Tissue Probability Based Registration of Diffusion-Weighted Magnetic Resonance Imaging

Area of Science:

• Neuroimaging
• Biomedical Engineering
• Computer Science

Background:

• Current diffusion MRI (dMRI) registration methods primarily focus on white matter (WM).
• Emerging dMRI applications for gray matter (GM) characterization necessitate registration methods that encompass all tissue types.
• This highlights a gap in existing dMRI registration techniques.

Purpose of the Study:

• To develop an advanced dMRI registration method.
• The method utilizes tissue probability maps (TPMs) for gray matter (GM), white matter (WM), and cerebrospinal fluid (CSF).

Main Methods:

• A joint segmentation-registration strategy was employed.
• Diffusion tensor imaging (DTI) maps were converted into GM, WM, and CSF TPMs using machine learning.
• These TPMs served as features for aligning dMRI data, with validation against existing tools.

Main Results:

• The proposed method demonstrated superior performance compared to mainstream registration tools.
• Key improvements included reduced voxel-wise variance in registered DTI maps (10% SD decrease) and enhanced similarity of registered TPMs (0.1-0.2 Dice increase).
• Statistical significance (P < 0.05) was observed for all tissue types and across participants.

Conclusions:

• A novel joint segmentation-registration approach using diffusion-derived TPMs offers more accurate dMRI data registration.
• This method effectively translates diffusion data into structural information (TPMs) for direct alignment of diffusion and structural images.
• The developed technique surpasses current registration tools in accuracy and consistency.