Estimating High-Order Brain Functional Networks in Bayesian View for Autism Spectrum Disorder Identification

Area of Science:

• Neuroscience
• Computational Biology
• Biostatistics

Background:

• Brain functional networks (BFNs) are crucial for understanding brain organization and neurological disorders.
• Pearson's correlation (PC) is standard for BFN construction, but higher-order methods like correlation's correlation (CC) lack theoretical grounding.
• Existing methods struggle with subtle connection changes and lack a robust theoretical foundation.

Purpose of the Study:

• To provide a rigorous Bayesian framework for understanding and improving high-order BFN estimation.
• To develop a novel Bayesian high-order method (BHM) for simultaneous estimation of high- and low-order BFNs.
• To enhance biomarker discovery for neurological disorders using advanced BFN analysis.

Main Methods:

• Reformulated correlation's correlation (CC) within a Bayesian framework using a matrix-variate normal distribution prior.
• Developed a Bayesian high-order method (BHM) for automated and simultaneous BFN estimation.
• Proposed an efficient optimization algorithm for the BHM framework.

Main Results:

• Achieved a probabilistic explanation for CC, providing a solid theoretical foundation.
• The BHM successfully identified subjects with autism spectrum disorder (ASD) from typical controls.
• Automatically learned high- and low-order BFNs using BHM outperformed conventional PC and CC methods.

Conclusions:

• The proposed Bayesian high-order method (BHM) offers a robust and theoretically sound approach to BFN analysis.
• BHM demonstrates superior performance in identifying neurological disorders like ASD compared to existing methods.
• This framework advances the field of BFN learning and biomarker discovery for neurological conditions.