Motivation: Amyloid PET imaging is important for Alzheimer's disease (AD) diagnosis but currently largely inaccessible. As an alternative, generating synthetic PET data from T1w MRI has been explored. However, current PET synthesis models lack rigorous evaluation in brain regions that are clinically relevant to AD pathology. Goal(s): To provide a clinically focused evaluation of T1w-to-amyloid PET synthesis models for AD. Approach: We will train a generative model to predict PET from T1w MRI and assess performance using standard metrics, alongside a clinical evaluation of reconstruction accuracy in brain regions where amyloid load has diagnostic importance for AD. Impact: Our findings could clarify the clinically relevant performance of existing amyloid PET synthesis techniques for Alzheimer's disease.
Baron et al. (Tue,) studied this question.