Similarity-guided multi-view functional brain network fusion.

Area of Science:

• Neuroscience
• Computational Psychiatry
• Medical Imaging Analysis

Background:

• Functional brain networks (FBNs) are vital for understanding mental disorders.
• Integrating diverse neuroimaging data offers comprehensive insights but faces challenges in maintaining data integrity.
• Heterogeneous data fusion is critical for accurate brain function analysis and disorder diagnosis.

Purpose of the Study:

• To develop an innovative fusion model for multi-view FBNs that preserves shared geometric structures.
• To address challenges in integrating heterogeneous neuroimaging data, including small sample sizes and high dimensionality.
• To improve the diagnostic accuracy and biomarker discovery for mental disorders using advanced data fusion techniques.

Main Methods:

• Proposed a novel manifold regularization term to capture and preserve common geometric structures across FBN views.
• Introduced a pairwise regularization function to maximize similarity between different views, integrating complementary information.
• Developed a dual-regularization framework for robust graph construction with heterogeneous neuroimaging data.

Main Results:

• The proposed model outperformed state-of-the-art diagnostic methods on the ABIDE dataset.
• Preserving geometric structures significantly improved diagnostic accuracy for Autism Spectrum Disorder (ASD).
• Identified key ASD biomarkers in the primary visual cortex, consistent with recent high-impact findings.

Conclusions:

• The dual-regularization fusion model effectively integrates multi-view FBN data while preserving essential geometric structures.
• This approach offers a robust framework for analyzing complex brain networks and advancing mental disorder diagnosis.
• The findings underscore the importance of geometric structure preservation in neuroimaging data fusion for biomarker discovery and clinical applications.