Introduction

Deep learning systems in gastrointestinal endoscopy are traditionally pretrained on large, diverse datasets comprising generic, non-medical images before being trained on scarce endoscopic data. Foundation models are models pretrained based on domain-specific datasets. For endoscopy, foundation models require large datasets of general endoscopic images. Yet datasets for developing such models remain limited. In this study, we present GastroNet-5M, a dataset comprising 4,820,653 endoscopic images of approximately 500,000 procedures.

Aims & Methods

GastroNet-5M is composed of anonymized endoscopic images collected from eight Dutch hospitals between 2012 and 2020. Using a self-supervised learning approach, this dataset enabled the development of a foundation model for various endoscopic AI applications. In this study, we compared GastroNet-5M-pretrained models against current state-of-the-art models with traditional pretraining approaches and all publicly available endoscopic foundation models.
We evaluated diagnostic performance on 9 classification and 9 segmentation tasks, including Barrett’s neoplasia detection, colorectal polyp characterization, and gastric cancer invasion depth prediction. All models were trained from scratch for each task. Data efficiency was assessed by repeating all classification tasks with progressively smaller training sets (i.e. 100%, 50%, 25%, 10% and 5%). Robustness was evaluated across 12 datasets reflecting data heterogeneity such as variation in endoscope manufacturer, imaging modality (e.g., virtual chromoendoscopy), and image quality.

Results

GastroNet-5M-pretrained models outperformed all benchmarks in 8 of 9 classification tasks, with an average AUC improvement of 6.8% (Table 1). For segmentation, they achieved higher Dice scores compared to all other models across all 9 tasks, with an average gain of 15.4%. In data efficiency experiments, GastroNet-5M models achieved the highest AUC in 5 of 9 tasks when training data was limited, while no other model led in more than one. In robustness testing, they outperformed other models in 8 of 12 tasks evaluating heterogeneous data conditions.

Table 1. Area-Under-the-Curve scores of the GastroNet-5M-pretrained model and benchmark models for all classification tasks.

Conclusion

GastroNet-5M, a multicenter dataset of over 5 million unlabeled endoscopic images, offers a valuable resource for pretraining deep learning models in endoscopy. The use of GastroNet-5M enhances diagnostic accuracy, reduces the need for scarce application-specific endoscopic imagery, and improves robustness of endoscopic AI systems to the inevitable data heterogeneity in clinical practice. GastroNet-5M is publicly accessible for further research and development.

References

1. Sanderson E, Matuszewski BJ. arXiv preprint arXiv:2401.06278. 2024.
2. van der Putten J, de Groof J, Struyvenberg M, et al. Artif Intell Med. 2020;107.
3. Wang Z, Liu C, Zhang S, et al. arXiv preprint arXiv:2306.16741. 2023.
4. Liu Z, Mao H, Wu C-Y, et al. arXiv preprint arXiv:2201.03545. 2022.
5. Dosovitskiy A, Beyer L, Kolesnikov A, et al. arXiv preprint arXiv:2010.11929. 2020.
6. Oquab M, Darcet T, Moutakanni T, et al. arXiv preprint arXiv:2304.07193. 2023.
7. Jocher G, Qiu J. Ultralytics YOLOv11, v11.0.0. 2024.
8. de Groof AJ, Struyvenberg MR, van der Putten J, et al. Gastroenterology. 2020;158(4):915–29.e4.
9. Fockens KN, Jukema JB, Boers T, et al. United European Gastroenterol J. 2023;11(4):324–36.
10. Jukema J, Kusters C, Jong M, et al. Endoscopy (ESGE Days). 2023.
11. GIANA Challenge. Available from: https://giana.grand-challenge.org
12. Borgli H, Thambawita V, Smedsrud PH, et al. Sci Data. 2020;7(1):283.
13. Polat G, Kani HT, Ergenc I, et al. Inflamm Bowel Dis. 2023;29(9):1431–9.
14. van der Zander QEW, Schreuder RM, Fonollà R, et al. Endoscopy. 2021.
15. Tang D, Zhou J, Wang L, et al. Front Oncol. 2021;11:654093.
16. Houwen BB, Hazewinkel Y, Giotis I, et al. Endoscopy. 2022;54(Suppl 1):eP133.

