-SWE) provides real-time, quantitative elasticity maps with improved reproducibility. This study aims to evaluate the performance of 2D-SWE during EUS for differentiating normal pancreas, CP, and solid pancreatic tumors.

Aims & Methods

This prospective study included patients undergoing EUS for pancreatic evaluation. All 2D-SWE measurements were made using the Aplio i800 EUS ultrasound machine. Four groups were analyzed: (1) individuals with a normal pancreas, (2) patients with CP (based on Rosemont criteria), (3) patients with PDAC, and (4) patients with CC (all diagnoses confirmed histopathologically). 2D-SWE was performed during EUS using a linear-array echoendoscope. Quantitative elasticity values (kPa) were recorded for the normal pancreas, CP-affected tissue, and tumor lesions. The diagnostic performance of 2D-SWE was assessed using intraclass correlation coefficient (ICC), receiver operating characteristic (ROC) analysis, and pairwise statistical comparisons.

Results

A total of 30 patients were included (6 normal, 5 CP, 17 PDAC, 2 CC). The intra-observer reproducibility of SWE measurements was excellent (ICC = 0.89). Mean elasticity values significantly differed among groups, with PDAC showing the highest variability (mean SWE = 54.1 ± 45.7 kPa, median = 33.1), CC presenting elevated stiffness (mean SWE = 77.4 ± 11.0 kPa), CP showing heterogeneous values (mean SWE = 57.5 ± 52.3 kPa, median = 43.6), and normal pancreas showing the lowest stiffness (mean SWE = 7.5 ± 3.9 kPa, median = 6.2).
ROC analysis for distinguishing malignant solid tumors (PDAC + CC) from normal pancreas showed excellent diagnostic performance (AUC = 0.93). The optimal cutoff of 17.0 kPa yielded a sensitivity of 84.2%, specificity of 100%, positive predictive value (PPV) of 100%, negative predictive value (NPV) of 66.7%, and overall accuracy of 88%.
Pairwise t-tests confirmed a statistically significant difference between PDAC and normal pancreas (p = 0.0007), with trends observed for CC vs normal (p = 0.0605) and CP vs normal (p = 0.099). No statistically significant differences were observed among solid tumor subtypes (PDAC vs CC, PDAC vs CP, CC vs CP).

Conclusion

2D-SWE during EUS provides highly reproducible and quantitative elasticity values that reliably differentiate normal pancreas from solid pancreatic tumors. This method demonstrates excellent diagnostic performance, particularly in identifying PDAC, and shows promising trends for distinguishing CP and CC, supporting its clinical integration in EUS evaluation protocols.

Introduction

Pancreatic ductal adenocarcinoma (PDAC) is a highly desmoplastic tumor that significantly affects its aggressiveness and treatment resistance. Cancer-associated fibroblasts (CAFs) are key components of the desmoplastic stroma and are now recognized as dynamic, heterogeneous cells that contribute to disease progression. While CAFs play a role in creating an immunosuppressive tumor microenvironment (TME), their precise impact on survival and cellular dynamics remains unclear. This study aims to provide an unbiased classification of CAF subtypes and investigate their functional relevance and dynamics in chronic pancreatitis (CP) and PDAC.

Aims & Methods

We conducted single-cell RNA sequencing (scRNA-seq) on 40,000 fibroblasts isolated from resected human CP and PDAC tissues. We clustered the CAFs and examined the driving genes for each subtype. Validating these subtypes involved immunohistochemistry (IHC)-multiplexing on tissue microarrays (TMA) from 126 PDAC and 66 CP patients. Additionally, we performed flow cytometry on human PDAC CAFs and utilized immunocytochemistry (ICC) with our newly identified markers.

Results

Our clustering revealed 11 distinct CAF subtypes. To facilitate future mechanistic experiments, we also analyzed the surface receptor signatures of these enriched subtypes. IHC multiplexing confirmed the existence of these subtypes, with End Stage inflammatory CAFs (iCAFs) (35.7% in PDAC vs. 56.6% in CP) and End Stage myofibroblastic CAFs (myCAFs) (35.3% in PDAC vs. 25.6% in CP) being the most prevalent. Flow cytometry results corroborated IHC findings. Notably, we identified a novel subtype, Lipofibroblast, in both PDAC and CP, with RNA velocity analysis indicating its ability to differentiate into other subtypes, driven by Periostin (POSTN).

