Learning functional brain networks with heterogeneous connectivities for brain disease identification

Area of Science:

• Neuroscience
• Computational Biology
• Biomedical Engineering

Background:

• Functional brain networks (FBNs) are crucial for understanding brain function and identifying neurological/mental disorder biomarkers.
• Current FBN estimation methods produce homogeneous networks, limiting the accurate encoding of complex brain interactions.
• This homogeneity hinders the discovery of subtle yet significant patterns in brain connectivity.

Purpose of the Study:

• To propose the existence of heterogeneous functional brain networks (FBNs) for the first time.
• To introduce a novel FBN estimation model capable of adaptively assigning heterogeneous connections.
• To enhance the accurate encoding of complex brain interactions and improve biomarker discovery.

Main Methods:

• Constructing multiple candidate correlation types from diverse perspectives or methods.
• Developing an improved orthogonal matching pursuit algorithm for adaptive connection selection.
• Utilizing label information to guide the selection of at most one correlation per brain region pair.

Main Results:

• The proposed heterogeneous FBNs significantly improved classification performance by 7.07% and 7.58% across two independent datasets.
• Demonstrated the ability of the novel method to distinguish individuals with neurological/mental disorders from healthy controls.
• Successfully identified potential biomarkers associated with these disorders using the enhanced network representations.

Conclusions:

• The findings support the hypothesis of heterogeneity in functional brain networks.
• The developed heterogeneous connection assignment algorithm is effective for encoding complex brain interactions.
• This approach offers a more accurate and sensitive method for neurological and mental disorder research.