Introduction

In colorectal epithelial neoplasms, lesions classified as type 2B according to the Japan Narrow Band Imaging (NBI) Expert Team classification (JNET), which are characterized by irregular microsurface or microvascular patterns, are often difficult to diagnose accurately using NBI magnifying endoscopy (ME) alone. Therefore, supplementary pit pattern diagnosis using crystal violet (CV)-stained ME (CV-ME) is recommended. However, CV is potentially carcinogenic, and CV-ME is relatively complex and time-consuming. Acetic acid (AA)-enhanced NBI ME (ANBI-ME) has been proposed as a simpler and safer alternative. Based on our previous study identifying 4.5% AA as the optimal concentration[1], this study prospectively evaluated the diagnostic utility of 4.5% ANBI-ME as a substitute for CV-ME (UMIN registration number: 00045621).

Aims & Methods

From October 2021 to March 2024, colorectal lesions classified as JNET type 2B or 3 during colonoscopy were prospectively enrolled until 300 cases were included. Each lesion underwent ANBI-ME with 4.5% AA, followed by CV-ME. The area of suspected deepest invasion was selected, and one representative image from ANBI-ME and one from CV-ME were independently assessed by two expert endoscopists blinded to histopathology. Intermethod and interobserver agreement were evaluated using Cohen’s weighted kappa. Diagnostic accuracy for each modality was calculated using the individual and consensus diagnoses of the two experts (McNemar’s test). AA whitening duration (AD), representing a reversible mucosal reaction, was measured during ANBI-ME and analyzed by histological depth (low-grade dysplasia [LGD], high-grade dysplasia [HGD], superficial submucosal invasive carcinoma [SMs], and deep submucosal invasive carcinoma [SMd]) using Kaplan–Meier analysis and the log-rank test.

Results

After excluding four unresected lesions, 296 lesions were analyzed. Median lesion size was 18 mm. Locations included the cecum (30), ascending colon (47), transverse colon (69), descending colon (13), sigmoid colon (73), and rectum (64); 122 lesions were elevated and 174 flat. Histological diagnosis included 77 LGD, 145 HGD, 18 SMs, and 56 SMd. Good agreement was observed between modalities (κ = 0.65 and 0.62) and between observers (κ = 0.78 for ANBI-ME, κ = 0.66 for CV-ME). Consensus was reached in 216 ANBI-ME and 192 CV-ME cases. Diagnostic accuracy was significantly higher with ANBI-ME than CV-ME for both observers (81% vs. 76%, p = 0.028; 75% vs. 69%, p = 0.025). Among consensus cases, ANBI-ME also showed higher accuracy (66% vs. 57%, p < 0.01). Median AD (in seconds) was 52 for LGD, 80 for HGD, 132 for SMs, and 107 for SMd, with a significantly shorter duration observed in LGD (p < 0.001).

Conclusion

ANBI-ME demonstrated diagnostic performance equivalent to or superior to CV-ME for colorectal lesions classified as JNET type 2B or 3. In addition, AA whitening duration (AD) may complement the assessment of histological depth of invasion.

References

[1]. Shimizu T, Yamagata T, Kanno Y, et al. A single-center pilot study on the efficacy and safety of acetic acid-enhanced magnifying endoscopy with narrow-band imaging for diagnosis of colorectal epithelial neoplasms. Gastrointest Endosc. 2025;101(2):416-424.

Introduction

Post-ESD surveillance guidelines remain heterogeneous, with many centers favoring early surveillance colonoscopy (SC) despite low recurrence rates after curative resection. This study aimed to evaluate the safety and outcomes of delayed SC after R0 colorectal ESD for benign lesions and to identify risk factors for metachronous advanced neoplasia.

