Functional Connectivity Metrics in Temporal Lobe Epilepsy: A Machine Learning Perspective With MEG

Abstract

Temporal Lobe Epilepsy (TLE) is a prevalent neurological disorder affecting millions worldwide, including a significant proportion in India. Precise diagnosis and effective treatment planning are critical for TLE patients, necessitating advanced neuroimaging techniques. Magnetoencephalography (MEG) offers a non-invasive method for evaluating brain function, providing detailed insights into TLE. In this study, we aim to evaluate the potential of functional connectivity metrics derived from MEG data at the source level for distinguishing TLE patients from healthy controls (HCs). We analyse the data across various brain frequency bands, including alpha, beta, delta, gamma, theta, broadband, and high-frequency oscillations (HFO), using amplitude envelope correlation and graph theory metrics. We employ machine learning algorithms to classify TLE and HC groups based on these metrics. Chi2 feature importance analysis reveals significant importance of connectivity metrics such as local efficiency, mean clustering coefficient, mean shortest path length, small worldness score, weighted degree centrality, binary degree centrality, global efficiency across frequency bands, particularly in theta, alpha, beta, broadband and HFO bands. Various machine learning models demonstrate high classification performance, with accuracies reaching up to 100% in particular frequency bands in agreement with the Chi2 feature importance analysis. Overall, the Subspace Discriminant Ensemble model, especially in the Theta and Alpha frequency bands, show exceptional potential for classifying TLE and HCs. Overall, this study underscores the potential of MEG and functional connectivity analysis using specific frequency bands and machine learning models for classifying TLE and HC with high accuracy, which may contribute to improved diagnosis and management of epilepsy.