Conclusion

The discovery of a novel Lipofibroblast subtype with differentiation potential highlights a new avenue for understanding CAF plasticity and therapeutic targeting in PDAC and CP.

Introduction

Pancreatic ductal adenocarcinoma (PDAC) is a highly aggressive cancer that poses significant therapeutic challenges. In this regard, the identification of prognostic biomarkers is crucial to improve the prediction of clinical outcomes and guide therapeutic decisions. Tumour-associated autoantibodies (TAAbs) have been identified as prognostic biomarkers in several types of cancer, as they can be generated before the first clinically detectable signs of disease.

Aims & Methods

The aim of this study was to identify TAAbs that can be used as prognostic biomarkers in PDAC. 80 sera from patients with PDAC, divided into four groups according to disease stage, were included in the study: patients with premalignant lesions (n = 20), patients undergoing curative resection after diagnosis (CR) (n = 20), patients with unresectable advanced locoregional disease (LA) (n = 20) and patients with metastatic disease (MT) (n = 20). Sera from each group were analysed divided into 4 pools of 5 samples each using the HuProt Human Protreome microarray v4.0 (CDI Laboratories, Baltimore, MD). Candidate TAAbs with high specific reactivity in MT and/or LA samples were selected according to the following criteria: z-score ≥ 3 in at least 2/4 pools and fold-change > 1.2. The RNAseq data collected in the Human Protein Atlas were used to determine protein expression levels and prognostic value. The GSE15471, GSE28735 and GSE16515 databases were used to compare expression levels with those of healthy subjects without the disease.

Results

43 TAAbs demonstrated high levels of specific reactivity in samples from MT and/or LA patients. Of these highly recognised proteins, 15 were selected for showing notable levels of expression in tumour tissue from PDAC patients and prognostic capacity in some types of cancer. Of these 15, 2 proteins involved in tumour development were identified. Survival analyses showed that six of the 15 selected proteins were significantly associated with a worse prognosis in PDAC patients (p < 0.05). Finally, in the comparison with control tumour tissue, 5 proteins were identified with significantly higher expression in patients diagnosed with the disease (p < 0.05 in at least 2 of the cohorts used).

Conclusion

15 TAAbs have been identified as promising candidates for prognostic biomarkers in PDAC. Further validation studies are required to confirm their prognostic value and to develop a robust biomarker signature with potential for clinical application and market translation.

Introduction

The aetiology of acute pancreatitis remains elusive in 10-30% of patients after clinical, biological and morphological investigations. Endoscopic ultrasound has shown promise and is particularly effective in exploring the bilio-pancreatic region.

Aims & Methods

The aim of our study was to evaluate the contribution of endoscopic ultrasound to the aetiological diagnosis of unexplained acute pancreatitis.
This is a retrospective descriptive study of 89 patients who underwent endoscopic ultra sound for unexplained acute pancreatitis between January 2010 and January 2025.

Results

The mean age of our patients was 55 +/- 16.3 years, with extremes ranging from 17 to 89 years. The sex ratio (M/F) was 0.81, with a slight female predominance of 55.1%. A history of previous acute pancreatitis was found in 31.9% of cases, and 16.3% of patients had undergone cholecystectomy. The diagnostic yield in our study was 68%. Biliary origin was retained in 40.4% (n=36) of patients, 19 of whom underwent complementary ERCP at the same anaesthetic time. Neoplastic pathology was found in 32.5% of cases (n=30). Chronic pancreatitis was reported in 22.4% of patients (n=20). Two cases of pancreas divisum, one case of sphincter of Oddi dysfunction and one case of choledocele were found. In the remaining 32% of cases, the endoscopic ultrasound was normal and no cause could be found.

Conclusion

Endoscopic ultrasound is currently an indispensable tool in the etiological investigation of unexplained acute pancreatitis. In our study, the diagnostic efficiency was around 68%.

Introduction

Acute pancreatitis (AP) is a common gastrointestinal disorder with significant morbidity and mortality, particularly in severe cases (SAP). Early prediction of SAP is crucial for timely intervention and improved patient outcomes. Traditional scoring systems like Ranson, BISAP, and APACHE II have limitations in accuracy and timeliness. Machine learning (ML) has shown promise in enhancing SAP prediction, but the potential of advanced architectures like Transformer models remains unexplored. This study aims to develop a Transformer-based model for SAP prediction and compare its performance with traditional ML methods.

