A Brain Connectivity Approach to Detect Diffusion-Weighted Imaging Changes in Post-Traumatic Epilepsy.

Area of Science:

• Neuroscience
• Medical Imaging
• Machine Learning

Background:

• Traumatic brain injury (TBI) is a primary cause of acquired epilepsy, leading to post-traumatic epilepsy (PTE) months or years after injury.
• Predicting PTE development using reliable imaging biomarkers is a significant clinical challenge.
• Diffusion-weighted imaging (DWI) and structural connectome analysis show promise in identifying brain network alterations linked to late seizures.

Purpose of the Study:

• To classify TBI patients into seizure-free and late seizure-affected groups.
• To identify anatomical regions associated with late seizure development after TBI.
• To explore the utility of machine learning in detecting predictive patterns in neuroimaging data for PTE.

Main Methods:

• Analysis of 59 DWI scans from the EpiBioS4Rx project (42 seizure-free, 17 late seizure-affected TBI patients).
• Application of a Random Forest classification algorithm.
• Incorporation of network feature importance using the Gini index for model interpretation.

Main Results:

• Achieved 69% ± 0.03 accuracy and 73% AUC ± 0.05 for binary classification between seizure-free and seizure-affected TBI patients.
• Identified specific brain regions potentially associated with epileptogenesis.
• Demonstrated accurate classification results compared to existing literature, despite dataset limitations.

Conclusions:

• The study successfully identified potential imaging biomarkers for predicting PTE in TBI patients.
• Machine learning analysis of DWI data can aid in understanding brain network alterations leading to late seizures.
• Further research with larger datasets is warranted to validate these findings and improve predictive accuracy for PTE.