Grant: Project Grant

Alzheimer's disease involves two key brain changes-amyloid "plaques" and tau "tangles." Today, the best way to see these directly is a PET scan with a special tracer, but PET is costly, requires an injection, exposes patients to radiation, and isn't widely available. By contrast, MRI is routine, safe, and accessible-but it doesn't directly show amyloid or tau. Our project will train computer models to estimate amyloid and tau from standard MRI scans, so that the information PET provides could be approximated when PET is hard to access. The intent is not to replace PET, but to help clinicians decide who most needs a PET scan or specialist follow-up, and to enable safer, repeatable monitoring over time. In practical terms, this could support timely conversations with patients, reduce uncertainty while waiting for specialty testing, and triage limited PET capacity to those most likely to benefit. We will first build and test the approach using large, well-known research datasets. Then we will check that MRI-based estimates match blood and spinal-fluid Alzheimer's biomarkers. Finally, we will evaluate the method in real-world clinic workflows in a "silent-mode" trial: doctors will be blinded to the model's outputs while we measure accuracy, safety, and potential impact on decisions. Only after rigorous validation would any use in care be considered. If successful, this work could make Alzheimer's assessment more accessible and less invasive-providing PET-like information at MRI appointments-while reducing costs and wait times for patients and health systems. People in rural and remote communities, who often travel far for PET, could especially benefit. We will also examine whether results differ by sex to ensure fairness and reliability.