Estimating Functional Connectivity Networks via Low-Rank Tensor Approximation With Applications to MCI Identification

Area of Science:

• Neuroscience
• Network Science
• Biomedical Engineering

Background:

• Functional connectivity networks (FCNs) are crucial for understanding brain function and diagnosing neurodegenerative diseases.
• Existing FCN estimation methods often analyze subjects individually, overlooking shared network structures.
• Current group-constrained methods assume identical FCN topologies, which is often unrealistic, especially with mixed subject groups.

Purpose of the Study:

• To develop a novel FCN estimation approach that accounts for similar, yet not identical, network topologies across subjects.
• To improve the accuracy of identifying biomarkers for neurodegenerative diseases like mild cognitive impairment (MCI).

Main Methods:

• A two-step learning framework was employed for FCN estimation.
• Initial FCNs were independently estimated using methods like Pearson's correlation and sparse representation.
• Refinement was achieved by stacking individual FCNs into a tensor and applying low-rank tensor approximation.

Main Results:

• The proposed method refines FCNs by leveraging shared topological similarities.
• Applied to mild cognitive impairment (MCI) detection, the improved FCNs led to higher classification accuracy compared to traditional methods.
• This demonstrates the potential of the approach for biomarker discovery in neurodegenerative diseases.

Conclusions:

• The novel FCN estimation approach effectively captures similar network topologies within a group.
• This method offers a more realistic and powerful tool for analyzing brain connectivity and diagnosing neurological conditions.
• The findings suggest improved diagnostic capabilities for neurodegenerative diseases through advanced FCN analysis.