A Novel Explainable Fuzzy Clustering Approach for fMRI Dynamic Functional Network Connectivity Analysis

Area of Science:

• Neuroscience
• Brain Imaging
• Computational Psychiatry

Background:

• Resting state functional magnetic resonance imaging (rs-fMRI) dynamic functional network connectivity (dFNC) analysis is crucial for understanding brain network interactions in neuropsychiatric disorders.
• Current methods using hard clustering to identify dFNC states and quantify feature importance are limited, as they only capture effects on a subset of samples.
• Existing approaches may not accurately reflect the overall impact of dFNC features on brain state clustering.

Purpose of the Study:

• To develop and validate a novel, explainable clustering approach for dFNC analysis.
• To more accurately identify the importance of each dFNC feature to the overall clustering of brain activity states.
• To compare the novel approach with existing methods and demonstrate its improved sensitivity.

Main Methods:

• Utilizing fuzzy clustering to assign probabilities of samples belonging to different brain states.
• Measuring Kullback-Leibler divergence after perturbing individual dFNC features to assess their impact.
• Applying the novel approach to default mode network analysis in individuals with schizophrenia (SZ) and healthy controls.

Main Results:

• The novel fuzzy clustering approach successfully identified significant differences in dFNC state dynamics between individuals with SZ and healthy controls.
• The new method demonstrated superior performance compared to an existing approach, capturing perturbation effects on a larger proportion of samples.
• Key interactions, including those involving the posterior cingulate cortex (PCC), were identified as important across both methods.

Conclusions:

• The developed fuzzy clustering method offers a more comprehensive and explainable approach to analyzing dFNC in rs-fMRI.
• This novel technique enhances the ability to detect group differences in brain network dynamics, particularly in conditions like schizophrenia.
• The approach holds promise for advancing rs-fMRI analysis and broader applications of clustering techniques.