An interpretable artificial intelligence system for measuring the size of small polyps (< 10 mm)
Yong Li1, Fujun Li2, Guanghui Lian1
1Gastroenterology, Xiangya Hospital Central South University, China.
Revista Espanola De Enfermedades Digestivas
|October 23, 2025
Summary
An artificial intelligence system, PolypM, accurately measures colonic polyp size, outperforming human endoscopists. This AI tool aids in determining treatment and surveillance strategies for better patient outcomes.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Gastroenterology
Background:
- Accurate colonic polyp size measurement is crucial for clinical decisions.
- Small polyps (<10 mm) pose measurement challenges for endoscopists.
- Current methods can lead to measurement ambiguity, impacting treatment and surveillance.
Purpose of the Study:
- To develop an artificial intelligence (AI) system named PolypM.
- To enable automated and precise polyp size measurement.
- To enhance clinical decision-making for colonic polyp management.
Main Methods:
- PolypM utilizes two models for automatic segmentation of transparent caps and polyps.
- The system was trained on a large dataset of 6486 endoscopic images.
- Performance was validated against endoscopist measurements on independent datasets.
Main Results:
- PolypM achieved high segmentation accuracy with an IoU of 0.91 for caps and 0.75 for polyps.
- The system demonstrated strong agreement with the gold standard (ICC=0.682).
- PolypM outperformed endoscopists in polyp size estimation accuracy during external validation.
Conclusions:
- PolypM provides an interpretable AI solution for colonic polyp size estimation.
- The system mitigates measurement ambiguity, improving diagnostic reliability.
- PolypM facilitates informed decisions on surgical intervention and surveillance timing.
