Graph-Based EEG Symmetry Features from the Temporal Lobe as Markers of Antidepressant Treatment Response

Abstract

Major Depressive Disorder (MDD) is a mental disorder that affects millions globally and has highly individualized responses to antidepressant treatment. Identifying objective early markers that can distinguish responders from non-responders remains a critical challenge in personalized psychiatry. In this study, we investigate the role of brain symmetry in EEG-derived functional connectivity as a potential early marker of treatment response. Using resting-state EEG recordings from 176 patients diagnosed with MDD at baseline (Visit 1) and after one week of treatment (Visit 2), we construct graph-based representations of functional connectivity and define region-wise and electrode-wise temporal-lobe symmetry features. We focus on symmetry change scores (Visit 2 - Visit 1) across typical EEG frequency bands in both weighted and binary forms. Statistical analyses reveal a consistent group-level pattern in the and {beta} bands: responders show negative symmetry change scores, whereas non-responders show relative stability or weakly positive change scores. The contrast is strongest in the {beta}1 band and is more pronounced for binary symmetry metrics, while the band shows the same direction with slightly weaker significance. Physiologically, these opposite trajectories are consistent with a treatment-related shift toward more lateralized temporal-lobe network organization in responders and a more bilateral coupling pattern in nonresponders; however, these neurophysiological interpretations remain hypothesis-generating given the resting-state design and the absence of direct behavioral or task-based validation. To quantify whether temporal symmetry change scores contain reproducible discriminative signal under strict validation, we benchmark the proposed features in a supervised classification pipeline using repeated nested cross-validation. The best-performing configuration achieved a median out-of-sample AUC of 0.690 with a 95% bootstrap confidence interval of [0.611, 0.711], indicating modest but reproducible separability. Overall, temporal-lobe symmetry change features provide interpretable candidate markers of early antidepressant-related neurophysiological change that may complement existing EEG predictors in future multi-marker models, but they are not yet sufficient for standalone clinical decision support and require external validation in independent cohorts.