Aims & Methods

The primary aim of this study was to develop a Transformer-based prediction model for acute pancreatitis severity and evaluate its performance compared to traditional machine learning models. Retrospective data from 722 AP patients admitted to the Second Affiliated Hospital of Guilin Medical University between January 2020 and June 2024 were analyzed. Patient data included 24-hour admission laboratory indicators such as electrolytes, liver function, renal function, and inflammatory markers. To construct the prediction model, a comprehensive workflow was designed. Data preprocessing involved feature vectorization using One-Hot encoding, missing value imputation using the K-Nearest Neighbor (KNN) algorithm, normalization of feature values to a [0, 1] range, and handling class imbalance using Synthetic Minority Over-Sampling Technique (SMOTE). Feature selection was performed by dividing patients into SAP and MAP groups, identifying features with significant differences using non-parametric tests, and ranking features by their relevance through correlation analysis. Features causing a decline in AUC during incremental addition were removed to optimize the feature set. The Transformer model, leveraging self-attention mechanisms to process sequential data, was trained alongside traditional ML models (Logistic Regression, Decision Tree, Random Forest, and XGBoost). Five-fold cross-validation was used to prevent overfitting, and model performance was evaluated using metrics such as AUC, accuracy, precision, recall, F1 score, and specificity. SHAP analysis was employed to interpret the Transformer model's predictions, providing insights into the contribution of each feature to the model's output.

Results

The Transformer model outperformed traditional ML models (Table 1), achieving an AUC of 0.9662 and exceeding 0.90 in all metrics. It maintained robust performance even with minimal training data (AUC: 0.8437 at 5% training). SHAP analysis highlighted key predictors (calcium, amylase, basophil percentage, CO2 binding capacity), aligning with clinical relevance.

Table 1. Performance metrics comparison of different models.

Conclusion

The Transformer model demonstrated superior accuracy, generalizability, and interpretability for SAP prediction, particularly in small datasets. Its clinical utility is enhanced by SHAP-driven insights into key predictors. Retrospective design limits prospective validity; future work should validate the model in larger cohorts and integrate additional biomarkers. This study supports the potential of transformer in medical prediction to improve patient outcomes through timely interventions.

References

[1]. Sinonquel P, Laleman W, Wilmer A. Advances in acute pancreatitis. Curr Opin Crit Care. 2021;27(2):193–200.
[2]. Thapa R, Iqbal Z, Garikipati A, et al. Early prediction of severe acute pancreatitis using machine learning. Pancreatology. 2022;22(1):43–50.
[3]. Luo Z, Shi J, Fang Y, et al. Development and evaluation of machine learning models and nomogram for the prediction of severe acute pancreatitis[J]. Journal of Gastroenterology and Hepatology, 2023, 38(3): 468-475.
[4]. Zhang C, Peng J, Wang L, et al. A deep learning-powered diagnostic model for acute pancreatitis[J]. BMC Medical Imaging, 2024, 24(1): 154.
[5]. Liao H, Huang C, Liu C, et al. Deep learning-based MVIT-MLKA model for accurate classification of pancreatic lesions: a multicenter retrospective cohort study[J]. La radiologia medica, 2025: 1-16.

Introduction

Acute pancreatitis (AP) is a common acute inflammatory disease of the abdomen. ATP citrate lyase (ACLY) has been shown to play a critical role in mitigating acute pancreatitis-associated injury. Lentinan, a multifunctional polysaccharide derived from shiitake mushrooms, exhibits biological activities such as anti-inflammatory, antioxidant, and immunomodulatory effects, and is widely used in clinical settings. However, the role of Lentinan in acute pancreatitis remains unclear.

