Wavelet-Fusion Diffusion Model for Multimodal Brain MRI Synthesis with Modality and Metadata Conditioning
Researchers have developed a new Wavelet-Fusion Diffusion Model (WFDM) for generating synthetic brain MRI scans. This model addresses limitations in existing methods by effectively handling uneven modality coverage and variations in acquisition protocols and metadata across diverse datasets. WFDM utilizes a latent diffusion approach with a Wavelet-Fusion variational autoencoder and a conditional 3D U-Net diffusion model, demonstrating superior distributional alignment compared to other synthetic MRI generators. AI
IMPACT Enhances capabilities for multimodal neuroimaging analysis and dataset augmentation by generating realistic synthetic MRI data.