A Novel Registration Framework for Aligning Longitudinal Infant Brain Tensor Images.

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Developmental Neuroscience

Background:

• Infant brain imaging presents registration challenges due to rapid developmental changes.
• Diffusion tensor imaging (DTI) offers consistent tissue properties for infant brain registration.
• Groupwise registration is preferred in infant studies to avoid atlas bias.

Purpose of the Study:

• To develop a novel groupwise registration method for longitudinal infant DTI images.
• To align infant DTI data to a sample-specific common space, reducing registration bias.

Main Methods:

• Proposed a novel groupwise registration approach for longitudinal infant DTI images.
• Clustered DTI images into subgroups using Louvain clustering based on image similarity.
• Applied tensor-based registration within subgroups and then aligned subgroups to a common space.

Main Results:

• The novel approach significantly improved global and local registration accuracy compared to standard methods.
• Clustering based on image similarity enhanced registration accuracy more than no clustering.
• Registration accuracy with image similarity clustering was comparable to chronological age clustering.

Conclusions:

• The developed groupwise registration framework accurately aligns longitudinal infant brain DTI images.
• This method capitalizes on consistent tensor map features in early infancy.
• Facilitates more precise alignment for infant neuroimaging research.