Classification of Epileptogenic networks in temporal lobe epilepsy patients in contrast to the healthy controls

Abstract

This study aims to investigate the classification of individuals with Left Temporal Lobe Epilepsy (LTLE) and Right Temporal Lobe Epilepsy (RTLE) in comparison to Healthy Controls (HC) based on machine learning approaches.  The dataset of patients and Healthy Cohorts of resting-state functional magnetic resonance imaging (rs-fMRI) is preprocessed using CONN software which works on MATLAB. Twelve Regions of Interest (ROIs) were selected in CONN.  Supervised learning algorithms, particularly the Random Forest Algorithm, were employed for categorizing the connection matrices of the 12 ROIs. The Random Forest Algorithm achieved the highest accuracy during five cross-validation folds, with 83% accuracy in classifying Right Healthy Controls (RHC)-RTLE and 72.10% in classifying Left Healthy Controls (LHC)-LTLE. Feature importance plots generated by the Random Forest Algorithm were utilized to identify critical relationships influencing the categorization, demonstrating distinct connection patterns between individuals with RTLE and RHC and LTLE and LHC, suggesting potential implications for understanding temporal lobe epilepsy.