DIFFUSION MODELS FOR MEDICAL IMAGE RECONSTRUCTION AND DENOISING IN LOW-RESOURCE SETTINGS
Keywords:
Diffusion Models, Denoising Diffusion Probabilistic Models (DDPM), Medical Image Reconstruction, Image Denoising, Deep Generative Models, Inverse Problems in ImagingAbstract
Medical imaging plays a vital role in diagnosis and treatment planning, but access to high-quality imaging technologies and reliable data acquisition pipelines remains limited in low-resource settings. Recent advances in generative models, especially diffusion models, offer new possibilities for robust medical image reconstruction and denoising even under severe data constraints. In this paper, we investigate the application of diffusion probabilistic models for medical image reconstruction and denoising tasks, with a focus on scenarios involving low signal-to-noise ratios (SNR), sparse data acquisition, and limited computational infrastructure. We demonstrate that these models can produce high-fidelity reconstructions from degraded or incomplete inputs and outperform conventional methods and other deep learning baselines in multiple medical imaging modalities including MRI, CT, and ultrasound. We further propose lightweight and compressed diffusion model architectures tailored for deployment in low-resource clinical environments.
