Related Experiment Video
Updated: Sep 11, 2025
13:01
Industrialized, Artificial Intelligence-guided Laser Microdissection for Microscaled Proteomic Analysis of the Tumor Microenvironment
Published on: June 3, 2022
3.9K
Multimodal artificial intelligence for subepithelial lesion classification and characterization: a multicenter
Jiao Li1,2, Xiaojuan Jing3, Qin Zhang4
1Department of Gastroenterology, The Second Affiliated Hospital of Chongqing Medical University, Linjiang Road 76#, Chongqing, Yuzhong District, China.
BMC Medical Informatics and Decision Making
|August 15, 2025
Summary
A new AI model, ECMAI-WME, integrates endoscopy and ultrasound to accurately classify gastrointestinal subepithelial lesions. This deep learning tool significantly outperforms human endoscopists in diagnosis and treatment decisions.
Area of Science:
- Gastroenterology
- Artificial Intelligence
- Medical Imaging
Background:
- Gastrointestinal subepithelial lesions (SELs) pose diagnostic challenges, especially distinguishing malignant from benign types.
- Misdiagnosis of SELs can lead to inappropriate interventions or delayed treatment.
- Accurate characterization of SELs is crucial for effective patient management.
Purpose of the Study:
- To develop and evaluate ECMAI-WME, a parallel fusion deep learning model integrating white light endoscopy (WLE) and microprobe endoscopic ultrasonography (EUS).
- To improve the classification and characterization of gastrointestinal subepithelial lesions.
- To enhance diagnostic accuracy and support clinical decision-making in SEL management.
Main Methods:
- Developed serial and parallel fusion AI models using data from 523 SELs across four hospitals.
- Designated the superior performing model as ECMAI-WME (a parallel fusion model).
- Validated ECMAI-WME on external (n=88) and multicenter (n=274) cohorts, comparing its performance against endoscopists.
Main Results:
- ECMAI-WME significantly outperformed endoscopists in diagnostic accuracy (96.35% vs. 63.87-86.13%) and treatment decision accuracy (96.35% vs. 78.47-86.13%).
- Achieved high accuracy in multiclass SEL classification and characterization (94.81% mean accuracy).
- Demonstrated robust performance and generalizability across validation cohorts and subgroup analyses.
Conclusions:
- The ECMAI-WME model shows excellent diagnostic performance and robustness for multiclass SEL classification and characterization.
- Supports potential for real-time deployment to improve diagnostic consistency.
- Aids in guiding clinical decision-making for gastrointestinal subepithelial lesions.
