Modeling a Shared Reality of Tractography through Varied Structural Imaging.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Biomedical Engineering

Background:

• Diffusion MRI (dMRI) is standard for white matter tractography, but its specificity is debated.
• Understanding what dMRI uniquely captures versus general structural properties is crucial.

Purpose of the Study:

• To investigate if white matter tractography patterns reflect diffusion-specific phenomena or general structural information.
• To develop and validate a framework for extracting white matter pathways from FLAIR images without subject-specific anatomical context.

Main Methods:

• Introduced a teacher-student model framework to guide FLAIR-based tractogram creation using dMRI-derived systemic information.
• The 'teacher' model trained on dMRI features, while the 'student' model used frozen teacher layers to extract tractography features solely from FLAIR input.
• Evaluated robustness on withheld subjects and compared FLAIR-generated streamlines to dMRI streamlines using bundle adjacency and Dice coefficient across 39 white matter bundles.

Main Results:

• The proposed method demonstrated statistically similar performance to other non-diffusion tractography algorithms when compared against diffusion streamlines.
• FLAIR-template generated streamlines showed comparable accuracy to gold-standard diffusion streamlines.
• Statistical evaluations confirmed the method's efficacy against alternative non-diffusion approaches.

Conclusions:

• Tractography patterns may originate from a shared latent space of structural information, not exclusive to any single imaging sequence.
• The developed method captures unconditional, subject-specific prior probabilities of tractography without diffusion data.
• This suggests alternative imaging modalities like FLAIR can yield valuable tractography insights.