Polyp-DDPM: Diffusion-Based Semantic Polyp Synthesis for Enhanced Segmentation

• DOI: 10.48550/arxiv.2402.04031
• Publication Date: 2024-02-06
• Authors: Zolnamar Dorjsembe; Hsing-Kuo Pao; Furen Xiao
• Keywords: FOS: Computer and information sciences; Computer Science - Machine Learning; Computer Vision and Pattern Recognition (cs.CV); Computer Science - Computer Vision and Pattern Recognition; Machine Learning (cs.LG)

This study introduces Polyp-DDPM, a diffusion-based method for generating realistic images of polyps conditioned on masks, aimed at enhancing the segmentation gastrointestinal (GI) tract polyps. Our approach addresses challenges data limitations, high annotation costs, and privacy concerns associated with medical images. By conditioning diffusion model masks-binary masks that represent abnormal areas-Polyp-DDPM outperforms state-of-the-art methods in terms image quality (achieving Frechet Inception Distance (FID) score 78.47, compared to scores above 83.79) performance an Intersection over Union (IoU) 0.7156, versus less than 0.6694 synthetic from baseline models 0.7067 real data). generates high-quality, diverse dataset training, thereby polyp be comparable offering greater augmentation capabilities improve models. The source code pretrained weights Polyp-DDPM are made publicly available https://github.com/mobaidoctor/polyp-ddpm.