Artificial intelligence in gastrointestinal endoscopy: a global clinical trial analysis

To the Editor, Gastrointestinal (GI) endoscopy constitutes a fundamental diagnostic and therapeutic modality in contemporary gastroenterology, enabling direct visualization and intervention within the digestive tract. Despite substantial technological evolution, conventional endoscopic procedures remain encumbered by inherent limitations, including operator dependence[1], variable detection rates[2], and missed diagnosis rate[3]. These metrics underscore the considerable unmet clinical needs within this domain. Artificial intelligence (AI) has emerged as a transformative paradigm across healthcare specialties, especially gastroenterology, demonstrating exceptional capacity in medical image analysis and pattern recognition[4]. GI endoscopy has a number of well-established AI applications in medicine, notably AI-assisted colorectal polyp detection (CADe) and characterization (CADx), which has gained approval through AI algorithms. Importantly, the clinical efficacy of CADe in colonoscopy is supported by high-quality evidence from multiple randomized controlled trials (RCT)[5]. Despite burgeoning research interest, a comprehensive landscape analysis of the clinical trial on the application of AI in GI endoscopy remains absent. This study aims to characterize the clinical trial landscape of AI applications in GI endoscopy, providing supportive data on the status of this emerging field and its potential translation into clinical practice. This research adheres to the TITAN (Transparency In The reporting of Artificial INtelligence) Guidelines 2025[6]. We conducted a comprehensive search of the ClinicalTrials.gov (https://clinicaltrials.gov/) database to identify all registered clinical trials investigating AI applications in GI endoscopy up to 1 April 2025. We employed the following search strategy: [(Computer-assisted) OR (Computer-aided) OR (Artificial Intelligence) OR (Machine learning) OR (Deep learning) OR (Large Language Models)] AND [(Gastrointestinal endoscopy) OR (Colonoscopy) OR (Duodenoscopy) OR (Balloon Enteroscopy) OR (Esophagoscopy) OR (Gastroscopy) OR (Proctoscopy) OR (Esophagogastroduodenoscopy) OR (Endoscopic ultrasound) OR (Cholangiopancreatography) OR (Capsule Endoscopy) OR (Endoscopic Mucosal Resection) OR (Chromoendoscopy) OR (Narrow band imaging) OR (Magnifying endoscopy) OR (Endoscopic submucosal dissection) OR (optical diagnosis) OR (Blue Light Imaging) OR (Linked Color Imaging) OR (I-SCAN)]. We excluded studies that were purely observational studies, primarily focused on non-GI endoscopic procedures, or did not involve AI applications. A total of 124 clinical trials met the inclusion criteria and were included in the final analysis (Supplemental Digital Content Table S1, available at: https://links.lww.com/JS9/G743). We systematically extracted the following data from each eligible trial: trial title, NCT number, study URL, study status, primary outcome, sponsor, start date, and locations. Additionally, we manually added several measures based on details from each trial:(1) AI system: the specific name or type of AI system employed in each trial; (2) Digestive tract: the anatomical region of investigation, categorized as colorectum, upper GI tract, GI tract, small bowel, or pancreatic; (3) Disease/condition: the specific disease or condition under investigation (e.g., colorectal adenomas, early gastric cancer, and endoscopy quality); (4) Sponsor type: classified as Academic, Industry, or Government based on the primary sponsor affiliation; (5) Study type: categorized as RCT, non-randomized controlled trial (NRCT), or single-arm trial based on study design characteristics; (6) Country: determined according to the primary sponsor’s location; and (7) Center: classified as single- or multi-center based on the number of participating centers. Since this analysis utilized exclusively publicly available data, ethical approval was not required. All data analyzed in this study are freely accessible from ClinicalTrials.gov. Our analysis of trials registered up to 1 April 2025 revealed substantial growth in AI-related GI endoscopy clinical trials over time, with a total of 124 trials included. Trial registrations demonstrated a marked acceleration from merely 2 trials in 2018 to a peak of 30 trials in 2021, representing a 15-fold increase (Fig. 1A). Regarding study design, RCT dominated the landscape with 97 trials (78.2%), while single-arm trials and NRCT comprised 20 (16.1%) and 7 (5.6%) studies, respectively. Trial completion status showed encouraging progress, with 46 studies (37.1%) successfully completed and 27 trials (21.8%) currently active or recruiting participants. Notably, only two trials (1.6%) were terminated, indicating high feasibility and sustained research momentum in this field (Fig. 1B). Academic institutions served as the predominant research sponsors, supporting 107 trials (86.3%; Supplemental Digital Content Figure S1, available at: https://links.lww.com/JS9/G742). Figure 1.: Clinical trial landscape of AI applications in GI endoscopy. (A) Annual distribution of AI-related GI endoscopy clinical trials by study design from 2018 to 2025. (B) Distribution of trial completion status across all registered studies. (C) Geographic distribution of clinical trials showing the top 10 contributing countries by study center type. (D) Distribution of the most frequently investigated AI systems in clinical trials categorized by clinical applications. (E) Sankey diagram illustrating the relationship between study center type and anatomical focus areas in registered trials. (F) Most common primary outcome measures utilized in AI gastrointestinal endoscopy clinical trials ranked by frequency. AI, artificial intelligence; GI: gastrointestinal; NRCT, non-randomized controlled trial; RCT, randomized controlled trial; UC, ulcerative colitis. Geographically, China demonstrated overwhelming dominance with 54 trials (43.5%), of which 44 were single-center studies and 10 were multi-center studies (Fig. 1C). The United States contributed nine trials (7.3%), comprising four single-center and five multi-center