Download PDF
Endoscopy
DOI: 10.1055/a-2750-5476
Editorial

Artificial intelligence in endoscopy: a yet underused but promising tool in underserved settings

Referring to Chiang TH et al. doi: 10.1055/a-2721-6552

Authors

  • Rui Morais 

    1   Gastroenterology Department, Unidade Local de Saúde de São João, Porto, Portugal (Ringgold ID: RIN285211)
    2   Faculty of Medicine, University of Porto, Porto, Portugal

  • Mário Dinis-Ribeiro

    2   Faculty of Medicine, University of Porto, Porto, Portugal
    3   Gastroenterology Department, Portuguese Oncology Institute of Porto (IPO Porto), Porto, Portugal
    4   RISE@CI-IPO (Health Research Network), Portuguese Oncology Institute of Porto (IPO Porto), Porto, Portugal
Further Information
Also available at
   Permissions and Reprints

10.1055/a-2721-6552

Gastric cancer is a leading cause of global mortality [1]. There is marked worldwide variability in the incidence of gastric cancer, with the incidence being highest in East Asia, Central Asia, and Latin America [1] [2]. Nevertheless, it is important to note that, even within regions, significant heterogeneity in incidence can occur. Furthermore, it is important not to forget that even in countries with a relatively low incidence and mortality, the burden is typically higher among the most vulnerable populations [3]. If the current trends continue, a 62% rise in the number of cases is expected within a 20-year period, irrespective of the level of human development index. Therefore, a significant effort is needed in terms of preventive and screening strategies to try to overcome this situation.

“This study addresses a significant and current challenge of the practical implementation of AI in endoscopic practice in low- and middle-income countries and/or underserved areas.”

Potential strategies include population-based Helicobacter pylori screening and treatment programs [4], and endoscopic screening in countries with high disease incidence (or synergistically associated with colorectal screening programs in intermediate regions), while in low-incidence regions identifying individuals with precancerous lesions and ensuring appropriate surveillance may represent a viable strategy by which to improve gastric cancer detection and management [5].

Artificial intelligence (AI) has emerged as a potential tool to assist with these processes. While the current main application of AI systems in gastrointestinal endoscopy is in lesion detection and characterization, recent studies using both machine learning and deep learning models have shown the potential of this technology for not only identifying high-risk patients who could benefit from endoscopic screening and surveillance [6] [7], but also accurately stratifying and tailoring surveillance in these patients [8]. Moreover, it should be noted that in resource-limited settings, important factors such as increasingly demanding clinical workloads, low funding, data availability, and limited computational infrastructure may place an additional burden on endoscopists and limit AI adoption in clinical practice [9].

In this issue of Endoscopy, Chiang et al. report the results of a rural-to-center AI model for diagnosing H. pylori infection and premalignant gastric conditions (namely atrophy and intestinal metaplasia) using static upper endoscopy images captured during routine clinical practice [10]. The study took place in both a rural hospital in the Matsu Islands (located 206 km from the Taiwan main island) and a tertiary reference center across Taiwan Strait. The study included a development phase (2012–2022) and a model implementation phase that took place between 2023 and 2024. In the development phase, AI algorithms were trained, validated, and tested for the desired outcomes, while in the model implementation phase endoscopy images from the rural hospital were transmitted to the tertiary center for analysis, with the results promptly returned. The authors found that the AI model showed fair to good accuracies in the implementation phases, with no significant differences observed between AI-predicted and clinically observed prevalence of H. pylori infection and premalignant conditions.

This study addresses a current and significant challenge of the practical implementation of AI in endoscopic practice in low- and middle-income countries and/or underserved areas. In fact, in these settings, AI use may be hindered by infrastructural, economic, regulatory, and training barriers [11]. The potential clinical impact of these findings should be highlighted, as they may enable physicians working in resource-limited settings to rapidly assess the H. pylori status and staging of gastritis, without the need for biopsies. As this may lead to additional diagnostic work-up, the authors in this study also performed a cost-effectiveness analyses comparing the AI-assisted approach with the standard approach, with the former being associated with an incremental gain of 2.02 life-years and a cost reduction of US $93.7 [10]. It should be noted that while a direct comparison was not established, the authors also included the possibility of performing rapid urease test for H. pylori in the non-AI approach. This test, while widely available and being able to provide results within 1–24 hours with reasonable sensitivity, is not routinely used by many clinicians [12]. This is of relevance because the results from the current study show that AI assistance may offer additional value, which is particularly relevant given the additional gains of AI in the detection of premalignant lesions.

Some significant limitations of the current study, recognized also by the authors, should be noted, however. First, virtual chromoendoscopy images were not used in the model development. This is of particular relevance as right now these techniques are almost universally available, and virtual chromoendoscopy scores such as Endoscopic Grading of Gastric Intestinal Metaplasia (EGGIM) have been reported as additional tools in the personalized management of gastric precancerous conditions, even when estimated by AI models [8] [13]. Nevertheless, as the authors state, the application of EGGIM may be challenging for less experienced endoscopists outside specialized centers and virtual chromoendoscopy may not always be available, especially in the outreach setting. However, further studies should address not only the additional value of this AI model in different clinical contexts but also possibly try to evaluate additional gains in accuracy of the model by also integrating virtual chromoendoscopy images.

In conclusion, while several questions remain, namely the long-term clinical impact of the implementation of an AI-assisted strategy, the current study shows a promising approach that can be particularly helpful for the diagnosis and management of gastric precancerous conditions, in outreach settings where health care providers may have limited experience or low resources. In addition, it opens the door to explore AI-assisted tools for training and endoscopic service delivery in underserved areas.



Publication History

Article published online:
15 December 2025

© 2025. Thieme. All rights reserved.

Georg Thieme Verlag KG
Oswald-Hesse-Straße 50, 70469 Stuttgart, Germany

Refers to:

A rural-to-center artificial intelligence model for diagnosing Helicobacter pylori infection and premalignant gastric conditions using endoscopy images captured in routine practiceEndoscopy eFirst
DOI: 10.1055/a-2721-6552