Asia-Pacific consensus on the use of artificial intelligence in colorectal cancer screening and surveillance

Frederick H. Koh, James Weiquan Li, Sunny H. Wong, Jonathan Lee, Sneha John, Vui Heng Chong, Kaichun Wu, Rashid N. Lui, Simon S.m. Ng, Thomas Yt. Lam, Louis Ho Shing Lau, Govind K Makharia, Murdani Abdullah, Hasan Maulahela, Nozomu Kobayashi, Masau Sekiguchi, Jeong‐Sik Byeon, Hyun‐Soo Kim, Yeong Yeh Lee, Han‐Mo Chiu, I‐Chen Wu, Somchai Leelakusolvong, Prateek Sharma, David A. Lieberman, Joseph J.�y. SungiGIE 2025 · DOI

TL;DR

This paper presents a consensus from Asia-Pacific experts on the use of Artificial Intelligence (AI) in colorectal cancer (CRC) screening and surveillance via colonoscopy. It highlights that while AI shows promise in improving adenoma detection and quality assurance, its role in polyp characterization and treatment decisions is still evolving, emphasizing the need for further validation and careful implementation.

What Problem Does This Paper Address?

Colorectal cancer (CRC) is a significant global health burden, with the Asia-Pacific region bearing a substantial portion of cases and mortality. Colonoscopy is a primary screening tool, but its effectiveness relies heavily on the endoscopist's skill in detecting and characterizing polyps. While Artificial Intelligence (AI) has shown potential to enhance colonoscopy, there's a lack of clear clinical guidance on its practical application in screening and surveillance. This paper aims to bridge that gap by developing consensus statements on when and how to use AI-assisted colonoscopy, based on the latest evidence.

Methods

The study employed a modified Delphi process, a rigorous method for reaching consensus among experts. A panel of 12 Asia-Pacific experts and 4 international experts, selected for their knowledge and experience in CRC screening and AI in colonoscopy, developed consensus statements. These statements were based on a comprehensive literature search of articles published between 2018 and October 2023, along with national and international guidelines. The consensus process involved anonymized electronic voting using a Likert scale, with specific criteria for achieving consensus (over 80% agreement) or refuting a statement (over 80% disagreement). Statements were graded for the quality of evidence and strength of recommendation.

Datasets

Not discussed in the paper. The paper focuses on synthesizing existing evidence and expert opinion rather than introducing new datasets or benchmarks.

Key Insights and Ideas

  • AI as an Adjunct, Not a Replacement: The consensus strongly suggests that AI-assisted colonoscopy should be viewed as an adjunct to conventional colonoscopy, enhancing its capabilities rather than replacing the endoscopist.
  • CADe Shows Promise for Detection: Computer-Assisted Detection (CADe) systems have demonstrated a consistent ability to improve adenoma detection rates (ADRs), which is a key quality metric for colonoscopy.
  • CADx for Characterization is Promising but Not Definitive: Computer-Assisted Diagnosis (CADx) shows potential for characterizing polyps, but it's not yet reliable enough to make definitive decisions about whether a polyp needs to be removed. Histological confirmation remains crucial.
  • Quality Assurance is a Strong Suit: Computer-Assisted Quality Assurance (CAQ) tools are well-received for their ability to monitor and improve the quality of colonoscopy procedures in real-time, such as withdrawal speed and bowel preparation.
  • Implementation Requires Engagement: Successful adoption of AI in colonoscopy hinges on engaging endoscopists and their assistants to build trust and address concerns. Training programs need to incorporate AI knowledge.
  • Patient Consent is Important: While not strictly mandated in all cases, informing patients about the use of AI in their colonoscopy is recommended to ensure transparency and address potential ethical considerations.
  • Sessile Serrated Lesions (SSLs) Remain a Challenge: Current AI systems lack sufficient accuracy to reliably differentiate SSLs from hyperplastic lesions, highlighting an area for future technological development.

Key Results

  • CADe Improvement: A meta-analysis of relevant trials showed a 24% improvement in adenoma detection rate (ADR) using CADe.
  • CADx Accuracy: While CADx can characterize polyps, its accuracy in differentiating specific lesion types (like SSLs) is still limited. For instance, one study showed a sensitivity for SSLs of only 17.1%.
  • CAQ Effectiveness: CAQ systems have demonstrated high accuracy (e.g., 93.33% for bowel preparation) and can significantly improve ADR when combined with CADe.
  • Agreement on Statements: The consensus achieved high levels of agreement for most statements, particularly regarding the utility of CADe for adenoma detection (74.1% "accept completely") and the use of CADx for polyp characterization (70.4% "accept completely").

Limitations and Future Work

The authors acknowledge that more validation studies are needed for AI-assisted colonoscopy, especially concerning its impact on reducing advanced adenoma and CRC incidence, as well as its cost-effectiveness. They also highlight the need for further research into the detection and characterization of sessile serrated lesions (SSLs) using AI. Future work should focus on developing AI technologies that can provide explainability for their decisions, thereby building greater trust among clinicians. Additionally, region-specific health economic analyses are crucial to inform healthcare authorities about the adoption of these technologies, considering the disparities in healthcare systems across the Asia-Pacific region. The paper also suggests that future studies should consider reporting both ADR and APC (adenoma per colonoscopy) to better assess the effectiveness of CADe.

Summary updated

Abstract

More validation studies on AI-assisted colonoscopy need to be done while AI technologies continue to improve.
BibTeX
@inproceedings{koh2025asia,
  author = {Frederick H. Koh and James Weiquan Li and Sunny H. Wong and Jonathan Lee and Sneha John and Vui Heng Chong and Kaichun Wu and Rashid N. Lui and Simon S.M. Ng and Thomas YT. Lam and Louis Ho Shing Lau and Govind K Makharia and Murdani Abdullah and Hasan Maulahela and Nozomu Kobayashi and Masau Sekiguchi and Jeong‐Sik Byeon and Hyun‐Soo Kim and Yeong Yeh Lee and Han‐Mo Chiu and I‐Chen Wu and Somchai Leelakusolvong and Prateek Sharma and David A. Lieberman and Joseph J.�Y. Sung},
  title = {{Asia-Pacific consensus on the use of artificial intelligence in colorectal cancer screening and surveillance}},
  booktitle = {iGIE},
  year = {2025},
  doi = {10.1016/j.igie.2025.04.001},
}
Figure