Prediction of Treatment Outcome to Transcranial Direct Current Stimulation in Major Depression Based on Deep Learning of EEG Data

Major Depressive Disorder (MDD) is a leading cause of disability worldwide. Current first line treatments are antidepressant medication and psychotherapy. However, they have limited effectiveness and there are no biomarkers that can predict treatment response at the individual level. Transcranial direct current stimulation (tDCS) is non-invasive brain stimulation method that is a potential novel treatment for MDD. The present study sought to investigate neural biomarkers for predicting response to tDCS at the individual level using portable EEG. The clinical trial was a double-blinded, placebo-controlled, randomized, superiority trial of home-based tDCS. Participants were randomized to a 10-week course of either active or sham tDCS sessions. Resting state, eyes closed EEG data were acquired at baseline, prior to starting tDCS, and at week 10. EEG data acquisition was conducted using a portable, 4-electrode EEG device (Muse). The baseline EEG data from 21 participants were used to train and test the deep learning models of 1D convolutional neural networks (1DCNNs), Long Short-Term Memory (LSTM), Gated recurrent units (GRU) and the hybrid models combining 1DCNN and LSTM/GRU. A prediction rule was proposed and applied to the classifier outputs of each participant and the treatment outcomes were predicted. Different combinations of power spectral density vectors extracted from the EEG frequency bands of four electrodes were selected to improve the treatment outcome prediction. Using 1DCNN model the work achieved a treatment outcome prediction accuracy 85.7%, with a specificity of 71.4% for predicting treatment remission and sensitivity of 92.8% for predicting residual depressive symptoms, which was based on the combined theta and alpha EEG band power spectral density from the TP10 electrode.