Clinical Report: Framework for Dual-Generation Synthesis in Colonoscopy

Overview

This study presents a dual generative synthesis framework aimed at improving polyp segmentation in colonoscopy images. The proposed method achieves a Dice score of 0.8786 and an Intersection over Union (IoU) of 0.7835.

Background

Colorectal cancer is a leading cause of cancer-related mortality, and effective screening through colonoscopy is essential for early detection and prevention. Accurate segmentation of polyps during colonoscopy is crucial, yet existing models struggle with small and flat polyps due to limited training data. This study addresses the need for improved segmentation techniques that can enhance the reliability of automated detection.

Data Highlights

Key Findings

• The dual generative synthesis framework improves segmentation performance for small and flat polyps.
• Realistic masks of small polyps are generated through procedural generation.
• A mask-conditioned GAN generates colonoscopy images that align with the generated masks.
• The proposed method outperforms baseline U-Net models in segmentation metrics.
• Data augmentation through generative methods addresses the issue of limited training data diversity.

Clinical Implications

The findings indicate that integrating generative models into training datasets can enhance the accuracy of polyp segmentation in colonoscopy.

Conclusion

The dual generative synthesis framework demonstrates potential in improving the robustness of automated polyp segmentation.

Related Resources & Content

1. Dorjsembe et al., 2026 -- Framework for Dual-Generation Synthesis: Improving Polyp Segmentation in Colonoscopy Using Mask-Conditional GANs
2. PrysmNet: A System for Polyp Refinement Utilizing Salience and Multimodal Approaches for Consistent Cross-Domain Segmentation, 2026
3. MAPSeg: self-supervised colorectal polyp segmentation via memory-augmented framework and synthetic polyp simulation, 2026
4. Colorectal cancer screening: An update to the American Cancer Society guideline, 2026
5. A Generative Multi-Adversarial Network Approach to Optimize Abdominal Image Segmentation Balance
6. Colorectal cancer screening: An update to the American Cancer Society guideline, 2026 - Wolf - 2026 - CA: A Cancer Journal for Clinicians - Wiley Online Library
7. Artificial intelligence–assisted colonoscopy and adenoma detection: An updated systematic review and meta-analysis of 42 studies. | Journal of Clinical Oncology