Skip to main content

Combination of Artificial Intelligence-Based Endoscopy and Methylation Panels for Early Stage of Gastric Cancer

Authors
Yoshiyuki Watanabe1,2,3,5, Hiroyuki Yamamoto1,4, Ritsuko Oikawa1, Seiji Futagami5, Muhammad Miftahussurur2, Kok-Ann Gwee6, Tomohiro Tada7, Keisuke Tateishi1

Affiliations

  1. Division of Gastroenterology, Department of Internal Medicine, St. Marianna University School of Medicine, Kawasaki, Japan
  2. Division of Gastroentero-Hepatology, Department of Internal Medicine, Faculty of Medicine-Dr. Soetomo Teaching Hospital, Universitas Airlangga, Surabaya, Indonesia
  3. Department of Internal Medicine, Kawasaki Rinko General Hospital, Kawasaki, Japan
  4. Department of Bioinformatics, St. Marianna University Graduate School of Medicine, Kawasaki, Japan
  5. Department of Internal Medicine, Division of Gastroenterology, Nippon Medical School, Tokyo, Japan
  6. Department of Medicine, Yong Loo Lin School of Medicine, National University of Singapore, and The Gastroenterology Group, Gleneagles Hospital, Singapore City, Singapore.
  7. Tada Tomohiro Institute of Gastroenterology and Proctology, Saitama, Japan.

Corresponding Author
Yoshiyuki WATANABE (ponponta@marianna-u.ac.jp)

Background
Esophagogastroduodenoscopy (EGD) and biopsy-based pathological evaluation are needed to diagnose early gastric cancer (EGC). However, since biopsy is only a topical procedure, we have been focusing on DNA methylation using gastric wash fluid as a molecular marker for gastric cancer. In addition, we have decided to conduct a comparative examination with the recently emerged and highly regarded artificial intelligence (AI)-based EGD.

Methods
Gastric wash fluid was collected before and after endoscopic submucosal dissection (ESD) for EGC cases, and changes in four DNA methylation panels related to EGC (MINT25, SOX17, miR34, BARHL2) were evaluated. In addition, a total of 4 kinds of endoscopic images (white light images (WL), narrow band images (NBI), magnify endoscopic images (Mag), and indigo carmine staining images (Indigo)) were evaluated for AI-based EGD diagnosis.

Results
DNA methylation of the 49 cases tended to decrease after treatment in 4 genes, but this was not significant (MINT25: 3.67+6.11%, p=0.662, SOX17: 9.82+3.83%, p=0.992, miR34: 7.06+4.33%, p=0.575, BARHL2: 12.39+7.02, p=0.066). On the other hand, in AI-based endoscopy, the AI score of tumor lesions was high in images under all four conditions (WL: 77.08+7.35, NBI: 73.13+6.71, Mag: 69.54+8.24, Indigo: 70.64+9.17); Scar lesions after ESD showed a significant decrease in AI score and high AUC (WL60.40+1.32, p<0.0001, AUC0.999). Conclusion: AI-based EGD have a potential modality for EGC diagnosis.

Publication
Cancer Sci. 2024 Mar 25. doi: 10.1111/cas.16160.
https://pubmed.ncbi.nlm.nih.gov/38528657/