Researchers Develop AI Tool to Reconstruct Missing Brain Connectivity Data

Diffusion MRI helps map nerve fiber pathways in the brain, but some scans lose signal in regions that are important for understanding connectivity. These gaps can affect areas such as the brainstem and orbitofrontal region, where magnetic field distortions may erase imaging data and disrupt brain wiring maps. This can limit research into how neural connections change with aging and disease. To recover these gaps, researchers have developed an AI model that restores missing information in brain connectivity maps.

Researchers at the Mark and Mary Stevens Neuroimaging and Informatics Institute (Stevens INI) at the Keck School of Medicine of USC (Los Angeles, CA, USA) developed the model, called FOD-Diffusion. The technology focuses on fiber orientation distributions, which consist of 45 linked three-dimensional brain maps for each scan. FOD-Diffusion restores one map at a time while using information from the remaining maps and surrounding well-measured tissue to maintain consistency in the reconstructed fiber patterns.

The team trained and tested FOD-Diffusion using scans from 1,315 participants in the UK Biobank and 679 participants in the Lifespan Human Connectome Project Aging study, both large studies involving mostly older adults. The model was described in a study published in Medical Image Analysis on November 3, 2026.

To evaluate accuracy, the researchers deliberately hid brainstem sections in scans with minimal signal loss and asked the model to reconstruct the missing information. Across 43 test participants, the primary fiber directions generated by FOD-Diffusion differed from the original data by about 1.4 degrees on average. By comparison, the closest-performing of six other artificial intelligence approaches differed by about 4.5 degrees.

Additional testing examined whether the model could correct signal loss in regions commonly affected by distortion. In 1,172 UK Biobank participants with signal loss, standard correction tools alone left brainstem fiber information significantly weaker than in participants with minimal signal loss. After restoration with FOD-Diffusion, that difference was no longer significant. Similar results were observed in the orbitofrontal region among 981 participants. Corticospinal tract traces in participants with severe signal loss also appeared more complete after FOD-Diffusion restoration than with standard-corrected data alone.

The researchers emphasized that the reconstructed data represent informed estimates rather than recovered measurements. Because no human diffusion MRI scan is completely free of distortion, the model’s accuracy was evaluated against scans from participants with minimal signal loss rather than against a perfect reference. The approach still requires testing across additional scanners and imaging settings and has not yet been evaluated for use in patient care.

“Some of the brain's most important pathways run through regions that have been especially difficult to image,” said Arthur W. Toga, Ph.D., director of the Stevens INI. “By sharing this tool openly, the team is giving researchers a way to strengthen analyses of the large data sets the field depends on.”

Related Links
Keck School of Medicine of USC