Related Experiment Video
Updated: Mar 10, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.6K
Two-stage deep learning framework for laterally spreading tumors detection using self-supervised learning and
Menghui Wang1, Zhanpeng Shi2, Yiwen Wang3
1Department of Gastroenterology, Shanghai General Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai, China.
Digital Health
|March 9, 2026
Summary
This study introduces an AI framework for detecting laterally spreading tumors (LSTs) in colonoscopy images, achieving 72.4% accuracy with limited expert data. The method combines self-supervised learning and few-shot classification for efficient LSTs recognition.
Area of Science:
- Gastroenterology
- Artificial Intelligence
- Medical Imaging
Background:
- Laterally spreading tumors (LSTs) are flat, precancerous colorectal lesions often missed during colonoscopy.
- Their subtle morphology and low prevalence present a significant early detection challenge.
Purpose of the Study:
- To develop a reliable AI model for LSTs detection using colonoscopy images.
- To overcome limitations of supervised learning with scarce annotated data.
Main Methods:
- A framework combining DINO self-supervised pretraining on 150,168 unlabeled images and Prototypical Networks for few-shot classification.
- Utilized 2799 labeled training images and evaluated on 601 test images.
Main Results:
- Achieved 72.4% overall accuracy, 74.7% sensitivity, and 70.2% specificity.
- Demonstrated a ROC-AUC of 0.798 and processed images at 50ms per image.
Conclusions:
- The AI framework effectively detects LSTs with limited annotations, reducing data requirements.
- This approach is suitable for resource-constrained settings and real-time colonoscopy support.