Dynamic Functional Network Connectivity Clustering and Harmonization Evaluation Metric

Area of Science:

• Neuroscience
• Brain Imaging
• Data Science

Background:

• Multi-site neuroscience datasets enhance research but introduce site effects.
• Dynamic functional network connectivity (dFNC) analysis relies on clustering temporal functional connectivity patterns.
• Site effects can confound dFNC state identification in combined datasets.

Purpose of the Study:

• To investigate the impact of ComBat harmonization on dFNC states in mild traumatic brain injury (mTBI) data from two sites.
• To introduce and utilize an 'Inclusivity' model for assessing site effect influence on dFNC clustering.

Main Methods:

• Applied ComBat harmonization to dFNC data from two mTBI studies.
• Clustered dFNC states to identify recurring connectivity patterns.
• Developed and employed an 'Inclusivity' model to quantify sample distribution across clusters concerning site effects.

Main Results:

• Site effects were found to significantly influence the formation and characteristics of dFNC states.
• ComBat harmonization reduced site-specific biases in the dFNC data.
• The 'Inclusivity' model effectively measured the impact of site effects on dFNC clustering before and after harmonization.

Conclusions:

• Data harmonization is crucial for mitigating site effects in multi-site dFNC studies.
• The 'Inclusivity' model provides a novel metric for evaluating clustering robustness against site variations.
• Findings support the use of harmonization techniques for reliable dFNC analysis in diverse datasets.