Multimodal surface matching: fast and generalisable cortical registration using discrete optimisation.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Brain Mapping

Background:

• Accurate surface-based, cross-subject alignment is crucial for group neuroimaging studies of the cerebral cortex.
• Current registration methods, relying on cortical folding, show suboptimal alignment for certain functional brain regions.
• Integrating multimodal brain data remains challenging due to limitations in existing alignment techniques.

Purpose of the Study:

• To develop a flexible framework for robust alignment of multiple neuroimaging modalities.
• To improve the accuracy of surface-based registration for investigating brain architecture, function, and connectivity.
• To enable better integration of complementary brain datasets in group studies.

Main Methods:

• Adapted the Fast Primal-Dual (Fast-PD) approach for discrete Markov Random Field (MRF) optimization to spherical registration.
• Reframed deformation labels as a discrete set of rotations, incorporating a novel regularization term based on geodesic distance between rotation matrices.
• Proposed a new multivariate cost function using a discretized graph-based mutual information measure.

Main Results:

• Demonstrated improved alignment using scalar metrics (curvature, myelination) and multivariate features (functional task performance).
• The proposed method shows potential for enhanced integration of complementary brain data.
• Results indicate superior alignment compared to traditional folding-based methods for specific functional subregions.

Conclusions:

• The novel spherical registration framework offers a flexible and robust solution for multimodal neuroimaging data alignment.
• This approach facilitates more accurate cross-subject analysis, advancing the study of brain architecture, function, and connectivity.
• The method holds significant promise for future research integrating diverse brain datasets.