Mutual Information Better Quantifies Brain Network Architecture in Children with Epilepsy

Area of Science:

• Neuroscience
• Medical Imaging
• Network Science

Background:

• Resting-state fMRI derived brain network architecture metrics offer physiologically relevant markers of IQ in children with epilepsy.
• Traditional functional connectivity (FC) measures, like Pearson correlation, assume linear relationships in BOLD time courses, which may be limiting.

Purpose of the Study:

• To compare network metrics derived from mutual information (MI)-defined FC with traditional correlation-defined FC for predicting patient-level IQ.
• To evaluate the capacity of MI-defined FC to capture complex associations in brain networks beyond linear relationships.

Main Methods:

• Retrospective analysis of 24 children with focal epilepsy and resting-state fMRI data.
• Brain networks constructed using anatomic parcellation (780 nodes) with edges defined by either Pearson correlation or mutual information of BOLD time courses.
• Calculation of network metrics (clustering coefficient, modularity, path length, global efficiency) and prediction of IQ using a machine learning algorithm.

Main Results:

• Mutual information-defined FC network metrics significantly outperformed Pearson correlation in predicting IQ.
• Fractional variation explained for IQ prediction was 49% using MI-defined FC versus 17% using Pearson correlation.
• All constructed brain networks exhibited small-world properties.

Conclusions:

• Mutual information-defined functional connectivity captures more physiologically relevant brain network features than Pearson correlation.
• Improved prediction of cognitive phenotypes using MI-defined FC is a critical step towards clinical utility of network-based biomarkers.