Motivation: Alzheimer's Disease hallmarks include tau deposits and brain atrophy, detectable via PET and MRI, respectively. PET is expensive, invasive, exposes patients to ionizing radiation, whereas MRI is a cheaper, non-invasive, free from ionizing radiation but limited to measuring structural changes. Goal(s): Develop a three-dimensional image-translation model synthesizing tau PET images from T1-weighted MRI, capitalizing on the relationship between tau and brain atrophy. Approach: The model was trained with 297 PET/MRI pairs and validated with 128 pairs Results: The model synthesized tau-PET images from T1-weighter MRI with high-degree of similarity reflected in high mean SSIM and PSNR (SSIM=0.96&PSNR=29.76), and low MAE in SUVR synthesis. Impact: Our model proves the feasibility of synthesizing AV-1451 PET images from T1-weighted MRI, an approach that could enhance the accessibility for large cohort-studies and early dementia detection, while also reducing costs, invasiveness, and enhancing patient safety by limiting radiation exposure.
Lara et al. (2025) studied this question.