Conditional Diffusion Models for Semantic 3D Medical Image Synthesis

• DOI: 10.36227/techrxiv.23723787.v2
• Publication Date: 2023-07-30T13:40:05Z
• Authors: Zolnamar Dorjsembe; Hsing-Kuo Pao; Sodtavilan Odonch...; Furen Xiao

<p>The demand for artificial intelligence (AI) in healthcare is rapidly increasing. However, significant challenges arise from data scarcity and privacy concerns, particularly medical imaging. While existing generative models have achieved success image synthesis image-to-image translation tasks, there remains a gap the generation of 3D semantic images. To address this gap, we introduce Med-DDPM, diffusion model specifically designed synthesis, effectively tackling issues. The novelty Med-DDPM lies its incorporation conditioning, enabling precise control during process. Our outperforms Generative Adversarial Networks (GANs) terms stability performance, generating diverse anatomically coherent images with high visual fidelity. Comparative analysis against state-of-the-art augmentation techniques demonstrates that produces comparable results, highlighting potential as tool enhancing accuracy. In conclusion, pioneers by delivering high-quality Furthermore, integration conditioning holds promise anonymization field biomedical imaging, showcasing capabilities addressing related to concerns. code weights are publicly accessible on our GitHub repository at <a href="https://github.com/mobaidoctor/med-ddpm/" target="_blank"><u>https://github.com/mobaidoctor/med-ddpm/</u></a>, facilitating reproducibility.</p>