Inter-subject Similarity Guided Brain Network Modeling for MCI Diagnosis

Area of Science:

• Neuroscience
• Computational Biology
• Medical Imaging

Background:

• Sparse representation models for brain networks show high variability, hindering inter-subject comparisons and disease diagnosis.
• Existing group sparse representation methods improve consistency but often lack sufficient separability between different subject groups (e.g., patients vs. controls).

Purpose of the Study:

• To develop an "inter-subject FC similarity-guided" group sparse network modeling method.
• To enhance the separability of brain functional networks between different groups for improved disease diagnosis.
• To balance individual network consistency with group-wise differences.

Main Methods:

• Proposed an "inter-subject FC similarity-guided" group sparse network modeling approach.
• Incorporated inter-subject functional connectivity (FC) similarity as a constraint.
• Estimated FC similarity by comparing Pearson's correlation-based FC patterns between subject pairs.
• Retained group sparsity constraints for individual network consistency.

Main Results:

• The proposed method effectively balances individual network consistency and group differences.
• Achieved improved separability of brain functional networks between different groups.
• Significantly enhanced connectomics-based diagnosis for mild cognitive impairment (MCI).

Conclusions:

• The novel method successfully addresses limitations of previous approaches by enhancing inter-subject similarity within groups and differences between groups.
• This approach offers a more robust framework for computer-aided diagnosis of neurological conditions.
• Improved diagnostic performance for mild cognitive impairment (MCI) demonstrates the clinical potential of this method.