Precise estimation of tissue microstructure with hybrid graph transformer.

Area of Science:

• Medical Imaging
• Neuroscience
• Machine Learning

Background:

• Accurate tissue microstructure estimation via Diffusion MRI (dMRI) requires substantial data, which is clinically challenging to acquire.
• Deep learning methods enhance microstructure inference from undersampled dMRI but often neglect joint spatial (x-space) and diffusion (q-space) information.

Purpose of the Study:

• To propose a novel hybrid graph transformer (HGT) for precise tissue microstructure estimation by integrating q-space learning and x-space guidance.
• To address limitations of existing methods by considering joint information across spatial and diffusion domains.

Main Methods:

• Developed a hybrid graph transformer (HGT) model incorporating a graph convolutional network for q-space learning and residual dense transformer blocks for x-space guidance.
• The x-space module leverages anatomical context to regularize microstructure estimation from undersampled q-space data.

Main Results:

• The HGT model demonstrated superior performance compared to state-of-the-art methods in extensive experiments.
• Evaluations were conducted on data from the Human Connectome Project and diffusion-weighted imaging of Parkinson's disease patients.

Conclusions:

• The proposed HGT effectively integrates spatial and diffusion information for improved tissue microstructure estimation from undersampled dMRI data.
• HGT offers a promising advancement for clinical applications requiring accurate microstructural analysis.