Aims & Methods

We retrospectively analyzed 1,097 patients who underwent colorectal ESD between 2013 and 2021. Clinical, procedural, and histological characteristics were recorded. Among 778 patients with available surveillance data, outcomes were compared between those who had early SC (<12 months) and those with delayed SC (>24 months). Rates of recurrence, advanced neoplasia, and colorectal cancer were assessed. Risk factors for metachronous advanced neoplasia were evaluated via univariate and multivariate analysis.

Results

Mean patient age was 74.3 ± 10.8 years; 54.1% were male. En bloc and R0 resection rates were 95.8% and 83.3%, respectively. Among patients with surveillance data, 82.6% underwent early SC, and 17.4% had late-only SC. The rate of advanced adenoma at first SC was low and comparable between early and late groups (3.9% vs. 6.7%, p = 0.167). The overall rate of metachronous advanced adenoma was 6.5%, with no significant difference between early and late strategies (6.1% vs. 8.9%, p = 0.25). All four colorectal cancers were detected in the early group; none occurred in the late-only group. Recurrence on the ESD scar was rare (1.5% overall), and no late recurrences occurred after a normal SC1. The only significant predictor of metachronous advanced neoplasia was the number of synchronous lesions at index colonoscopy (OR 1.16, 95% CI 1.07–1.27; p < 0.01).

Conclusion

Delayed surveillance colonoscopy 3 years after R0 colorectal ESD for benign lesions appears safe and effective in selected patients. The multiplicity of synchronous lesions could guide risk stratification. These findings support extending SC intervals after curative ESD, with potential to reduce unnecessary procedures.

References

Pimentel-Nunes P, Libânio D, Bastiaansen BAJ, Bhandari P, Bisschops R, Bourke MJ, et al. Endoscopic submucosal dissection for superficial gastrointestinal lesions: European Society of Gastrointestinal Endoscopy (ESGE) Guideline - Update 2022. Endoscopy. juin 2022;54(6):591‑622.
Dai M, Xiao X, Guo CLT, Lui RN, Yip HC, Chu S, et al. The Long‐Term Risk of Metachronous Advanced Adenoma Recurrence After Endoscopic Submucosal Dissection for Colorectal Neoplasia: A Propensity‐Score Matched Longitudinal Cohort With 5‐Year Follow‐Up. United Eur Gastroenterol J. 21 déc 2024;13(2):210‑9.
Yoshida N, Naito Y, Siah KTH, Murakami T, Ogiso K, Hirose R, et al. High incidence of metachronous advanced adenoma and cancer after endoscopic resection of colon polyps ≥20 mm in size. Dig Endosc Off J Jpn Gastroenterol Endosc Soc. mars 2016;28(2):194‑202.

Introduction

Etrasimod is an oral, once-daily, selective sphingosine 1-phosphate (S1P)_1,4,5 receptor modulator for the treatment of moderately to severely active ulcerative colitis (UC), with demonstrated efficacy in patients (pts) with baseline modified Mayo Score (MMS) 4–9.¹ No prior clinical trials of advanced therapies in UC have studied efficacy only in pts with baseline MMS ≤6.

Aims & Methods

GLADIATOR (NCT04607837) was a proof-of-concept, global, randomised, double-blind, placebo-controlled trial to evaluate the efficacy and safety of etrasimod in adults with mildly to moderately active UC (MMS 4–6; endoscopic subscore ≥2; rectal bleeding subscore ≥1) and inadequate response, loss of response or intolerance to ≥1 UC therapy. Pts were randomised 2:1 to etrasimod 2 mg or placebo once daily, stratified by previous biologic/Janus kinase inhibitor exposure and baseline corticosteroid use. GLADIATOR used a treat-through design (12-week [wk] induction period followed by a 40‑wk maintenance period). From Wk 12, pts whose disease had worsened or not improved vs baseline may have been eligible to enrol in an open-label extension trial (NCT03950232). The primary efficacy endpoint was the proportion of pts achieving clinical remission at Wk 52; additional endpoints are shown in the Table. Safety was monitored throughout the study.