Aims & Methods

This study aimed to investigate the protective effects of Lentinan against acute pancreatitis and whether this protection depends on ATP citrate lyase. Adult healthy male C57BL/6 mice were randomly divided into control group, acute pancreatitis group, and Lentinan + acute pancreatitis group. The cerulein-LPS induced AP model was established by intraperitoneal injection of cerulein solution (100 μg/kg body weight) 10 times at hourly intervals, followed by a single injection of lipopolysaccharide (5 mg/kg) immediately after the last cerulein injection. In the lentinan + acute pancreatitis group, mice were pretreated with intraperitoneal injections of of Lentinan solution (50 mg/kg, 100 mg/kg, or 200 mg/kg) for three consecutive days before AP induction. The severity of pancreatitis was assessed by comparing plasma activities of amylase, lipase, alanine aminotransferase, aspartate aminotransferase, and plasma levels of urea, and creatinine, as well as pancreatic histopathological scores. Systemic inflammatory response was evaluated by measuring plasma IL-1β and TNF-α levels through enzyme-linked immunosorbent assay. Pancreatic infiltration of macrophages and neutrophils was detected by immunofluorescence, and protein levels of ACLY, NLRP3, and IL-1β in pancreatic tissues were measured by Western blot. To further validate the critical role of ACLY, AP was induced in mice with pancreas-specific ACLY knockout, with Lentinan pretreatment.

Results

Lentinan significantly alleviated pancreatic histopathological damage caused by AP, including edema, necrosis, and inflammatory cell infiltration. It also reduced plasma aminotransferase activity, creatinine, and urea levels, but did not significantly decrease plasma amylase or lipase activity. A higher dose of Lentinan (200 mg/kg) was less effective than lower doses in reducing plasma aminotransferase, creatinine, and urea in AP mice, possibly due to greater hepatorenal toxicity. Therefore, a moderate dose of Lentinan (100 mg/kg) provided effective protection against acute pancreatitis and was therefore selected for subsequent experiments. Additionally, Lentinan reduced pancreatic infiltration of macrophages and neutrophils, decreased plasma pro-inflammatory cytokine levels, upregulated pancreatic ACLY protein expression, and downregulated the levels of NLRP3 inflammasome and IL-1β proteins in pancreatic tissues. In mice with pancreas-specific ACLY knockout, both local pancreatic injury and systemic inflammation caused by AP were significantly exacerbated, and the protective effect of Lentinan against acute pancreatitis was diminished.

Conclusion

Lentinan exerts protective effects against acute pancreatitis by improving local and systemic inflammatory responses. The expression of pancreatic ACLY is indispensable for mediating this protective effect, which was partially achieved through the inhibition of the NLRP3 inflammasome pathway. The present study establishes a novel mechanism for this clinically available immunomodulator, which could have potential therapeutic implications in acute pancreatitis.

Introduction

The learning curve for EUS-guided Gastroenterostomy (EUS-GE) has been assessed through retrospective studies involving few elite senior endoscopists (SE), using procedural time as the main surrogate outcome. This study aims to evaluate the training course of junior endoscopists (JE) approaching EUS-GE using the WEST technique, considering both technical efficiency and clinically relevant patient outcomes.

Aims & Methods

From a prospective single-centre registry of EUS-GE (PROTECT, ClinicalTrials.gov NCT04813055) using the Wireless Simplified Technique (WEST), the learning curve of 3 JE, experienced in pancreatobiliary endoscopy (PBE) was analysed to identify inflection points (“knots”), applying adaptive regression and cumulative sum control chart (CUSUM) on several continuous and dichotomous variables. EUS-GE performed by a SE and those on post-surgical anatomy were excluded.

Results

From a prospective database of 165 EUS-GE, 100 were performed by a JE with previous EUS-GE training (JE1) and 26 by two naïve JEs.
In JE1’s curve, neither procedural nor fluoroscopy time showed any improvement while radiation dose (mean Air Kerma: 392 ± 317 Gy) significantly decreased (F-test p=.0005) after 57 (adaptive regression) and 72 interventions (CUSUM), stabilizing thereafter. The need for supervision/backup from a SE showed a significant knot after 32 procedures (p<.0001). Technical success (100%), clinical success (95.0%), adverse events (AEs; 6.0%) and moderate/severe AEs (4.0%) remained constant through the learning curve, clustered according to the aforementioned knots. Conversion from EUS-GE to enteral stenting was rare (7.2%) and non-significantly decreased over time.
Despite requiring significantly higher fluoroscopy time (234 vs. 187 seconds, p=.009) and procedural time (55 vs. 43 minutes, p=.002), naïve endoscopists achieved efficacy and safety outcomes comparable to the remaining cohort.

Conclusion

From the analysis of a prospective learning curve, performing ≈ 30 EUS-GE seems required for independent practice, and ≈ 70 to stabilize the fluoroscopy usage. EUS-GE training of naïve PBE endoscopists in high-volume institutions, using a standardized technique, does not compromise clinical and safety standards.