Learning Pairwise-Similarity Guided Sparse Functional Connectivity Network for MCI Classification

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Medical Diagnostics

Background:

• Sparse representation (SR) and weighted SR (WSR) show promise for diagnosing Alzheimer's disease and mild cognitive impairment (MCI) using resting-state functional MRI (RS-fMRI).
• Traditional SR/WSR methods independently process brain regions, neglecting potential inter-regional relationships crucial for accurate network analysis.

Purpose of the Study:

• To develop an improved functional connectivity (FC) modeling approach for RS-fMRI data.
• To incorporate inter-regional relationships into SR/WSR frameworks via regularization for joint representation learning.
• To enhance the diagnostic accuracy for MCI and improve the modularity of brain FC networks.

Main Methods:

• Proposed a novel FC modeling approach integrating two types of inter-regional relationships as regularization terms into SR/WSR.
• Developed an efficient alternating optimization algorithm to solve the proposed model.
• Evaluated the method's performance in diagnosing MCI subjects using RS-fMRI data.

Main Results:

• The proposed method significantly outperformed traditional SR and WSR in diagnosing MCI subjects.
• The novel approach resulted in brain FC networks with improved modularity structure.
• Joint learning of regional representations captured inter-regional relationships more effectively.

Conclusions:

• The novel FC modeling approach offers superior performance for MCI diagnosis compared to existing SR/WSR methods.
• Incorporating inter-regional relationships enhances the biological relevance and diagnostic utility of brain networks derived from RS-fMRI.
• The developed method provides a more robust framework for understanding brain connectivity in neurodegenerative diseases.