Results

In the primary analysis set for efficacy, 127 and 60 pts received etrasimod and placebo, respectively. At Wk 52, 26.0% and 18.3% of pts receiving etrasimod and placebo, respectively, were in clinical remission (p=0.2524; Table). At Wk 12, a greater proportion of pts receiving etrasimod achieved clinical remission, endoscopic improvement (EI) and endoscopic improvement-histologic remission (EIHR) vs placebo (p<0.05; Table). Pts receiving etrasimod were more likely to achieve sustained clinical remission vs placebo (p=0.0104; Table). Etrasimod was generally well tolerated and findings were consistent with the known safety profile.

Conclusion

In pts with mildly to moderately active UC, etrasimod demonstrated higher rates of clinical remission, EI and EIHR at Wk 12, as well as sustained clinical remission at Wk 52, vs placebo. However, the primary endpoint of clinical remission at Wk 52 was not met. Etrasimod demonstrated efficacy in pts with UC with an MMS 4–9, including those with an MMS 4–6, in the ELEVATE UC clinical programme.¹ The proof-of-concept GLADIATOR trial further supports etrasimod’s efficacy in active UC and provides important insights on studying pts with mildly to moderately active UC.

References

1. Sandborn WJ et al. Lancet 2023; 401: 1159–1171.

Disclosure

SD: Lecture fees: AbbVie, Amgen, Ferring Pharmaceuticals Inc, Gilead, Janssen, Mylan, Pfizer Inc, Takeda; Consultancy fees: AbbVie, Allergan, Amgen, AstraZeneca, Biogen, Boehringer Ingelheim, Celgene, Celltrion, Ferring, Gilead Sciences, Hospira, Janssen, Johnson&Johnson, MSD, Mundipharma, Pfizer Inc, Roche, Sandoz, Takeda, TiGenix, UCB, Vifor; Directorship: Gastroenterology and Endoscopy.
MG, KL: Employees of Pfizer AG; shareholders of Pfizer Inc.
LPB: Consultancy fees: AbbVie, Abivax, Alimentiv, Alma Bio Therapeutics, Amgen, Applied Molecular Transport, Arena, Biogen, BMS, Celltrion, CONNECT Biopharm, Cytoki Pharma, Enthera, Ferring, Fresenius Kabi, Galapagos, Genentech, Gilead, Gossamer Bio, GlaxoSmithKline, HAC-Pharma, IAG Image Analysis, Index Pharmaceuticals, Inotrem, Janssen, Lilly, Medac, Mopac, Morphic, MSD, Norgine, Novartis, OM Pharma, ONO Pharma, OSE Immunotherapeutics, Pandion Therapeutics, Par’Immune, Pfizer Inc, Prometheus, Protagonist, Roche, Roivant, Samsung, Sandoz, Sanofi, Takeda, Theravance, Thermo Fisher, Tigenix, Tillotts, Viatris, Vifor, Vectivbio, Ventyx, Ysopia; Grants: Takeda, Fresenius Kabi, Celltrion; Lectures: AbbVie, Amgen, Arena, Biogen, Celltrion, Ferring, Galapagos, Genetech, Gilead, Janssen, Lilly, Medac, MSD, Pfizer Inc, Sandoz, Takeda, Tillotts, Viatris, Vifor.
AJY: Consultancy fees: AbbVie, Arena, BMS, Celltrion, Pfizer Inc, Takeda; Lectures: AbbVie, BMS.
MCD: Consultancy fees: AbbVie, Abivax, Arena Pharmaceuticals, AstraZeneca, BMS, Eli Lilly, Galapagos, Genentech, Gilead Sciences, Janssen Pharmaceuticals, Johnson&Johnson, Merck, Pfizer Inc, Prometheus Laboratories, Prometheus Biosciences, Sanofi, Sphyre, Takeda, UCB; Shareholder: Trellus Health; Directorship: Trellus Health.
