Fusing Multiview Functional Brain Networks by Joint Embedding for Brain Disease Identification.

Area of Science:

• Neuroscience
• Medical Imaging
• Data Science

Background:

• Resting-state functional MRI (rs-fMRI) is crucial for identifying brain disorders like autistic spectrum disorder (ASD).
• Existing methods for estimating functional brain networks (FBNs) often use a single view, limiting their ability to capture complex brain interactions.
• There is a need for advanced methods to integrate information from multiple FBN estimation strategies.

Purpose of the Study:

• To develop a multiview FBN fusion strategy using joint embedding for improved ASD identification from rs-fMRI data.
• To leverage common information across different FBN estimation methods for a more comprehensive brain network analysis.
• To identify potential neuroimaging biomarkers for ASD diagnosis.

Main Methods:

• Proposed a novel method for fusing multiview FBNs through joint embedding.
• Utilized tensor factorization to learn a common factor (joint embedding) for each region of interest (ROI) across different FBNs.
• Reconstructed a new FBN by calculating Pearson correlations between embedded ROIs.

Main Results:

• The proposed fusion method achieved superior performance in automated ASD diagnosis compared to state-of-the-art single-view FBN methods on the ABIDE dataset.
• The framework attained an accuracy of 74.46% in ASD identification.
• Demonstrated significant performance improvement over other multi-network methods, with an accuracy increase of at least 2.72%.

Conclusions:

• Introduced an effective multiview FBN fusion strategy via joint embedding for fMRI-based ASD identification.
• The method offers a robust approach to analyzing complex functional brain connectivity.
• The framework provides theoretical grounding from the perspective of eigenvector centrality and aids in discovering potential ASD biomarkers.