Motivation: Alzheimer's disease (AD) is marked by amyloid-beta plaques, typically detected through amyloid-PET imaging, which is expensive and limited in availability. Developing a more accessible diagnostic tool is essential for early detection and monitoring. Goal(s): This project aims to create a deep learning model that synthesizes amyloid-PET images from widely available MRI scans, offering a cost-effective, non-invasive alternative. Approach: A specialized VQGAN is trained to map MRI features to amyloid-PET images, utilizing datasets from patients across various stages of AD. Results: The model demonstrates accurate amyloid-PET synthesis, showing potential for early diagnosis and broad clinical application. Impact: This MRI-based deep learning method provides a cost-effective, non-invasive alternative to amyloid-PET imaging, potentially expanding diagnostic tools in clinical settings, especially where PET imaging is unavailable.
Zhang et al. (Tue,) studied this question.