🤖 AI Summary
Medical image data scarcity, high annotation costs, and privacy constraints severely limit polyp detection performance. To address this, we propose the first semantic-guided diffusion model conditioned on binary segmentation masks—termed Mask-Conditioned Diffusion Model—for high-fidelity and diverse synthetic polyp image generation. Our method innovatively integrates mask-image joint modeling with a semantic-guided sampling mechanism and introduces a multi-objective evaluation framework combining Fréchet Inception Distance (FID) and Intersection-over-Union (IoU). On gastrointestinal polyp synthesis, our model achieves an FID of 78.47—significantly outperforming all baselines by over 95.8—and yields a downstream segmentation IoU of 0.7156, matching the performance of models trained exclusively on real data. To foster reproducibility and further research, we fully open-source our code, pre-trained weights, and the generated synthetic dataset.
📝 Abstract
This study introduces Polyp-DDPM, a diffusion-based method for generating realistic images of polyps conditioned on masks, aimed at enhancing the segmentation of gastrointestinal (GI) tract polyps. Our approach addresses the challenges of data limitations, high annotation costs, and privacy concerns associated with medical images. By conditioning the diffusion model on segmentation masks—binary masks that represent abnormal areas—Polyp-DDPM outperforms state-of-the-art methods in terms of image quality (achieving a Fréchet Inception Distance (FID) score of 78.47, compared to scores above 95.82) and segmentation performance (achieving an Intersection over Union (IoU) of 0.7156, versus less than 0.6828 for synthetic images from baseline models and 0.7067 for real data). Our method generates a high-quality, diverse synthetic dataset for training, thereby enhancing polyp segmentation models to be comparable with real images and offering greater data augmentation capabilities to improve segmentation models. The source code and pretrained weights for Polyp-DDPM are made publicly available at https://github.com/mobaidoctor/polyp-ddpm.