Multi-subject Manifold Alignment of Functional Network Structures via Joint Diagonalization.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Medical Image Analysis

Background:

• Functional magnetic resonance imaging (fMRI) group studies require accurate cross-subject correspondence for comparing brain function.
• Current registration methods often rely on morphology, neglecting functional localization variability, which can introduce confounds in disease studies.

Purpose of the Study:

• To develop a novel multi-subject functional registration method that accounts for functional localization variability.
• To improve the specificity and accuracy of cross-subject functional brain comparisons in fMRI studies.

Main Methods:

• Encoding subject-specific functional network structure using diffusion maps to decouple functional relationships from spatial position.
• Employing a two-step manifold alignment to establish initial correspondences between functionally equivalent regions.
• Utilizing coupled joint diagonalization to create common eigenbases across subjects and refine functional correspondences.

Main Results:

• The proposed coupled joint diagonalization method significantly enhances matching accuracy compared to existing functional alignment techniques.
• Demonstrated superior performance over methods relying solely on structural correspondences for functional alignment.
• Validated on fMRI data from a language task, showing improved alignment precision.

Conclusions:

• Multi-subject functional registration via manifold alignment and coupled joint diagonalization offers a more accurate approach to cross-subject brain analysis.
• This method addresses limitations of morphology-based registration, reducing confounds in disease research.
• The technique shows promise for advancing group-level fMRI studies, particularly in understanding functional brain organization.