To present a dual generative synthesis framework for data-centric augmentation to improve segmentation performance of small and flat polyps in colonoscopy images, addressing challenges such as subtle edges and poor contrast.
Approach:
Step 1: Procedural generation is used to create realistic masks of small, flat polyps, addressing the issue of limited diversity in training data.
Step 2: A mask-conditioned GAN generates colonoscopy images that match the generated masks in texture and lighting conditions, ensuring anatomical realism and alignment.
Key Findings:
Achieved a Dice score of 0.8786 and an Intersection over Union (IoU) of 0.7835.
Model obtained a precision of 0.8930 and a recall of 0.8648, significantly higher than baseline U-Net model.
Interpretation:
The dual generative synthesis framework improves segmentation robustness by generating realistic and aligned training data for small and flat polyps.
Limitations:
The study does not address the performance of the framework on larger or more complex polyps.
Potential computational inefficiencies in the GAN training process are not discussed.
Conclusion:
The results demonstrate the framework's potential to enhance the reliability of automated polyp segmentation in colonoscopy images.
