Explainable fuzzy clustering framework reveals divergent default mode network connectivity dynamics in schizophrenia

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Psychiatric Disorders

Background:

• Dynamic functional network connectivity (dFNC) analysis reveals insights into neurological and neuropsychiatric disorders.
• Hard clustering methods in dFNC analysis obscure network dynamics and limit subject comparison.
• Fuzzy clustering offers a dimensional approach to connectivity patterns, mitigating limitations of hard clustering.

Purpose of the Study:

• Introduce an explainable fuzzy clustering framework for dFNC analysis.
• Combine fuzzy c-means clustering with explainability metrics and novel summary features.
• Apply the framework to analyze default mode network dynamics in schizophrenia.

Main Methods:

• Extracted dFNC from individuals with schizophrenia (SZ) and controls.
• Identified 5 dFNC states using a perturbation-based clustering explainability approach.
• Quantified state dynamics using traditional and novel fuzzy clustering features, examining differences between SZ and controls and correlating with symptom severity.

Main Results:

• Individuals with SZ spent more time in states with moderate and strong anticorrelation between specific brain regions (anterior/posterior cingulate cortex, precuneus).
• SZ individuals exhibited more rapid transitions between low-magnitude and high-magnitude dFNC states compared to controls.
• Identified key dFNC features crucial for characterizing states in SZ.

Conclusions:

• Presented a novel dFNC analysis framework with enhanced insight into network dynamics.
• Demonstrated the framework's potential for identifying effects of SZ on network dynamics.
• Highlighted the framework's ease of implementation and potential for future dFNC studies.