Stacking multiple optimal transport policies to map functional connectomes.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Brain Atlases

Background:

• Connectomics relies on brain atlases, but differences in atlas size, topology, and scale prevent cross-atlas analysis.
• This limitation hinders increasing sample sizes and demonstrating generalizability in neuroimaging studies.
• Existing methods like Cross Atlas Remapping via Optimal Transport (CAROT) address this by creating spatial mappings between atlases.

Approach:

• This work introduces Stacking CAROT, an extension of CAROT that combines multiple pre-trained mappings between source and target atlases.
• Stacking CAROT allows for the creation and reuse of mappings, enabling novel combinations for new atlas pairs.
• This method does not require raw neuroimaging data, avoiding time-consuming and costly re-processing.

Key Points:

• Stacking CAROT integrates existing CAROT mappings to generate new ones, enhancing flexibility.
• Reconstructed connectomes using Stacking CAROT demonstrate comparable performance to CAROT in downstream analyses.
• Stacking CAROT significantly reduces training time and storage requirements compared to the original CAROT.

Conclusions:

• Stacking CAROT offers an efficient and flexible extension of the CAROT framework for neuroimaging connectomics.
• This approach improves upon previous CAROT versions by enabling the combination of mappings, reducing computational burden.
• Stacking CAROT facilitates more robust and generalizable connectomic analyses across diverse neuroimaging datasets.