Mapping informative clusters in a hierarchical [corrected] framework of FMRI multivariate analysis.

Area of Science:

• Neuroimaging
• Cognitive Neuroscience
• Machine Learning

Background:

• Multivariate pattern analysis (MVPA) is widely used in functional magnetic resonance imaging (fMRI) for brain activity discrimination.
• Voxel-based functional brain mapping using MVPA often yields inconsistent results, complicating the interpretation of brain activity localization.
• Existing methods struggle to provide robust and interpretable functional brain maps.

Purpose of the Study:

• To develop a hierarchical multivariate framework for reliable functional brain mapping in fMRI.
• To enhance the robustness and interpretability of brain mapping by focusing on informative clusters instead of individual voxels.
• To maintain discriminative power while improving the stability of functional mapping results.

Main Methods:

• Proposed a hierarchical framework integrating cluster analysis and multivariate pattern classification.
• Identified local homogeneous clusters with similar voxel response profiles.
• Utilized multi-voxel classifiers within clusters and multivariate ranking to identify informative clusters based on inter-cluster interactions.

Main Results:

• The hierarchical approach demonstrated superior robustness in functional brain mapping compared to traditional voxel-based methods on both simulated and real fMRI data.
• Mapped clusters showed high overlap for perceptually equivalent object categories, validating the approach's reliability.
• The method effectively identified informative clusters, improving the localization of brain activity patterns.

Conclusions:

• The proposed hierarchical framework offers a robust and reliable method for functional brain mapping in fMRI studies.
• This approach effectively balances discriminative power with improved localization accuracy.
• The framework is suitable for both pattern classification and robust brain mapping applications in neuroscience research.