Diffusion MRI harmonization via personalized template mapping.

Area of Science:

• Neuroimaging
• Medical Physics
• Biomedical Engineering

Background:

• Diffusion MRI (dMRI) harmonization faces challenges separating scanner effects from brain anatomy.
• Conventional methods struggle with persistent neuroanatomical misalignment post-registration, especially near cortical boundaries.
• Existing approaches using common templates limit accuracy in dMRI data harmonization.

Purpose of the Study:

• To develop a personalized framework for dMRI harmonization that overcomes limitations of conventional methods.
• To effectively address confounding effects from neuroanatomical misalignment in dMRI harmonization.
• To improve the accuracy and reliability of dMRI data harmonization across different scanning sites.

Main Methods:

• Proposed a personalized framework for dMRI harmonization, adapting personalized templates for source and target sites.
• Integrated the framework with rotation invariant spherical harmonics (RISH) features for dMRI signal harmonization.
• Applied and compared the method to dMRI data from Siemens Prisma and GE MR750 platforms within the Adolescent Brain Cognitive Development dataset.

Main Results:

• The personalized harmonization framework significantly reduced inter-site variations caused by scanner differences.
• The method demonstrated superior performance in preserving sex-related biological variability in original cohorts.
• Harmonization using the personalized procedure showed robustness in maintaining original fiber orientation distributions.

Conclusions:

• The proposed personalized framework offers a more effective solution for dMRI harmonization compared to conventional methods.
• This approach enhances the reliability of dMRI data by reducing scanner-induced variability while preserving biological signals.
• The personalized harmonization method is robust and preserves crucial information for downstream analyses, such as fiber orientation estimation.