Reduction in PCL-5 Scores With rTMS

Background: Repetitive transcranial magnetic stimulation (rTMS) is widely used for patients with treatment-resistant depression; however, therapeutic outcomes differ substantially across individuals, and clinically practical predictors of response remain limited.Resting-state electroencephalography (EEG), analyzed with machine learning techniques, may offer an objective and scalable biomarker to support individualized treatment decisions.Methods: Pre-treatment resting-state EEG was collected from 100 patients with major depressive disorder (MDD), including 51 responders and 49 non-responders to rTMS.EEG recordings were acquired using either an 8-channel or a 30-channel montage.Spectral and connectivity features-band power, coherence, phaselag index (PLI), and phase-locking value (PLV)-were extracted.Predictive models were developed using a two-stage feature selection strategy incorporating recursive feature elimination with a support vector classifier and sequential backward selection, followed by linear discriminant analysis (LDA).Model performance was evaluated using 5-fold cross-validation with 1,000 permutation tests.Results: Models integrating multiple EEG features demonstrated superior predictive performance compared with single-feature approaches.The highest accuracy was achieved using the 8-electrode configuration, yielding 78.9% classification accuracy and an area under the curve of 73.3%, outperforming the higher-density montage.Among individual features, PLI showed the strongest predictive value, while feature integration further enhanced model robustness.LDA provided stable generalization in the context of limited sample size.Discussion: Resting-state EEG combined with machine learning can effectively identify patients likely to respond to rTMS in MDD.Importantly, reliable prediction was achieved using a low-density EEG montage, underscoring the feasibility of cost-efficient and clinically deployable EEG-based tools for personalized neuromodulation treatment.