Introduction

Recently, red dichromatic imaging (RDI), a novel image-enhanced endoscopy, which comprises wavelengths of amber (600 nm), red (630 nm) and green (540 nm), was launched. Previous studies have indicated that RDI improved the visibility of gastrointestinal bleeding. However, visibility assessments have been based on the subjective decision of evaluators. Herein, we investigated the recognition of bleeding points during endoscopic submucosal dissection (ESD) under RDI compared with that under white light imaging (WLI).

Aims & Methods

Consecutive patients scheduled to undergo esophageal or gastric ESD at a single center were enrolled. Paired videos of active bleeding during ESD under both WLI and RDI were created. Six endoscopists identified the virtual hemostasis point on still images after random video viewing. The distance between the virtual hemostasis and actual bleeding points was scored in four levels (0–3 points), and the association with the color value was analyzed in both WLI and RDI.

Results

We retrospectively evaluated 116 videos for 58 bleeding points. The median visibility score and recognition rate for RDI were significantly higher than those for WLI (2.17 vs. 1.42, p<0.001 and 62.1% vs 27.6%, p<0.001). The median visibility scores were significantly higher for RDI than for WLI under blood pooling (2.00 vs. 1.00, p<0.0016) and no blood pooling (2.50 vs. 1.67, p<0.001). The median visibility scores for RDI were significantly higher than those for WLI in oozing hemorrhage (2.17 vs. 1.33, p<0.001) and pulsating hemorrhage (2.17 vs. 1.50, p=0.030). The median visibility scores in RDI were significantly higher than those in WLI between trainees (2.00 vs. 1.33, p<0.001) and experts (2.67 vs. 1.67, p<0.001). The median color difference of RDI was significantly higher than that of WLI (8.97 vs. 3.69, p<0.001). The correlation coefficient between the visibility score and color difference was 0.712 (strong correlation).

Conclusion

RDI can provide better recognition of bleeding points than WLI during ESD, regardless of the bleeding pattern and experience of the endoscopist. Further studies are warranted to investigate whether RDI improves ESD outcomes.

Disclosure

The authors declare no conflict of interest for this article.

Introduction

Clostridioides difficile infection (CDI) is among of the most common causes of nosocomial diarrhea with high morbidity and mortality. Recurrent and refractory disease courses in particular represent a special challenge in the management of these patients. The identification of patients at risk for recurrent courses could improve management, e.g. through the use of new and effective therapies in patients with CDI.

Aims & Methods

The aim of this study was to investigate the relationship between the presence of different comorbidities and the course of CDI.
In a retrospective, monocentric cohort study of a tertiary centre over a period of 10 years, patients with CDI-related hospitalizations were examined. The study included 336 patients with available information on comorbidities (diverticulosis, biliary diseases (cholecysto-, docholithiasis, cholecystitis, cholangitis or cholangiectasis) inflammatory bowel diseases (IBD), liver cirrhosis and malignancies) and the diagnosis of the recurrent CDI (rCDI) vs. non-recurrent course of CDI (nrCDI). Patients on dialysis or immunosuppressed patients were also considered as comorbid.

Results

Among 336 patients included in the study, 227 (67.6%) presented with at least one comorbidity. The most common were malignancy (30.4%), biliary diseases (19.4%), diverticulosis (15.8%), liver cirrhosis (14.3%), and inflammatory bowel disease (IBD, 8.9%). rCDI occurred more frequently in patients with diverticulosis and biliary disease. Diverticulosis was present in 24.4% of rCDI cases vs. 12.6% in nrCDI cases (OR 2.24, 95% CI 1.22–4.17). Biliary diseases were observed in 26.7% of rCDI vs. 16.7% of nrCDI patients (OR 1.82, 95% CI 1.01–3.26). No significant associations with rCDI were found for IBD (OR 0.66, 95% CI 0.27–1.65), malignancy (OR 1.13, 95% CI 0.67–1.88), or liver cirrhosis (OR 0.79, 95% CI 0.38–1.61).

Conclusion

These findings suggest diverticulosis and biliary disease may be independent risk factors for rCDI, whereas other common comorbidities are not significantly associated with recurrence. Structured assessment of risk conditions may help to select patients for tailored management to prevent rCDI disease course.

Introduction

With about 40% of adults infected with Helicobacter pylori (H. pylori), China has the highest number of gastric cancer (GC) cases worldwide.¹ Household transmission is the primary route for H. pylori infection,² and family-based testing and treatment has been proposed to mitigate GC cancer risk and prevent transmission.^1,3,4 A household-based transmission model for H. pylori would enhance understanding of H. pylori dynamics and inform intervention strategies. However, few models⁵ have been developed for H. pylori, and none incorporates household transmission.

