Artificial intelligence (AI) is increasingly entering colonoscopy practice, with computer-aided detection (CADe) systems improving polyp and adenoma detection. However, the next challenge is not only to detect more lesions, but to determine in real time which lesions require resection, histopathological assessment, surveillance adjustment or surgical referral. Computer-aided diagnosis (CADx) systems aim to support this step by predicting colorectal polyp histology before resection and enabling optical diagnosis strategies such as “resect-and-discard” and “diagnose-and-leave”.The clinical value of CADx should not be judged by diagnostic accuracy alone. To become clinically meaningful, CADx must safely guide management decisions, meet established ASGE and ESGE thresholds, integrate into real-time workflow and remain valid across endoscopy platforms, imaging modalities, lesion subtypes and operator expertise. Although systems such as CAD EYE, GI Genius, POLAR and endocytoscopy-based algorithms show promising diagnostic performance, recent meta-analyses suggest that CADx has not yet provided clear incremental benefit for “diagnose-and-leave” or “resect-and-discard” strategies when added to endoscopist assessment. This may reflect high baseline confidence among endoscopists, but also current limitations including binary classification schemes, inconsistent handling of sessile serrated lesions, lack of calibrated confidence scores and limited explainability.Beyond diminutive-polyp characterisation, AI is also being explored for invasion-depth prediction in larger or suspicious colorectal lesions, where incorrect predictions may lead to undertreatment or overtreatment. Moreover, AI-assisted colonoscopy remains dependent on high-quality mucosal exposure, adequate bowel preparation, careful inspection and trained endoscopists. Overall, CADx remains promising, but broader implementation requires prospective real-world validation, explainable and interoperable systems, robust human-AI interaction and clinically relevant outcomes before it can safely substitute histopathology in selected settings.

OBJECTIVES:To quantify the safety impact of using Microstream capnography monitoring during procedural sedation for gastrointestinal procedures and bronchoscopy and to identify a risk-stratification method. METHODS:A quality improvement initiative at hospitals in Belgium, Canada, Spain, Turkey, and the UK included 6734 procedures in adults undergoing procedural sedation, who received either Microstream capnography monitoring added to standard monitoring, or standard monitoring only. The primary outcome was the incidence of any intraoperative adverse event. Incidence of any severe adverse event was assessed separately. American Society of Anesthesiologists classification, sedatives, procedure time, and procedure type were used to adjust odds ratios (OR) derived from logistic regressions for adverse events and applied in risk stratification. RESULTS:Adverse event incidence decreased from 10.8% standard monitoring to 6.7% using capnography, with a number needed to treat (NNT) of 24. The adjusted ORs were 0.54 (95% confidence interval (CI): 0.44-0.66, P  < 0.001) for any adverse event and 0.46 (95% CI: 0.26-0.83, P  = 0.01) for any severe adverse event. Procedures were stratified into low- ( N  = 1292, 25.6%), medium- ( N  = 2383, 47.2%), and high-risk ( N  = 1376, 27.2%) groups. Capnography monitoring significantly decreased adverse events in all risk groups with absolute risk reductions (NNT) of 3.8% (26), 4.6% (22), and 6.7% (15) for low- to high-risk groups, respectively. Severe adverse events significantly decreased with capnography in the high-risk group only (3.5% vs. 0.9%; OR 0.25, 95% CI: 0.10-0.60; NNT = 38). The risk stratification model should be considered exploratory and requires further validation. CONCLUSION:Capnography in procedural sedation significantly reduced both any and severe adverse events and should benefit patients in routine practice. We recommend considering American Society of Anesthesiologists level, sedation type, and planned procedure time in future risk stratification evaluations.

