Remodeling Pearson's Correlation for Functional Brain Network Estimation and Autism Spectrum Disorder Identification.

Area of Science:

• Neuroscience
• Computational Biology
• Medical Imaging

Background:

• Functional brain networks (FBNs) model neural dependencies and serve as biomarkers for neurological disorders.
• Pearson's Correlation (PC) is widely used for FBN construction but often yields dense, noisy networks requiring arbitrary sparsification.
• Existing sparsification methods lack flexibility and biological plausibility.

Purpose of the Study:

• To develop a novel, flexible approach for estimating FBNs by reformulating PC as an optimization problem.
• To incorporate biological/physical priors into FBN construction for more accurate network representation.
• To improve the identification of neurological disorders using enhanced FBNs.

Main Methods:

• Remodeled Pearson's Correlation (PC) as an optimization problem incorporating an L1-norm regularizer for sparse FBNs.
• Developed a weighted counterpart for learning sparse and scale-free networks.
• Applied the method to identify autism spectrum disorders (ASD) from normal controls (NC) using constructed FBNs.

Main Results:

• The proposed L1-norm regularized optimization framework provides an elegant and flexible method for sparsifying PC-based FBNs.
• The weighted network counterpart allows for learning both sparse and scale-free FBNs.
• Achieved 81.52% classification accuracy in identifying ASD from NC, outperforming baseline and state-of-the-art methods.

Conclusions:

• The novel optimization approach offers a more mathematically rigorous and biologically informed method for FBN construction.
• This framework provides a versatile platform for integrating prior knowledge into FBN analysis.
• The improved FBN estimation demonstrates significant potential for enhancing diagnostic capabilities in neurological and psychological disorders, particularly ASD.