Aims & Methods

We developed an agent-based household transmission model for H. pylori in China to study transmission patterns from 1990 to 2021. Building on our previous modeling framework,⁶ the model includes demographic and transmission components. The demographic simulation captures births, deaths, marriages, and mobility among youth, couples, and the elderly. Mobility parameters were calibrated using joint distributions of household size and generation from Chinese censuses of 1990, 2000, 2010, and 2020, allowing realistic household dynamics in China. In the transmission simulation, household and community transmissions were modeled using a susceptible-infected structure, applying density-dependent transmission within households and frequency-dependent transmission in the community. Contact rates, defined as the number of contacts adequate for transmission by an individual per year, were allowed to vary over time, and were dependent on the susceptible individual’s age. Contact rate variation was modeled by adjusting base rates with age- and year-specific multipliers, each following a continuously decreasing function. Simulations began in 1900, with the first 90 years as a burn-in period. Transmission parameters were calibrated and validated using data from our meta-analyses and a large national family-based survey from China in 2021⁷.

Results

The model aligned well with calibration targets. Key parameters and calibration results are summarized in the Table. Consistent with observed data, larger households had higher infection rates than smaller households. Three-generation households were more likely to have infected members than one- or two-generation households. Model calibration showed best fit when the within-household contact rate (0.2) was higher than the community rate (0.1), and when most infections occurred before age 20. Validation confirmed model accuracy: children’s infection probabilities were 31% with both parents infected, 21% with one infected parent, and 9% with neither. Our model found that family clustering became more pronounced in recent years with lower prevalence compared to earlier periods with higher prevalence.

Table. Calibration and validation results for the transmission model

Conclusion

We developed a novel agent-based model simulating community and household transmission of H. pylori. Our results confirm that household transmission is the primary infection route, and that childhood is the predominant period of acquisition. Notably, family clustering becomes more prominent in low-prevalence contexts. This model provides a valuable tool for evaluating various control strategies and offers insights difficult to obtain in real-world data. Future extensions could incorporate antibiotic resistance dynamics and adapt the model to other country contexts.

References

1. Ding, S.-Z. et al. Chinese Consensus Report on Family-Based Helicobacter pylori Infection Control and Management (2021 Edition). Gut 71, 238–253 (2022).
2. Malfertheiner, P. et al. Helicobacter pylori infection. Nat. Rev. Dis. Primer 9, 1–24 (2023).
3. Chey, W. D. et al. ACG Clinical Guideline: Treatment of Helicobacter pylori Infection. Off. J. Am. Coll. Gastroenterol. ACG 119, 1730 (2024).
4. Shah, S. C., Wang, A. Y., Wallace, M. B. & Hwang, J. H. AGA Clinical Practice Update on Screening and Surveillance in Individuals at Increased Risk for Gastric Cancer in the United States: Expert Review. Gastroenterology 168, 405-416.e1 (2025).
5. Rupnow, M. F. T., Shachter, R. D., Owens, D. K. & Parsonnet, J. A Dynamic Transmission Model for Predicting Trends in Helicobacter pylori and Associated Diseases in the United States. Emerg. Infect. Dis. 6, 228–237 (2000).
6. Blok, D. J. et al. The long-term impact of the Leprosy Post-Exposure Prophylaxis (LPEP) program on leprosy incidence: A modelling study. PLoS Negl. Trop. Dis. 15, e0009279 (2021).
7. Zhou, X.-Z. et al. Large-scale, national, family-based epidemiological study on Helicobacter pylori infection in China: the time to change practice for related disease prevention. Gut 72, 855–869 (2023).B1

Introduction

Steatotic liver disease (SLD) encompasses a spectrum of conditions, including metabolic dysfunction-associated SLD (MASLD), alcohol-related liver disease (ALD), and the overlapping category of MetALD. SLD affects more than a third of adults globally, but the majority remain undiagnosed, partially due to its lack of symptoms and limited awareness among patients and physicians. Adults undergoing screening colonoscopy, typically aged ≥50 years, represent a key population for targeted SLD detection, as they have a higher prevalence of cardiometabolic risk factors and is already engaged with preventive healthcare.

Aims & Methods

We examined 610 consecutive individuals aged ≥50 years with no known history of chronic liver disease who were referred for a colonoscopy for colorectal cancer screening. The protocol included sonographic assessment of liver steatosis and point shear wave elastography (pSWE) for liver fibrosis quantification, anthropometric measurements, biochemistry, medical history and alcohol intake evaluation. From the collected data, non-invasive fibrosis scores were calculated and used for aetiological and risk stratification.