BACKGROUND:Artificial intelligence (AI)-based computer-aided detection (CADe) systems improve adenoma detection in average-risk colorectal cancer screening. Meanwhile, evidence in Lynch syndrome surveillance is sparse and inconsistent. We assessed the effect of CADe on adenoma detection during Lynch syndrome surveillance. Computer-aided optical diagnosis (CADx) performance for optical differentiation of colorectal lesions was evaluated as a secondary aim. METHODS:CADLY2 was an international, multicentre, open-label, randomised controlled superiority trial at nine specialised hereditary cancer surveillance centres in Belgium, Germany, the Netherlands, and Spain. Adults aged 18 years or older with genetically confirmed Lynch syndrome scheduled for surveillance colonoscopy were randomly assigned (1:1) to high-definition white-light (HD-WL) colonoscopy alone or to HD-WL colonoscopy with computer-aided assistance from CAD EYE (Fujifilm, Tokyo, Japan). CAD EYE was used for CADe during withdrawal and for CADx after lesion detection. Randomisation was done centrally through a secure web-based system using Pocock's minimisation algorithm with a stochastic component and was stratified by centre, sex, previous colorectal cancer, underlying pathogenic variant, and interval since previous colonoscopy. Allocation concealment was ensured through the centralised web-based system. Patients were masked to group allocation until the start of withdrawal in procedures with mild sedation, or until completion of the procedure in procedures with propofol-based sedation. Endoscopists were not masked. The primary outcome was adenoma detection rate, defined as the proportion of patients with at least one histopathologically confirmed adenoma, analysed in the full analysis set (defined as all randomly allocated patients with available data for the primary outcome). The diagnostic performance of the CADx system was evaluated as a secondary outcome. The safety analysis set comprised all randomly allocated patients who underwent a study colonoscopy. This study is registered with the German Clinical Trials Register, DRKS00030695, and is completed. FINDINGS:Between May 9, 2023, and Oct 30, 2025, 757 patients were randomly allocated to HD-WL colonoscopy (377 patients) or to AI-assisted colonoscopy (380 patients); 733 patients were included in the full analysis set (369 HD-WL and 364 AI-assisted). The median age was 49 years (IQR 38-59) in the HD-WL group and 50 years (38-59) in the AI-assisted group; 213 (58%) were female and 156 (42%) male in the HD-WL group, and 207 (57%) were female and 157 (43%) male in the AI-assisted group. The adenoma detection rate was 30·9% (114 of 369 patients) with HD-WL versus 33·8% (123 of 364 patients) with CADe assistance (odds ratio 1·14 [95% CI 0·83-1·57], p=0·41). For CADx differentiation of neoplastic versus non-neoplastic lesions in the paired lesion-level analysis, with histopathology as the reference standard and sessile serrated lesions and traditional serrated adenomas classified as non-neoplastic, CADx sensitivity was 85·9% (95% CI 82·0-89·1) and specificity was 91·4% (89·4-93·0). Three adverse events occurred in the AI-assisted group: two mild post-polypectomy bleedings and one serious pulmonary embolism or deep venous thrombosis unrelated to the procedure. No adverse events occurred in the HD-WL group. INTERPRETATION:CADe-assisted colonoscopy did not show the absolute improvement in adenoma detection rate that was assumed in the prespecified sample-size calculation. CADx did not clearly improve lesion differentiation beyond expert optical diagnosis in expert Lynch syndrome surveillance settings. FUNDING:Third-party research funding of the National Center for Hereditary Tumor Syndromes, University Hospital Bonn.

