Uncovering Neural Signatures of Convulsive Therapy in Depression Using Massive EEG Time-Series Feature Extraction

BACKGROUNDConvulsive therapies, including electroconvulsive therapy (ECT) and magnetic seizure therapy (MST), are highly effective for treatment-resistant depression, however, their neural mechanisms remain incompletely understood. We tested whether a data-driven framework applying a massive time-series feature library to electroencephalography (EEG) could reveal novel insights into changes in functional brain dynamics following convulsive therapy and provide preliminary markers of response. METHODSResting-state EEG was analysed before and after a course of ECT or MST in 42 patients. Data were pooled and reduced to three principal components (PCs) capturing 78.1% of total variance, then >7,000 time-series features per PC were extracted from each patient using the highly comparative time-series analysis (hctsa) framework. Linear support vector machines (SVMs) classified pre-versus post-treatment EEG for each PC. A separate linear SVM, was also trained on 18 representative baseline features to predict clinical response. RESULTShctsa-based classifiers distinguished pre-from post-treatment EEG for each PC (accuracy 73.8-76.2%, pFDR<0.01). Between 986-1,414 features significantly differentiated post-stimulation from baseline time-series across the three PCs (pFDR<0.05). The most discriminative features indexed linear and non-linear autocorrelation, correlation, multiscale entropy, and spectral properties. The baseline SVM combining 18 features showed modest but statistically reliable prediction of treatment response (balanced accuracy=0.69, area under the curve [AUC]=0.61, p=0.014). Single-feature ROC analyses further identified several top features with AUCs=[~]0.7. CONCLUSIONSData-driven analysis using a diverse time-series feature library can uncover novel EEG signatures of brain changes following convulsive therapy. This holds potential for delineating treatment-related mechanisms and developing predictive biomarkers to support precision psychiatry.