Estimating high-order brain functional networks by correlation-preserving embedding.

Area of Science:

• Neuroscience
• Medical Imaging
• Computational Biology

Background:

• Brain functional networks (FNs) are crucial for understanding neurological diseases.
• Traditional methods (e.g., Pearson's correlation, sparse representation) capture limited, low-order brain interactions.
• High-order FNs (HoFNs) show promise for early disease diagnosis but face challenges with noisy or redundant node features.

Purpose of the Study:

• To develop a novel method, correlation-preserving embedding (COPE), for refining low-order FNs (LoFNs) before constructing HoFNs.
• To address the issue of redundant/noisy information in node feature vectors used for HoFN construction.
• To ensure the preservation of original low-order relationships during HoFN generation.

Main Methods:

• Construct traditional LoFNs using sparse representation (SR).
• Apply COPE to embed LoFNs, generating new node representations that remove noise and preserve low-order relationships.
• Estimate HoFNs using SR based on the refined node representations.

Main Results:

• The proposed COPE scheme effectively removes redundant/noisy information from node features.
• COPE successfully maintains the integrity of low-order relationships within the functional networks.
• Experiments on the Alzheimer's Disease Neuroimaging Initiative (ADNI) database demonstrated superior performance in identifying mild cognitive impairment (MCI) compared to baseline methods.

Conclusions:

• COPE offers an effective approach to enhance the construction of HoFNs for improved neurological disease diagnosis.
• The method's ability to preserve low-order relationships while reducing noise is key to its improved performance.
• This technique shows significant potential for the early detection of conditions like MCI.