DCLNet: Double Collaborative Learning Network on Stationary-Dynamic Functional Brain Network for Brain Disease...

Area of Science:

• Neuroscience
• Medical Imaging
• Machine Learning

Background:

• Stationary functional brain networks (sFBNs) and dynamic functional brain networks (dFBNs) from resting-state functional MRI (rs-fMRI) offer complementary insights into brain function.
• Current analyses often focus on sFBNs or dFBNs individually, limiting comprehensive brain disease analysis.
• Existing methods integrating sFBNs and dFBNs overlook crucial inter- and intra-category subject distribution information.

Purpose of the Study:

• To develop a novel method, the Double Collaborative Learning Network (DCLNet), for enhanced brain disease classification.
• To leverage both sFBNs and dFBNs, along with subject distribution information, for improved diagnostic performance.
• To extract complementary features from different levels of brain network representations.

Main Methods:

• Constructed sFBNs and dFBNs from rs-fMRI data using correlation-based methods.
• Employed a collaborative encoder with a prune-graft transformer module to extract and integrate multi-level brain network features (connectivity-based, region-based, network-based).
• Utilized a collaborative contrastive learning module to capture subject distribution patterns for learning discriminative features.

Main Results:

• The DCLNet effectively integrates complementary information from sFBNs and dFBNs.
• The method successfully captures subject distribution information, enhancing feature discriminability.
• Experimental results on two real brain disease datasets demonstrate the superiority of DCLNet over conventional methods.

Conclusions:

• DCLNet offers a superior approach for brain disease classification by synergistically utilizing sFBNs, dFBNs, and subject distribution information.
• The proposed method advances the integration of static and dynamic brain network analysis for clinical applications.
• DCLNet holds significant potential for improving the accuracy and robustness of neuroimaging-based disease diagnosis.