Machine Learning of Diffusion Weighted Imaging for Prediction of Seizure Susceptibility Following Traumatic Brain Injury

Abstract

Post-traumatic epilepsy (PTE) is a consequence of traumatic brain injury (TBI) and can drastically decrease quality of life. Currently, there is no method available to predict which TBI patients will develop epilepsy. The present study aims to use a machine learning model that can accurately predict the risk of developing PTE from white-matter alterations following trauma. We used diffusion weighted imaging of 39 patients from the Epilepsy Bioinformatics Study for Antiepileptogenic Therapy to analyze fractional anisotropy from a tractography-based analysis. Next, we utilized a Random Forest model to classify seizure outcomes in TBI patients. Our model, assessed with 100 rounds of cross-validation, classified seizure outcome with 61% accuracy. The discrimination between seizure-free and seizure-affected subjects suggests that the classifier could improve characterization and diagnosis of PTE. These results may be instrumental in predicting PTE risk and may be implemented in future research of antiepileptic therapies.