atlas-free-brain-network-transformer

To generate subject-level embeddings from rs-fMRI data without an atlas, use atlas-free individualized parcellation and ROI-to-voxel connectivity features. This approach minimizes atlas bias and feeds into a transformer model to create subject-specific brain network embeddings.

Atlas-free individualized parcellation is a neuroimaging technique that derives brain regions directly from subject rs-fMRI data. It avoids fixed parcellations that misalign with individual anatomy, thereby improving sensitivity to subject-specific brain network patterns.

You can use transformer models for cross-subject comparisons by inputting ROI-to-voxel connectivity features derived from individualized parcellation. The transformer generates subject-level embeddings that capture unique network properties for direct comparison across individuals.

Yes, the atlas-free brain network transformer works for brain-age estimation and biomarker discovery. By generating subject-level embeddings from rs-fMRI data, it captures individualized network patterns suitable for these downstream neuroimaging tasks.

Fixed atlas-based parcellations reduce sensitivity because they can misalign with individual anatomy. This misalignment hinders cross-subject comparability and masks subject-specific patterns, which atlas-free parcellation solves by deriving regions directly from rs-fMRI data.

Transformer-based brain network embedding generation requires ROI-to-voxel connectivity features. These features, derived from atlas-free individualized parcellation of rs-fMRI data, provide the spatial relationship inputs needed for the transformer model.