Background Delayed bleeding (DB) after endoscopic mucosal resection (EMR) and endoscopic submucosal dissection (ESD) remains a significant clinical problem. This study aimed to evaluate the safety and efficacy of a new gel hemostatic powder (GHP) for the prevention of DB in high-risk patients. Methods A prospective, multicentric, open-label study was conducted at seven Belgian hospitals from 2023 to 2024 in patients undergoing EMR or ESD for lesions over 20 mm with high risk of DB (either anticoagulant or P2Y12RA treated, or undergoing duodenal EMR). The Nexpowder GHP preventive spray was applied to the resected area at the end of the procedure. The primary outcome was the rate of DB within 4 weeks post-procedure. Secondary outcomes included safety, procedure duration, post-procedural pain, and incidence of adverse events. Results Fifty patients were enrolled in the study and treated with endoscopic resection followed by GHP application. The observed DB rate was 22% (11/50), higher than the literature-reported rate of 16%. The hypothesized reduction in DB below 5% was not achieved ( p = 0.2472). DB was most common in rectal ESD ( n = 3/5; 60%) and duodenal EMR ( n = 6/20; 30%). DB typically occurred within the first 10 days post-procedure (median onset 7d). On univariate analysis, the only factor associated with DB risk was lesion size ( p = 0.042). No adverse events related to the use of the GHP were reported. Conclusion In this prospective study, GHP demonstrated limited efficacy for the prevention of DB in high-risk patients, with a 22% observed DB rate. Lesion size and location were key predictors of DB.

BACKGROUND & AIMS:Peroral endoscopic myotomy (POEM) demonstrated higher efficacy at 1-year follow-up compared with pneumatic dilation (PD) in patients after failed laparoscopic Heller myotomy (LHM). The aim of this study was to evaluate the 5-year efficacy and safety of POEM compared with PD in patients with achalasia after failed LHM. METHODS:This study reports the long-term follow-up results of a previously published multicenter randomized controlled trial performed in the Netherlands, Belgium, and Italy. Adult patients with achalasia confirmed by esophageal manometry were included when they had recurrent symptoms after LHM. Patients were randomized to PD 30 to 35 mm or POEM in a 1:1 ratio. The primary outcome was treatment success at 5-year follow-up, defined as Eckardt score ≤3 without retreatment. Patients who underwent additional dilation once up to 40 mm within 1 year after initial PD were not considered as treatment failure. Intention-to-treat analysis was applied for the primary outcome. RESULTS:In total, 90 patients were included between January 2014 and June 2020 and randomized to PD (n = 45) or POEM (n = 45). At 5-year follow-up, treatment success was 11% after PD (n = 5/45) and 40% after POEM (n = 18/45) with a relative risk of 3.60 (95% confidence interval, 1.46-8.66; P = .002). Median time to treatment failure was 12 months for PD and 60 months for POEM (P = .005). Reflux esophagitis grade B to D was observed in 11% of the patients after POEM (n = 4/37) and in 9% after PD (n = 3/34) (P = .548). No procedure-related serious adverse events were reported between 1-year and 5-year follow-up. CONCLUSIONS:POEM is more effective than PD at 5-year follow-up in patients with achalasia after failed LHM. CLINICAL TRIAL REGISTRY:Dutch Trial Registry (NL4361; NTR4501).

Abstract Topic Esophageal Cancer: Surgical Treatment of Esophageal Cancer – early outcomes and complications Background The etiopathogenesis of anastomotic leakage (AL) after esophagectomy is inadequately understood. In colorectal surgery, AL has been linked with phenotypic switching of commensal bacteria to pathogenic collagenase-producing bacteria. We aimed to explore for the first time if collagenase activity might also play a role in AL after esophagectomy for cancer. Methods Samples were endoscopically collected with cytology brushes at diagnosis of AL (n=4) or delayed conduit emptying (control group,n=3). In AL-patients, anastomoses was first sampled on the side opposite to the AL and second on the AL itself. In control patients two opposite sides were sampled. Collagenase activity was evaluated on a multimode microplate reader using fluorescein-labeled dye quenched collagen by comparing activity at 90-minutes to collagenase from Clostridium histolyticum, and by using Relative Fluorescence Units (RFU) at 12-hours. Samples were grouped according to sampling sequence. Collagenase activity was compared between AL and control patients using the Mann-Whitney U test. Results Patients with AL were sampled on postoperative day 11-24, on average 2.5±2.1 days after diagnosis, and patients in the control group on days 5-7. AL-severity was grade 2-3 and Clavien-Dindo IIIb-IVb. A non-significant trend for higher average collagenase activity (both sampling sites) was found in the AL group versus in the control group (4.60±3.8 and 0.73±0.5 respectively). Differences in average RFU were however significant in this exploratory cohort (37551±13403 in the AL-group,13835±8573 in the control group; p= 0.034). Of note, in patients with AL, the collagenase activity was similar between samples taken at the AL and on the opposite side. Conclusion Patients with AL after esophagectomy for cancer had higher collagenase activity on the anastomosis than patients without AL, mirroring earlier research in colorectal surgery suggesting a bacterial etiopathogenesis.

