Enhancing Atlas Construction via Edge-Assisted 3D Latent Diffusion Models for Brain MRI Generation

Atlas construction is a cornerstone in medical imaging, facilitating research, diagnosis, and personalized treatment planning. In this paper, we introduce an edge-assisted 3D latent diffusion model, Med3Diffusion, trained on the Alzheimer’s Disease Neuroimaging Initiative (ADNI) dataset, to enhance atlas construction by synthesizing high-quality brain MRI scans across ages. These generated images augment the original ADNI dataset, addressing the issue of age-related data imbalance and improving the performance of the current atlas-building methods. We employ two state-of-the-art atlas construction methods, i.e., Aladdin and a VoxelMorph-based model, to construct the atlases with this enriched dataset. Our approach not only compensates for the scarcity of certain age groups in the existing dataset but also improves the diversity and quality of the atlas construction. Additionally, experimental results demonstrate the potential of our Med3Diffusion model in generating synthetic 3D medical images for broader applications in medical fields. Our source code is released at https://github.com/sjtu-media/Med3Diffusion.