DTR: Consultancy fees: AbbVie, Abivax SA, Altrubio, Avalo Therapeutics, Bausch Health, Bristol Myers Squibb, Buhlmann Diagnostics Corp, Celltrion, ClostraBio, Connect BioPharma, Douglas Pharmaceuticals, Eli Lilly & Co., Foresee, Genentech, Image Analysis Group, InDex Pharmaceuticals, Iterative Health, Janssen Pharmaceuticals, Odyssey Therapeutics, Pfizer Inc, Sanofi, Takeda Pharmaceuticals, Throne, Vedanta; Grants: Takeda.
BGF: Senior Scientific Director of Alimentiv Inc, which provides central reading services. He is not a company employee and has no equity stake in the organisation, which is owned by a medical trust; Speaker fees: AbbVie, Janssen, Takeda; Consultancy fees: AbbVie, AbolerIS, AgomAB Therapeutics, Allianthera, Amgen, AnaptysBio, Applied Molecular Transport Inc, Arena Pharma, Avoro Capital Advisors, Atomwise, BioJamp, Biora Therapeutics, Boehringer Ingelheim, Boxer, Celsius Therapeutics,Celgene/BMS, Connect BioPharma, Cytoki, Disc Medicine, Duality, EcoR1, Eli Lilly, Equillium, Ermium, First Wave, First Word Group, Galapagos, Galen Atlantica, Genentech, Gilead, Gossamer Pharma, GSK, Hinge Bio, Hot Spot Therapeutics, Index Pharma, Imhotex, Immunic Therapeutics, JAKAcademy, Janssen, Japan Tobacco Inc, Kaleido Biosciences, Landos Biopharma, Leadiant, L.E.K. Consulting, Lenczner Slaght, LifeSci Capital, Lument AB, Millennium, MiroBio, Morgan Lewis, Morphic Therapeutics, Mylan, OM Pharma, Origo BioPharma, Orphagen, Pandion Therapeutics, Pendopharm, Pfizer Inc, Prometheus Therapeutics and Diagnostics, Play to Know AG, Progenity, Protagonist, PTM Therapeutics, Q32 Bio, Rebiotix, REDX, Roche, Sandoz, Sanofi, Seres Therapeutics, Silverback Therapeutics, Surrozen Inc, Takeda, Teva, Thelium, Tigenix, Tillotts, Ventyx Biosciences, VHSquared Ltd, Viatris, Ysios, Ysopia, Zealand Pharma; Shareholder: Gossamer Pharma.
GRDH: Advisory fees: AbbVie, AstraZeneca, Alimentiv, BMS, Boehringer Ingelheim, Celltrion, Cosmo, Eli Lilly, Galapagos, GSK, Johnson&Johnson, Takeda, Pfizer Inc, Polpharma, Prometheus Biosciences, Tillotts, Ventyx; Speaker fees: AbbVie, Boehringer Ingelheim, Celltrion, Eli Lilly, Johnson&Johnson, Takeda, Pfizer Inc, Tillotts.
FB: Research/grants: AbbVie, Amgen, Eurogenerics, Janssen, Takeda; Lectures: AbbVie, Arena Pharmaceuticals, Celltrion, Ferring, Galapagos, Janssen, Merck Sharp & Dohme, Pfizer Inc, Takeda; Consultancy fees: AbbVie, Abivax, Amgen, Arena Pharmaceuticals, Celgene, Celltrion, Ferring, Fresenius Kabi, Janssen, Merck Sharp & Dohme, Pfizer Inc, Sandoz.
GG, JY, NL, CZ, WW, SM, EHL, IM, SS, WN: Employees/shareholders of Pfizer Inc.
SV: Lectures: AbbVie, Dr. Falk Pharma, Ferring, Galapagos, Hospira, Janssen, MSD, Takeda, Tillotts; Consultancy fees: AbbVie, AbolerIS Pharma, Alimentiv, Arena, AstraZeneca, Avaxia, Bristol Myers Squibb, Boehringer Ingelheim, Celgene, CVasThera, Dr Falk Pharma, Eli Lilly, Ferring, Galapagos, Genentech/Roche, Gilead, Hospira, Imidomics, Janssen, Johnson&Johnson, Materia Prima, MiroBio, Morphic, MrMHealth, MSD, Mundipharma, Pfizer Inc, Prodigest, Progenity, Prometheus, Robarts Clinical Trials, Second Genome, Shire, Surrozen, Takeda, Theravance Biopharma, Tillotts Pharma AG, Zealand Pharma; Grants: AbbVie, MSD, Pfizer Inc, Galapagos, Janssen, Takeda.

