Related Experiment Video
Updated: Jun 20, 2025
16:59
Computer-assisted Large-scale Visualization and Quantification of Pancreatic Islet Mass, Size Distribution and Architecture
Published on: March 4, 2011
12.3K
Diagnosing Solid Lesions in the Pancreas With Multimodal Artificial Intelligence: A Randomized Crossover Trial
Haochen Cui1, Yuchong Zhao1, Si Xiong1
1Department of Gastroenterology and Hepatology, Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology, Wuhan, China.
JAMA Network Open
|July 19, 2024
Summary
A novel multimodal artificial intelligence (AI) model integrating clinical data and endoscopic ultrasonography (EUS) images significantly improved the diagnosis of solid pancreatic lesions, enhancing accuracy for novice endoscopists.
Area of Science:
- Medical imaging analysis
- Artificial intelligence in healthcare
- Gastroenterology diagnostics
Background:
- Diagnosing solid pancreatic lesions using endoscopic ultrasonography (EUS) is challenging.
- Existing AI models for EUS diagnosis are limited to single modalities.
- A multimodal approach integrating clinical data and EUS images is needed.
Purpose of the Study:
- To develop and evaluate a multimodal AI model for diagnosing solid pancreatic lesions.
- To assess the impact of AI assistance on endoscopist diagnostic performance.
- To integrate clinical information and EUS images for improved diagnostic accuracy.
Main Methods:
- A randomized crossover trial involving 12 endoscopists across 4 centers.
- Development and internal/external validation of a joint AI model using retrospective data (628 patients).
- Prospective evaluation of AI-assisted diagnosis in 130 patients.
Main Results:
- The joint AI model achieved high diagnostic performance, with AUCs ranging from 0.924 to 0.996 across internal and external test datasets.
- AI assistance significantly improved diagnostic accuracy for novice endoscopists (0.69 to 0.90).
- AI interpretability features reduced skepticism among experienced endoscopists.
Conclusions:
- The multimodal AI model shows potential for facilitating clinical diagnosis of solid pancreatic lesions.
- Positive human-AI interaction was observed, suggesting clinical utility.
- Further randomized clinical trials are recommended to confirm findings.
