TractoFormer: A Novel Fiber-level Whole Brain Tractography Analysis Framework Using Spectral Embedding and Vision Transformers

TractoFormer: A Novel Fiber-level Whole Brain Tractography Analysis Framework Using Spectral Embedding and Vision Transformers.

Introduction. Diffusion MRI (dMRI) tractography in vivo mapping of white matter (WM) connections study of the brain: health and disease Challenges: Defining a good data representation of tractography L imited sample size (number of subjects) Result i nterpretability.

Proposed method. TractoFormer ： A novel parcellation-free whole brain tractography (WBT) analysis framework Leverage tractography information at the level of individual fiber streamlines Provide a natural mechanism for interpretation of results using the self-attention scheme of Vision Transformers ( ViTs ) Major contributions: A novel 2D image representation of WBT, referred to as TractoEmbedding , based on a spectral embedding of fibers from tractography A ViT - based network that performs effective and interpretable group classification..

4. (a) Tractography data Group Subject (b) Embedding space D3 (c) Embedding grid 02 (e) TractoEmbedding image Groupwise embedding New data embedding Coordinate discretization Feature mapping An example and its pint (d) Fiber measure of interest (e.g., mean FA).

5. ViTL Patches Right ViTR Commissural ViTc mean Ensemble Group classification logits Attention maps showing TractoEmbedding regions for classification Right Discriminative identification Commissural.

6. Results: Synthetic data experiment. (a) Left TractoEmbedding FA imagewith synthetic (b) Group-wise attention map Discriminative TractoEmbedding pixels Subject-specific attention map Fib synthetic ges Right Gro discriminative fibers Commissural Su -s— discriminativu fibers Right Commissural.

7. A picture containing chart Description automatically generated.

8. Conclusion. A novel parcellation-free whole brain tractography analysis framework Analysis whole brain tractography data at individual fiber level Result interpretation via ViTs Effective data augmentation.

Thank you!. This work is supported by the following NIH grants: R01MH119222, R01MH125860, P41EB015902, R01MH074794..