Search papers, labs, and topics across Lattice.
ACADiff, a novel latent diffusion framework, is introduced to address missing modality imputation in multimodal neuroimaging for Alzheimer's disease diagnosis. It leverages adaptive fusion of available imaging data and clinical metadata, encoded via GPT-4o prompts, to guide the denoising process in the latent space. Experiments on ADNI data demonstrate that ACADiff achieves state-of-the-art generation quality and diagnostic performance, even with 80% missing data, for sMRI, FDG-PET, and AV45-PET modalities.
Even with 80% of brain scan data missing, ACADiff can accurately generate the missing modalities and maintain robust diagnostic performance for Alzheimer's disease.
Multimodal neuroimaging provides complementary insights for Alzheimer's disease diagnosis, yet clinical datasets frequently suffer from missing modalities. We propose ACADiff, a framework that synthesizes missing brain imaging modalities through adaptive clinical-aware diffusion. ACADiff learns mappings between incomplete multimodal observations and target modalities by progressively denoising latent representations while attending to available imaging data and clinical metadata. The framework employs adaptive fusion that dynamically reconfigures based on input availability, coupled with semantic clinical guidance via GPT-4o-encoded prompts. Three specialized generators enable bidirectional synthesis among sMRI, FDG-PET, and AV45-PET. Evaluated on ADNI subjects, ACADiff achieves superior generation quality and maintains robust diagnostic performance even under extreme 80\% missing scenarios, outperforming all existing baselines. To promote reproducibility, code is available at https://github.com/rongzhou7/ACADiff
