Weighted Graph Regularized Sparse Brain Network Construction for MCI Identification.

Area of Science:

• Neuroscience
• Medical Imaging
• Computational Biology

Background:

• Resting-state functional magnetic resonance imaging (rs-fMRI) is crucial for analyzing brain diseases.
• Brain functional networks (BFNs) are key for diagnosing conditions like Alzheimer's disease and mild cognitive impairment (MCI).
• Traditional sparse representation (SR) methods for BFNs lack the ability to capture intrinsic data locality and similarity.

Purpose of the Study:

• To develop an optimized framework for constructing robust BFNs.
• To address the limitations of independent coding in SR methods for BFN construction.
• To improve the accuracy of identifying neurological disorders like MCI through enhanced BFN analysis.

Main Methods:

• Proposed a novel weighted graph (Laplacian) regularized sparse representation (SR) framework.
• Optimized BFN construction by integrating intrinsic correlation similarity and local manifold structure.
• Addressed and resolved the non-convergence issue of graph Laplacian in self-representation models.
• Implemented a pipeline combining sparse feature selection and classification for MCI identification.

Main Results:

• The proposed method effectively constructs BFNs by considering data similarity and local structure.
• The framework successfully overcomes the limitations of standard SR in capturing data characteristics.
• The developed pipeline demonstrated effectiveness in identifying MCI using the constructed BFNs.
• The graph Laplacian non-convergence problem within the self-representation model was successfully solved.

Conclusions:

• The novel weighted graph regularized SR framework offers an improved approach to BFN construction.
• This method enhances the analysis of brain connectivity for neurological disease diagnosis.
• The study successfully identified MCI, highlighting the potential of the proposed technique in clinical applications.