Introduction

Crohn’s disease (CD), as a type of inflammatory bowel disease (IBD), is characterized by chronic inflammation in the intestinal tract and its incidence is rising globally. The factors involved in causing or triggering Crohn’s disease (CD) have yet to be confirmed. We recently demonstrated that the intestinal microbiome composition is associated with the risk of CD development; however, its potential causal effect remains unknown. Our study investigated whether stool microbiome might contribute to or trigger dysregulated immune response by transplanting stool into germ-free mice (GFM).

Aims & Methods

Fecal samples were collected from 12 discordant pairs of siblings (one who did not develop CD referred to as Healthy Matched Control (HMC) and another who subsequently developed CD, referred to as pre-CD). These subjects were recruited as part of the Crohn's and Colitis Canada Genetic, Environmental, Microbial Project (CCC-GEM) project, a cohort of prospectively followed healthy first-degree relatives. The stool was transplanted into RAG1-/- GFM before T-cell transfer-induced colitis (3 transplanted mice per donor). Body weight was monitored for 6 weeks after T-cell transfer. Fecal Lipocalin-2 (LCN-2) was measured by ELISA to assess gut inflammation. Intestinal tissue damage was assessed by a validated histology scoring. Pro-inflammatory cytokine gene profiles of intestinal tissue were measured by qPCR. The feces of colitic mice were collected after 6 weeks of T-cell transfer and used to investigate metabolic profile measured by UHPLC-MS/MS.

Results

At week 2 after T-cell transfer, mouse recipients of pre-CD stool experienced greater weight loss than mice receiving stool from HMC (p=0.004). LCN-2 of pre-CD mice was significantly higher after T-cell transfer at week 3. Intriguingly, pre-CD stool recipient mice had a higher histological damage score in the ileum (p=0.031). Mice receiving pre-CD stool had increased expression of selected pro-inflammatory cytokine genes in the ileum including Tnfα, Il-1β, Ifnγ, Il-10 and Tgfβ compared to mice that received HMC stool. Colonic tissue showed significant increases in TNFα and Tgfβ in the mice recipient of the pre-CD stool. Moreover, pre-CD stool led to significant changes in 40 fecal metabolites compared to HMC microbiome (0.01<p≤0.05). Spearman correlation between certain metabolites and colitis markers in mouse was observed. Fecal metabolomics in humans revealed two shared metabolic pathways associated with future onset of CD and murine inflammatory markers. Notably, there was a positive correlation between sphingolipids in humans and mice that was associated with increased colitis in mice. Sphingolipid metabolites also demonstrated positive correlation with mouse colitis marker and human CD subclinical markers.

Conclusion

These results showed that transplanted pre-CD stool worsened T cell transfer colitis in mice compared to HMC stool. This suggests that stool microbiome of individuals with CD is functionally altered years before diagnosis, exhibiting greater inflammatory potential when transferred to susceptible mice. Modulating pre-CD microbiome composition and/or function could lead to novel methods of CD prevention.