BACKGROUND:Available achalasia therapies include peroral endoscopic myotomy (POEM), laparoscopic Heller myotomy (LHM), pneumatic dilation (PD), and botulinum toxin injection (BT). Head-to-head trials comparing these therapies are limited. We conducted this network meta-analysis to evaluate the efficacy and safety of these interventions. METHODS:We searched online databases for randomized controlled trials (RCTs) evaluating therapies for achalasia. The primary outcomes were treatment success at 1, 2, and 5 years. Secondary outcomes included the Eckardt score, symptoms of gastroesophageal reflux disease (GERD), erosive esophagitis, and serious adverse events. Risk ratios (RRs) were used in the analysis, using a random-effects model, and the quality of evidence was assessed using the GRADE approach. RESULTS:A total of 14 RCTs with 1171 patients were included. At 1 year, all treatments were more effective than botulinum toxin, and POEM and LHM outperformed PD, with POEM ranked highest (SUCRA 86%). At 2 and 5 years, POEM remained superior to PD.POEM and LHM showed comparable effectiveness across all time points. PD and LHM were associated with a lower risk of GERD than POEM at 5 years: PD versus POEM (RR: 0.35; 95% CI: 0.14-0.89) and LHM versus POEM (RR: 0.49; 95% CI: 0.29-0.84). Erosive esophagitis was more frequent with POEM than PD at 2 years. CONCLUSIONS:POEM and LHM demonstrated comparable success rates, and both had significantly higher success rates than PD at all time points. However, the quality of the evidence was rated low to very low. POEM was associated with higher rates of GERD at 5 years. Future trials should directly compare these modalities using standardized protocols to enable meaningful comparisons, and follow-up beyond 5 years is needed to establish their long-term outcomes.

Artificial intelligence (AI) is rapidly transforming the management landscape of inflammatory bowel disease (IBD). While early applications in endoscopy, digital pathology and cross-sectional imaging drew substantial attention, next-generation AI systems that enable deeper disease understanding, personalized treatment and streamlined clinical workflows are now emerging. These advances encompass the multimodal integration of endoscopic, histological and molecular data ('endo-histo-omics'); AI-assisted assessment of the intestinal barrier; remote monitoring via wearables; and the incorporation of large language models for decision-making support and patient interactions. This Perspective traces the evolution of AI in IBD from domain-specific tools to foundational platforms supporting data-driven precision medicine. We highlight validated AI applications across diagnosis, monitoring, outcome prediction and neoplasia surveillance. We also explore the expectations of key stakeholders, including clinicians, patients, regulatory bodies and industry, and discuss unresolved challenges such as explainability, integration into workflows, reimbursement and environmental sustainability. By aligning innovation with ethical and clinical priorities, AI holds the potential to redefine IBD care. Its future will be shaped by collaboration, transparency and responsible implementation, ushering in a new era of personalized, efficient and equitable care for individuals with IBD.