studies, while Italy also registered nine trials (7.3%) with seven single-center and two multi-center studies. Among AI systems investigated, GI Genius emerged as the most frequently studied platform with 13 trials (10.5%), primarily focusing on polyp detection (Fig. 1D). CAD-EYE and EndoAngel each appeared in seven trials (5.6%). Among the 124 trials, single-center studies predominated with 89 trials (71.8%), while multi-center studies comprised 35 trials (28.2%; Fig. 1E). Regarding anatomical focus, colorectal applications dominated with 98 trials (79.0%), followed by upper GI tract studies with 14 trials (11.3%). Within the colorectal domain, research efforts concentrated primarily on adenoma detection and characterization, followed by colorectal polyps. Clinical endpoints demonstrated a clear emphasis on adenoma detection capabilities, with 39 trials (31.5%) using adenoma detection rate as the primary outcome (Fig. 1F). Additional adenoma-related endpoints included adenomas per colonoscopy in 11 trials (8.9%) and adenoma miss rate in 8 trials (6.5%). Collectively, adenoma-related endpoints represented approximately 52.0% of all primary outcomes. Secondary areas of investigation encompassed optical diagnosis accuracy in nine trials (7.3%) and bowel preparation quality assessment in seven trials (5.6%), reflecting the broader applications of AI technology in endoscopic practice. Our analysis of 124 clinical trials registered up to 1 April 2025 demonstrates rapid growth in AI applications for GI endoscopy, with registrations increasing 15-fold from 2018 to 2021, positioning gastroenterology at the forefront of AI integration in clinical medicine. The high proportion of RCTs indicates that the application of AI in digestive endoscopy values rigorous clinical validation, which establishes a strong standard of evidence that will likely influence regulatory approval pathways[7]. China’s commanding research dominance represents a concentration of AI endoscopy research in a single country, reflecting healthcare digitization policies, vast patient populations, and substantial research funding prioritization. While this leadership has accelerated technological development, it raises important questions about global generalizability across diverse healthcare systems, regulatory frameworks, and patient demographics. The geographic concentration, combined with limited international collaboration, may create challenges for worldwide clinical implementation and regulatory harmonization. The predominance of single-center studies (71.8%) amplifies concerns about external validity. From a practical standpoint, AI algorithms developed within single institutional environments may exhibit performance variability when deployed across diverse clinical settings due to data heterogeneity arising from population diversity and different device manufacturers[8]. Additionally, differences in image quality standards, bowel preparation protocols, and operator expertise levels may further compromise AI system generalizability. The limited international collaboration evident in our findings suggests potential barriers to multi-center study design, including regulatory complexities, data sharing restrictions, and varying clinical practice standards. This geographic imbalance underscores the critical need for international collaborative frameworks to ensure that AI endoscopy technologies demonstrate robust performance across diverse clinical environments before widespread implementation. The predominant focus on adenoma detection reflects clinical priorities in colorectal cancer screening. Adenoma detection rate is an established quality metric with demonstrated correlation to interval cancer risk, providing a clinically meaningful validation target for AI systems[9]. Recent meta-analyses demonstrate that AI-assisted colonoscopy significantly improves adenoma detection rates and reduces miss rates[5]. The success of commercially approved systems like GI Genius, which received FDA clearance in 2021, validates the clinical translation potential of AI-assisted endoscopy[10]. Similarly, China’s approval of 59 AI medical devices as of July 2023 reflects the accelerating global adoption of AI-driven diagnostic technologies[11]. Several limitations affect our findings. First, our analysis relied exclusively on ClinicalTrials.gov, potentially underrepresenting trials registered in other databases such as the Chinese Clinical Trial Registry (ChiCTR), the European Union Clinical Trials Register (EU-CTR), or the Japan Registry of Clinical Trials. Given that China already demonstrates substantial research activity in our analysis (43.5%), the true magnitude of Asian-region contributions may be even greater if trials registered solely in ChiCTR were included. Similarly, European contributions may be underestimated due to preferential registration in EU-CTR. This database limitation affects the geographic generalizability of our findings and warrants consideration when interpreting the global distribution of AI endoscopy research. Second, the rapid evolution of AI technology means that recently developed systems may not yet be reflected in clinical trial registrations, creating a temporal lag in our analysis. Finally, our analysis is subject to clinical trial registration bias, with inconsistent study registration and incomplete trial updates potentially affecting the accuracy of our landscape assessment. This comprehensive landscape analysis demonstrates accelerating clinical research activity in AI applications for GI endoscopy, with notable trends toward diversification of clinical applications. Future technological developments will likely emphasize multi-modal AI integration, combining endoscopic imaging with clinical data, biomarkers, and patient histories to enable personalized diagnostic and therapeutic strategies[12]. For successful real-world implementation, multi-institutional trials and regulatory harmonization will be crucial milestones. As the field matures toward widespread clinical adoption, standardized evaluation metrics, comprehensive cost-effectiveness analyses, and streamlined regulatory frameworks will become essential prerequisites.