Constructing Dynamic Functional Networks via Weighted Regularization and Tensor Low-Rank Approximation for Early Mild...

Area of Science:

• Neuroscience
• Medical Imaging
• Machine Learning

Background:

• Brain functional networks are crucial for early mild cognitive impairment (eMCI) classification.
• Existing methods often overlook connection strength and topological attributes in functional networks.
• This limits the accurate reflection of similarities and differences in functional connections among individuals.

Purpose of the Study:

• To develop a novel method for constructing dynamic functional networks (DFN) that incorporates weighted regularization (WR) and tensor low-rank approximation (TLA).
• To apply this method for improved identification of eMCI subjects from healthy controls.
• To enhance the classification performance compared to existing DFN construction techniques.

Main Methods:

• Constructed WR-based DFNs (WRDFN) by introducing a WR term.
• Combined WRDFNs into a third-order tensor for TLA processing to obtain WR-TLA DFNs (WRTDFN).
• Extracted features using the weighted-graph local clustering coefficient and employed t-test for selection, followed by linear SVM classification.

Main Results:

• The proposed method generated DFNs exhibiting scale-free properties.
• Achieved high classification performance: Accuracy (87.07%), Sensitivity (83.44%), Specificity (90.70%), and AUC (0.9431).
• Outperformed comparable methods in eMCI classification.

Conclusions:

• The novel WR-TLA method effectively constructs DFNs for eMCI classification.
• This approach offers significant improvements over existing methods for eMCI detection.
• The findings hold potential reference value for the early diagnosis of Alzheimer's disease (AD).