Synthesizing Aβ-PET based on multi-modal neuroimaging fusion for pathological diagnosis of Alzheimer’s disease

Synthesizing Aβ-PET images from cross-modal neuroimaging for diagnosing Alzheimer’s disease through multi-modal medical image fusion is highly significant. However, there are relatively few studies in this area. We propose a multi-modal fusion adversarial method for Aβ-PET medical image synthesis, named MultiMed, to generate Aβ-PET images by fusing features from T1 and Flair images. Our approach employs an end-to-end method for image generation, incorporating a Multi-Feature Fusion Stream module for feature fusion. Integrating multimodal features enhances the model’s representation capability and preliminarily designed a visual domain classifier for Alzheimer’s disease diagnosis. We constructed an internal dataset for model training and validated our model on two external validation sets, achieving notable performance.