Enhanced hyperalignment via spatial prior information.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Statistical Modeling

Background:

• Functional alignment is crucial for group-level functional magnetic resonance imaging (fMRI) analysis but is often imperfect.
• Existing hyperalignment methods, while useful, have limitations in interpretability, uniqueness, and whole-brain application.

Purpose of the Study:

• To introduce the ProMises (Procrustes von Mises-Fisher) model for statistically robust functional alignment in fMRI.
• To address shortcomings of current hyperalignment techniques, enabling unique and interpretable transformations.
• To facilitate efficient whole-brain functional alignment for improved inter-subject analysis.

Main Methods:

• Reformulating functional alignment as a statistical model using Procrustes transformations.
• Incorporating a von Mises-Fisher prior distribution on orthogonal parameters.
• Embedding anatomical information by penalizing spatially distant voxels in shared space creation.

Main Results:

• ProMises provides unique and interpretable alignment transformations and aligned images.
• The method allows for efficient whole-brain functional alignment.
• Simulations and real fMRI data analysis show improved inter-subject classification accuracy and interpretability compared to standard hyperalignment.

Conclusions:

• The ProMises model offers a statistically principled and more effective approach to functional alignment in fMRI.
• This advancement enhances the reliability and interpretability of group-level fMRI analyses.
• ProMises represents a significant improvement for inter-subject functional data analysis in neuroscience.