Explainable Fuzzy Clustering Framework Reveals Divergent Default Mode Network Connectivity Dynamics in Schizophrenia

Area of Science:

• Neuroscience
• Computational Neuroscience
• Psychiatry

Background:

• Dynamic functional network connectivity (dFNC) analysis using resting-state fMRI aids in understanding neurological and neuropsychiatric disorders.
• Traditional hard clustering methods in dFNC analysis obscure network dynamics and pose challenges for subject comparison.

Approach:

• Developed an explainable fuzzy clustering framework combining fuzzy c-means with explainability metrics.
• Applied the framework to analyze default mode network dynamics in schizophrenia (SZ).
• Identified 5 distinct brain states and characterized them using novel explainability approaches.

Key Points:

• The framework extracts features comparable to hard clustering but offers unique metrics for state dynamics.
• Demonstrated the framework's ability to identify effects of SZ on network dynamics.
• Uncovered relationships between SZ symptom severity and specific brain region interactions (precuneus, anterior/posterior cingulate cortex).

Conclusions:

• The explainable fuzzy clustering framework provides enhanced insights into brain network dynamics compared to hard clustering methods.
• The framework's ease of implementation and detailed analysis capabilities make it valuable for future dFNC research in clinical populations.