A deep learning approach to evaluating sex differences in antidepressant response to neuromodulation using EEG in major depressive disorder

      Sex differences in the antidepressant response to repetitive transcranial magnetic stimulation (rTMS) for major depressive disorder (MDD) are poorly understood. Electroencephalography (EEG) can serve as a biomarker for treatment response, and deep learning (DL) can classify EEG signals to predict response. In this study, five DL models were created to classify males and females as rTMS responders/nonresponders using EEG.
      Fifty MDD patients (M 25, F 25) were enrolled in an open-label rTMS study. RTMS (1 Hz) was applied to the right dorsolateral prefrontal cortex (DLPFC) for 10 sessions. EEG was recorded for 10 minutes before and after treatment.
      Twenty-eight (57.1%) patients exhibited treatment response. Significant decreases on the Beck Depression Index (BDI) were observed from pre- to post-treatment in males and females (both p<.01). Neither the difference between the decreases for males and females or the difference for remission were significant. Using pre-and post-rTMS data, the first two models classified females as responders/nonresponders with 94.52%, and males with 92.52%, accuracy. Using only pre-rTMS EEG, the third and fourth models classified females and males as responders/nonresponders with accuracies of 91.54% and 98.15%, classifying responses in pre-TMS data better in males than females. The fifth model classified all participants at pre-TMS as responders/nonresponders with 94.13% accuracy.
      This is the first study to apply DL to sex differences in EEG response to rTMS. We were able to predict TMS response with 94% accuracy from the baseline’s pre-TMS EEG data.
      DL models and EEG can be used to predict antidepressant rTMS response in males and females with high accuracy, and the predictive power of this approach is marginally dependent on sex.
      Keywords: rTMS, EEG, deep learning, sex differences