Locality preserving non-negative basis learning with graph embedding.

Area of Science:

• Neuroimaging
• Network Neuroscience
• Computational Neuroscience

Background:

• High-dimensional brain connectivity networks require methods to identify key sub-network patterns.
• Understanding brain pathology and population variations necessitates robust analytical tools.

Purpose of the Study:

• To develop a non-negative component analysis framework for learning localized and sparse sub-network patterns.
• To extract discriminative and reconstructive bases from connectivity matrices.
• To identify population-specific connectivity differences using graph-theoretical approaches.

Main Methods:

• Non-negative component analysis (NCA) framework.
• Decomposition of connectivity matrices into discriminative and reconstructive bases.
• Graph-theoretical scheme for preserving locality and inter-node distances.
• Application to Diffusion Tensor Imaging (DTI) derived connectivity matrices.

Main Results:

• Successfully identified localized and sparse sub-network patterns.
• Demonstrated effectiveness in characterizing brain pathology in Autism Spectrum Disorder (ASD).
• Revealed gender differences in structural brain connectivity during development.

Conclusions:

• The proposed NCA framework is effective for dissecting complex brain connectivity networks.
• This approach facilitates the identification of clinically relevant sub-network biomarkers.
• The method offers insights into population-specific variations in brain structure and function.