Multi-modal endoscopic artificial intelligence system improving classification of colorectal subepithelial lesions: a multicenter study

Shurong Chen; Louzhe Xu; Yihong Zhang; Jinghua Wang; Xinjue He; Tianlian Yan; Han Ma; Yizhen Zhang; Lan Li; Shu Bian; Yue Huang; Chao Lu; Yangyang Xiong; Jiang Liu; Qiang Chen; Lina Ye; Qiaona He; Yini Ke; Youming Li; Xinli Mao; Chaohui Yu; Chunxiao Li; Ting Li; Yi Chen

doi:10.1016/j.dld.2025.06.003

Multi-modal endoscopic artificial intelligence system improving classification of colorectal subepithelial lesions: a multicenter study

Dig Liver Dis. 2025 Jun 27:S1590-8658(25)00807-2. doi: 10.1016/j.dld.2025.06.003. Online ahead of print.

Authors

Shurong Chen¹, Louzhe Xu², Yihong Zhang¹, Jinghua Wang¹, Xinjue He¹, Tianlian Yan¹, Han Ma¹, Yizhen Zhang¹, Lan Li¹, Shu Bian³, Yue Huang¹, Chao Lu¹, Yangyang Xiong¹, Jiang Liu⁴, Qiang Chen⁵, Lina Ye⁶, Qiaona He⁷, Yini Ke⁸, Youming Li¹, Xinli Mao⁹, Chaohui Yu¹⁰, Chunxiao Li¹¹, Ting Li¹², Yi Chen¹³

Affiliations

¹ Department of Gastroenterology, the First Affiliated Hospital, College of Medicine, Zhejiang University, Hangzhou, China.
² Institute of Biomedical Engineering, Chinese Academy of Medical Sciences & Peking Union Medical College, Tianjin, China.
³ Department of Gastroenterology, Tianjin Union Medical Center, the First Affiliated Hospital of Nankai University, Tianjin, China.
⁴ Department of Gastroenterology, Huzhou Central Hospital, Affiliated Central Hospital Huzhou University, Huzhou, China.
⁵ Department of Gastroenterology, Sanmen County People's Hospital, Taizhou, China.
⁶ Department of Gastroenterology, Shengzhou People's Hospital, Shaoxing, China.
⁷ Department of Gastroenterology, Zhejiang Hospital, Hangzhou, China.
⁸ Department of Rheumatology, the First Affiliated Hospital, College of Medicine, Zhejiang University, Hangzhou, China.
⁹ Department of Gastroenterology, Taizhou Hospital, Taizhou, China.
¹⁰ Department of Gastroenterology, the First Affiliated Hospital, College of Medicine, Zhejiang University, Hangzhou, China. Electronic address: zyyyych@zju.edu.cn.
¹¹ Department of Gastroenterology, Ningbo First Hospital, Ningbo, China. Electronic address: 11418130@zju.edu.cn.
¹² Institute of Biomedical Engineering, Chinese Academy of Medical Sciences & Peking Union Medical College, Tianjin, China. Electronic address: t.li619@foxmail.com.
¹³ Department of Gastroenterology, the First Affiliated Hospital, College of Medicine, Zhejiang University, Hangzhou, China. Electronic address: zyyyychen@zju.edu.cn.

PMID: 40581582
DOI: 10.1016/j.dld.2025.06.003

Abstract

Background: Colorectal subepithelial lesions (SELs) present with similar endoscopic appearance but diverse in histology and prognosis, making precise diagnosis crucial but challenging.

Aim: This study aims to develop a multi-modal endoscopic artificial intelligence (AI) system to facilitate colorectal SELs classification.

Methods: Patients with histologically confirmed colorectal SELs were retrospectively enrolled from 7 hospitals. Single-modal models [Model W (WLE), Model E (EUS)] and multi-modal AI system, AIOSCOPE-WE, were constructed. The classification performance was evaluated both internally and externally. Endoscopists' performance with and without AIOSCOPE-WE support was compared.

Results: 510 patients (364 neuroendocrine tumors (NETs), 72 lipomas, 23 leiomyomas and 51 inflammatory hyperplasia) with 5118 WLE images and 4950 EUS images were included. In classifying colorectal SELs, AIOSCOPE-WE achieved higher accuracy than EUS experts and non-experts (86.0 % vs 75.8 % and 65.4 %, P < 0.01). With AI support, endoscopists' performance was improved from 70.6 % to 83.1 % (P < 0.001). AIOSCOPE-WE also significantly outperformed Model W and Model E in accuracy in both internal (92.1 % vs 71.9 % and 78.6 %, P < 0.001) and external tests (84.9 % vs 69.4 % and 63.9 %, P < 0.001), with consistent performance across lesion sizes and NET grades.

Conclusion: AIOSCOPE-WE holds great promise as a potent tool for endoscopists in classifying colorectal SELs and optimizing clinical decisions.

Keywords: Artificial intelligence; Endoscopy; Multi-modal image; Subepithelial lesion.