Post-operative recurrence (POR) of Crohn’s disease (CD) remains a major challenge, affecting up to 80% of patients within 10 years after surgery. Conventional endoscopy struggles to distinguish early inflammatory recurrence from post-surgical ischaemic changes, delaying effective treatment. Virtual chromoendoscopy (VCE) offers precise visualisation of mucosal and vascular changes. The PROSPER study aims to develop and validate a novel endoscopic score integrating advanced imaging features to predict early POR in CD. This analysis is part of the international, multicentre, prospective PROSPER study (NCT06505304). Adult CD patients undergoing ileocolic resection were enrolled pre-operatively or within 3 months post-operatively and underwent endoscopic assessment at 3 or 6 months after surgery, guided by faecal calprotectin levels (cutoff 150 µg/g). Endoscopic videos were obtained using white-light and VCE modalities —TXI, RDI, and NBI with the Olympus (EVIS-X1) platform, and i-scan1 and OE mode 1–2 with the Pentax (INSPIRA) platform— following a standardised protocol. A three-round modified Delphi process defined the PROSPER domains: two in-person consensus meetings and an online survey (Qualtrics platform) where experts rated each feature. Consensus required >75% agreement. Seven core experts participated in both the first (10 videos) and second (15 videos) rounds, scoring the colon proximal to the anastomosis, the anastomotic region (including anastomotic line, ileal inlet, ileal body, and, when present, blind loops), and the neo-terminal ileum. All anastomotic configurations -end-to-end, end-to-side, side-to-end, side-to-side, and Kono-S- were assessed (Fig 1). Inter-rater agreement was evaluated using Fleiss’ κ, Gwet’s AC1/AC2, and ICC(2,k). Mucosal and vascular domains were selected, including erosions, superficial and deep ulcers, villi morphology, vessel dilatation, and bleeding (Tab 1). Agreement across both in-person rounds was moderate to substantial for features in the neo-terminal ileum and colon proximal to the anastomosis (AC1/AC2 = 0.61–0.78) and lower for the anastomotic region (fair to moderate), particularly for villi and erosions (AC2 ≈ 0.40–0.50). Ulcers and vascular patterns showed higher reproducibility (AC1 ≈ 0.70–0.80). Moderate agreement was observed for the anastomosis type (κ 0.51). The Delphi survey confirmed selected domains, with over 80% expert agreement on inclusion and weighting. The novel PROSPER score is a multimodal tool integrating advanced imaging for comprehensive assessment of the anastomosis, showing good inter-expert agreement across most domains. These results support its feasibility for detecting early POR. The next phase will involve real-life validation. Conflict of interest: Iacucci, Marietta: Grant: Pentax, Olympus, Eli lilly, Helmsley Personal Fees: Pentax, Pfitzer, Janssen, EliLilly, J&J Dr. Santacroce, Giovanni: No conflict of interest Zammarchi, Irene: No conflict of interest Nardone, Olga Maria: Advisory board fees from Eli Lilly, Nestlè, Janssen Speaker fees from AbbVie, Janssen, Eli Lilly, Ferring, Alfa Sigma, Recordati, Noòs, and Pfizer Cannatelli, Rosanna: No conflict of interest Doherty, Glen: Research or Education Grants (last 36 months):Abbvie, Pfizer, Janssen, Takeda, Tillotts, Celltrion, Abbott, Dr Falk, Amgen Speaker/Meeting Honoraria (Last 36 months):Abbvie, Dr Falk, Galapagos/Alfa Sigma, GSK Furfaro, Federica: Grant: IG-IBD Personal Fees: Pfizer, Biogen, J&J, Abbvie, Amgen, Janssen Gabbiadini, Roberto: No conflict of interest Lu, Cathy: Advisory board - Abbvie, JnJ, Takeda, Ferring, Merck, Celltrion, Pfizer Research Funding - Abbvie, JnJ Pugliano, Cecilia Lina: No conflict of interest Snir, Yifat: No conflict of interest Ghosh, Subrata: No conflict of interest Tontini, Gian Eugenio: • Speaker honoraria from Pentax (2022), Medtronic (2021), NTC Pharma (2021), Ferring (2024) • Consultant fees from NTC Pharma (2022, 2024), F. Hoffmann-La Roche Ltd (2022), Fujinon (2023), Invicro (2025). Bisschops, Raf: Grant: Pentax Europe, Medtronic, Norgine Personal Fees: Pentax Europe, Fujifilm, Norgine, Medtronic, Ipsen, Alfasigma, Viatris, Endostart, Falk Foundation Non-financial Support: Pentax Europe, Fujifilm, Boston Scientific, Olympus, Erbe, Norgine

