Assessing Multi-Site rs-fMRI-Based Connectomic Harmonization Using Information Theory.

Area of Science:

• Neuroimaging
• Network Neuroscience
• Data Science

Background:

• Site-specific biases in multi-site resting-state functional magnetic resonance imaging (rs-fMRI) data complicate analysis and compromise results.
• Harmonization techniques are crucial for large cohorts requiring data independence from scanner variations.
• No comprehensive assessment of harmonization technique efficacy has been previously available.

Purpose of the Study:

• To evaluate the effectiveness of existing harmonization techniques for rs-fMRI connectomics data.
• To introduce a novel methodology for comparing different harmonization models using Information Theory.
• To identify which harmonization frameworks successfully mitigate site-specific biases.

Main Methods:

• Application of Information Theory tools to assess harmonization effectiveness.
• Development of a comparative methodology for evaluating multiple harmonization models.
• Testing the methodology on widely used harmonization frameworks and datasets.

Main Results:

• Demonstrated the utility of the developed Information Theory-based methodology.
• Identified that some widely used harmonization techniques are ineffective.
• Showed that the acquisition site can still be determined from rs-fMRI data after applying certain harmonization methods.

Conclusions:

• The proposed Information Theory approach provides a robust method for assessing rs-fMRI harmonization efficacy.
• Current harmonization techniques exhibit variable effectiveness, with some failing to remove site-specific information.
• Further development of robust harmonization strategies is necessary for reliable multi-site rs-fMRI network analysis.