Results

A total of 543 patients were included in the final analysis, after excluding patients with an invalid pSWE measurement (5.3%, n=29) and then excluding patients with incomplete results (n=38). The median age was 64 years (IQR 57-70) and 54.0% were male. The median BMI was 28.7 kg/m² (IQR 25.8–31.4) in males and 28.3 kg/m² (IQR 24.2-31.5) in females. SLD was identified in 63.4 % (n=344) of participants. Elevated liver stiffness (pSWE ≥ 1.30 m/s) was found in 26.9 % (n=146), and 4.8% (n=26) had values ≥1.70 m/s, suggestive of compensated advanced chronic liver disease. Among patients with pSWE ≥ 1.30 m/s, MASLD was the predominant aetiology (69.9%, n=102), followed by MetALD (15.8%, n=23), ALD (1.4%, n=2), cryptogenic (11.6%, n=17) and other etiologies (1.4%, n=2). Preselection using EASL-EASD-EASO criteria (i.e. presence of type 2 diabetes or central obesity plus ≥1 cardiometabolic risk factor and elevated liver enzymes) would have detected 365 at-risk individuals (67.2% of the cohort) of whom 55.9% (n=204) had a FIB-4 index ≥1.30. This approach would have identified only 30.4% (n=38) of all the patients with MASLD/MetALD with elevated liver stiffness (n=125).

Conclusion

Opportunistic screening for SLD in patients undergoing screening colonoscopy could be a useful strategy for detecting asymptomatic and potentially progressive liver disease. This approach identified a high prevalence of SLD (63.4%) and elevated liver stiffness (26.9%) in this population, primarily due to MASLD and MetALD. The integration of ultrasound and elastography added minimal burden to the existing workflow while enabling early risk stratification without the need for additional patient visits or referrals.

References

1 Tacke F, Horn P, Wong VWS, Ratziu V, Bugianesi E, Francque S, et al. EASL–EASD–EASO Clinical Practice Guidelines on the management of metabolic dysfunction-associated steatotic liver disease (MASLD). Journal of Hepatology [Internet]. 2024 Jun 7;81(3):492–542.
2 Lazarus JV, Miralles-Sanchez JE, Agirre-Garrido L, Brachowicz N, Nicolàs A, Armenta A, et al. A call to action to address the steatotic liver disease public health threat in Barcelona. The Lancet Regional Health - Europe [Internet]. 2025 Mar 22;52:101272.
3 Barr RG, Wilson SR, Rubens D, Garcia-Tsao G, Ferraioli G. Update to the Society of Radiologists in Ultrasound Liver Elastography Consensus Statement. Radiology [Internet]. 2020 Jun 9;296(2):263–74.

Introduction

Endoscopy is essential for the early detection and management of upper gastrointestinal (UGI) tumors, yet variability in diagnostic expertise and workload-related fatigue often compromise the quality of report documentation. These challenges underscore the urgent need for innovative technological approaches to assist endoscopists in optimizing the report generation process.
Multimodal large language models (MLLMs) enable the synergistic integration of visual and semantic representations, demonstrating significant clinical value in medical text generation. This study introduces MERGS, an innovative endoscopy report generation system powered by MLLM. MERGS annotates medical terminology and generates natural language descriptions based on endoscopic images, ultimately producing structured reports that conform to clinical standards.

Aims & Methods

MERGS is built on the pre-trained MLLM Qwen-VL-2.5-7B. It adopts a multimodal architecture integrating visual encoding and cross-modal alignment to enable medical term annotation and natural language generation from endoscopic images. A pathological classification module further enhances lesion characterization. The language model synthesizes image-derived descriptions to generate a comprehensive report with a diagnostic conclusion.
UGI endoscopy reports were collected from the Sun Yat-sen University Cancer Center between 2019 and 2021. A standardized terminology dictionary was developed based on clinical guidelines, supporting a three-tier annotation framework: endoscopic image → standardized medical terminology → natural language text. Model-assisted pre-annotation combined with expert validation substantially reduced manual effort. The resulting dataset was split into training and test sets at an 8:2 ratio.
Model performance was assessed via multi-label classification metrics (accuracy, specificity, F1-score), as well as semantic similarity measures (Jaccard index) to evaluate its ability to recognize anatomical sites and endoscopic features. Natural language generation quality was assessed using BLEU-2 and ROUGE-L scores, and expert satisfaction ratings were collected for 100 randomly sampled outputs to assess readability and clinical relevance. A human–AI comparison was conducted on 50 test reports, comparing system-generated outputs with those authored by experienced endoscopists.