Multimodal large language models (MLLMs) can automatically analyze clinical video, but evidence from full esophagogastroduodenoscopy (EGD) and the impact of on-screen computer-aided detection/diagnosis (CAD) overlays on MLLM behavior remain unclear. We tested whether an MLLM can produce clinically adequate EGD reports and whether a CAD overlay changes performance. We analyzed five complete EGD videos with Gemini 2.5 Pro in paired versions: (1) clean video and (2) the same video with a CAD overlay. Five blinded endoscopists rated report adequacy in three domains. MLLM accuracy for landmarks/lesions was further assessed by two blinded expert endoscopists using the time-window rule (a model detection counted as correct if it occurred within ±2 s of the expert-annotated timestamp). In this retrospective pilot study, five archived diagnostic EGD procedures from five patients were available as full-length videos. Across five raters, MLLM Completeness was judged adequate in 56.0% (14/25 ratings) with Clean-Video versus 48.0% (12/25 ratings) with Overlay-Video (p = 0.500). Visualization was identical (36.0% [9/25 ratings] for both; p = 1.000). Lesions characteristics were identical (16.0% [4/25] for both; p = 1.00). For the Landmark agreement, the overall accuracy of the MLLM with Clean-Video vs. Overlay-Video was: 0.55 [95% CI 0.43-0.67] vs. 0.33 [0.23-0.46], p = 0.029; sensitivity 0.53 [0.40-0.66] vs. 0.35 [0.24-0.49], p = 0.122; specificity 0.67 [0.35-0.88] vs. 0.22 [0.06-0.55], p = 0.125. In this pilot study, Gemini 2.5 Pro demonstrated inadequate performance for clinical EGD reporting. These hypothesis-generating findings suggest substantial optimization and larger-scale validation are required before deployment.

BACKGROUND & AIMS:Tools that can predict recurrence in patients with Barrett's esophagus (BE)-related neoplasia treated with endoscopic eradication therapies (EET) to guide surveillance decisions are needed. We aimed to develop and validate a machine learning (ML)-based prediction tool to predict the risk and timing of recurrence status post EET. METHODS:Three prospective United States databases of patients who underwent EET for BE-related neoplasia and achieved complete eradication of intestinal metaplasia (CE-IM; n = 1114) were utilized to develop and internally validate a ML-based prediction tool using the Random Forest model and imputation techniques. Predictors incorporated in this model included demographics, endoscopy and pathology results, and EET details. A Cox proportional hazards model was utilized to predict the time to recurrence. External validation was performed using the United States Radiofrequency Ablation database (n = 1397). RESULTS:BE recurrence occurred in 29.2% (n = 734) of patients and BE-related neoplasia recurrence in 10.6% (n = 265), with a mean time to recurrence of 21.3 months (mean follow-up, 37.7 months). The top predictors for recurrence included BE length, body mass index, age, sessions needed to achieve CE-IM, and baseline histology. The model was well-calibrated, and area under the receiver operating characteristic curve (AUC) was 0.92 (95% confidence interval [CI], 0.85-0.95) on internal validation and 0.91 (95% CI, 0.87-0.95) on external validation for BE recurrence. For BE-related neoplasia recurrence, the AUC was 0.90 (95% CI, 0.88-0.93). The model had moderate discriminative performance to predict timing of recurrence with a C-index of 0.701 at 1 year (AUC 0.71), 0.68 (AUC 0.69) at 3 years, and 0.66 (AUC 0.69) at 5 years. CONCLUSIONS:This United States-based externally validated tool accurately predicts BE and BE-related neoplasia recurrence and timing post EET. This practical tool may help provide a personalized approach to surveillance strategies.