TRANSFORMER-BASED T1-TRACTOGRAPHY.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Medical Image Analysis

Background:

• Diffusion MRI (dMRI) streamline tractography is the standard for mapping brain white matter (WM) pathways.
• Deep learning now enables streamline generation from T1-weighted (T1w) MRI, a more accessible imaging modality.
• Current T1w tractography methods face accuracy limitations due to their recurrent neural network architectures.

Purpose of the Study:

• To enhance the accuracy and performance of T1w-based streamline tractography.
• To investigate the utility of Transformer modules for improving fiber orientation distribution prediction.
• To overcome the limitations of recurrent architectures in current T1w tractography methods.

Main Methods:

• Modified a state-of-the-art T1w tractography method (CoRNN) by replacing recurrent units with Transformer modules.
• Altered both the representation and prediction networks for fiber orientation distributions.
• Applied the modified method to generate streamlines from T1w MRI data in healthy adults.

Main Results:

• The Transformer-based approach demonstrated substantial performance improvements over the baseline recurrent method.
• Achieved high angular consistency between generated streamlines and gold-standard dMRI tractograms.
• Successfully generated accurate white matter pathway estimations from T1w MRI.

Conclusions:

• Transformer modules offer significant advantages over recurrent architectures for T1w tractography.
• This novel approach provides a more accurate and efficient method for non-invasive brain pathway estimation.
• The findings pave the way for improved non-invasive neuroimaging analysis using T1w MRI.