Dynamic functional connectivity in temporal lobe epilepsy: a graph theoretical and machine learning approach.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Medical Diagnostics

Background:

• Resting-state functional magnetic resonance imaging (fMRI) assesses brain organization without tasks.
• Functional connectivity (FC) is dynamic and varies over time.
• Temporal lobe epilepsy (TLE) diagnosis requires identifying the epileptogenic hemisphere.

Purpose of the Study:

• To leverage dynamic FC characteristics and graph theory for TLE laterality identification.
• To apply machine learning for distinguishing left vs. right epileptogenic zones in TLE.

Main Methods:

• Extracted six global graph measures from static and dynamic FC matrices using fMRI data from 35 TLE subjects.
• Quantified alterations in graph measure time trends.
• Employed Random Forest for feature selection and Support Vector Machine for classification of epileptogenic hemisphere.

Main Results:

• Dynamic FC features improved classification accuracy to 88.5% compared to 82% for static features.
• Optimized feature selection further elevated classification accuracy to 91.5%.
• Demonstrated the superior performance of dynamic over static FC analysis for TLE laterality.

Conclusions:

• Dynamic connectivity analysis of graph measures effectively captures non-stationary FC characteristics.
• Machine learning integration with dynamic FC analysis can identify temporal network feature trends.
• These network features show promise as potential imaging biomarkers for determining the epileptogenic hemisphere in TLE.