Results

A multimodal dataset was constructed after quality control, comprising 22,373 reports and 135,503 associated images from 21,619 patients. Internal validation demonstrated that MERGS achieved robust performance in automated image interpretation, with 97.3% accuracy and 98.6% specificity for standard anatomical site recognition. In feature identification, it showed high clinical relevance (F1 scores: smooth mucosa 0.915, polyp 0.857, tumor 0.878, erythematous mucosa 0.766).
Generated reports showed strong alignment with reference reports (BLEU-2 = 0.587, ROUGE-L = 0.661) and received high expert satisfaction ratings (mean score: 4.34 ± 0.63 on a 5-point scale). In a human–AI comparison, MERGS-generated reports were considered non-inferior to those authored by endoscopists (scores: MERGS 4.52 ± 0.81 vs. manual 4.68 ± 0.72), with the upper limit of the one-sided 95% CI (0.37) below the non-inferiority margin (0.5, p = 0.0045).

Conclusion

This study developed an automated system for UGI endoscopy reporting powered by an MLLM. Both internal and clinical validation demonstrated the system's high performance in endoscopic image feature recognition and report generation, underscoring its strong potential for clinical deployment.

References

Luo H, Xu G, Li C, He L, Luo L, Wang Z, Jing B, Deng Y, Jin Y, Li Y, Li B, Tan W, He C, Seeruttun S R, Wu Q, Huang J, Huang D W, Chen B, Lin S B, Chen Q M, Yuan C M, Chen H X, Pu H Y, Zhou F, He Y, Xu R H. Real-time artificial intelligence for detection of upper gastrointestinal cancer by endoscopy: a multicentre, case-control, diagnostic study[J]. Lancet Oncol, 2019,20(12):1645-1654.

Introduction

Upper gastrointestinal bleeding (UGIB) remains a common and potentially life-threatening condition in emergency medicine. Early identification of patients requiring urgent esophagogastroduodenoscopy (EGD) is crucial to improve outcomes and optimize healthcare resources management. Traditional risk stratification tools, such as the Glasgow-Blatchford Score (GBS), have been widely used to guide clinical decision-making. However, the growing availability of artificial intelligence (AI) offers new perspectives in diagnostic and therapeutic strategies. Recent advancements in large language models (LLMs), such as ChatGPT® and Google GEMINI®, suggest potential applications in clinical triage.

Aims & Methods

This study aims to evaluate the performance of these AI tools in determining the appropriate timing for endoscopy in UGIB patients, compared to standard clinical scoring systems.
The medical records of 100 patients who presented to the emergency department with upper gastrointestinal bleeding (UGIB) between April and September 2024 were reviewed retrospectively. Each case was evaluated by two AI models: Google GEMINI® (September 2024 version) and ChatGPT® (version 3.5). These models, acting as virtual gastroenterology experts, were asked to recommend whether urgent or delayed esophagogastroduodenoscopy (EGD) was necessary. For each patient, clinical data such as age, medical history, clinical examination findings, and laboratory results at admission were provided. The AI recommendations were recorded in an Excel file and compared with the Glasgow-Blatchford Score (GBS), which was calculated by a gastroenterologist at the time of admission.
An urgent endoscopy was deemed necessary for a GBS ≥ 3. Statistical analysis was performed using Jamovi® software (version 2.3), and the agreement between the AI models and human evaluations was assessed using Cohen’s kappa coefficient (κ).

Results

The epidemiological characteristics of the patient cohort revealed a median age of 63 years (range: 37.5 to 78), with a male-to-female ratio of 1.7. Among the patients, 73% had at least one comorbidity, including hypertension (33%), diabetes (22%), ischemic heart disease (17%), and chronic liver disease (12%). At the time of admission, melena was the predominant symptom, reported in 78% of patients. The average Glasgow-Blatchford Score (GBS) was 7.07 ± 3.47.
In terms of AI recommendations, ChatGPT suggested urgent endoscopy in 48% of cases, while Google GEMINI recommended it in 49%. A moderate level of agreement was observed between the two AI models for urgent endoscopy recommendations (kappa = 0.540, p < 0.001). However, the agreement between the AI models and the human evaluator was weak. The level of agreement between the gastroenterologist’s assessment and ChatGPT’s recommendation was weak (kappa = 0.222, p < 0.001), with similar results for Google GEMINI (kappa = 0.231, p < 0.001).

Conclusion

Artificial intelligence is a promising tool that is already beginning to transform various fields of medicine. However, our study demonstrates that the publicly available AI models currently in use are not yet capable of effectively triaging patients in emergency settings, particularly in cases of upper gastrointestinal bleeding. A model specifically trained on clinical gastroenterology cases, using validated scientific guidelines, could produce more accurate results and potentially revolutionize patient triage in emergency departments.