A generative AI model is rebuilding brain wiring where MRI scans lose signal.

FOD-Diffusion repairs blank brain scans by guessing from patterns. A plausible pathway can look more real than the missing signal.

A missing patch in the pons becomes a finished-looking pathway when FOD-Diffusion fills it. The USC model may recover structure distortion hides, but it can also impose what scans from 1,315 UK Biobank and 679 HCP-Aging participants taught it to expect. That matters because tractography already produces false positives. In brain mapping, cleaner evidence can quietly become invented anatomy.

• What it could changeReliable reconstruction could reveal how brain connections shift with aging and disease.
• How it touches todayMRI researchers already risk mistaking model-made tissue for pathways in noisy scans.
• Our futureBrain maps could grow more complete while the line between measurement and invention fades.

• Local tractography is prone to error accumulation and high false-positive rates on noisy or low-resolution data, the problem generative global methods aim to reduce.arxiv
• A 2024 preprint specifically evaluates synthetic diffusion MRI maps made with diffusion denoising probabilistic models, showing that synthetic-map fidelity is an active research problem.doi
• A systematic review reports that deep-learning MRI systems can erase lesions or invent tissue, while common pixel-averaged metrics may fail to detect those errors.arxiv