A Deep Learning-Based Framework for Detecting Depression from Electroencephalogram Signals

Depression is a serious mental disorder that affects millions of people around the world. One of the challenges in dealing with depression is to detect it early and accurately. In this study, we use electroencephalography (EEG) to measure the brain activity of depressed and non-depressed individuals. We aim to find out which EEG features can serve as reliable biomarkers for depression recognition. We processed the raw EEG data from 128 electrodes by selecting 16 relevant electrodes, removing outliers and wrong formatted data. Then we applied a Gramian angular field (GAF) transformation to the relevant electrodes. We fed the resulting matrix into a recurrent neural network (RNN) model with two hidden layers using ReLU activation function and a sigmoid activation function as output layer, which produces binary results. We used this model to classify 53 patients with different brain disorders based on their EEG features (16 electrodes * 3 features). The comparison of the classification accuracy of our proposed work with other dataset is found to be better. It is observed that the overall depression classification performance is found to be very promising as compared to previous research works being conducted on both the datasets with an accuracy of 90.57%. In this paper, we have presented a novel approach to identify depression from EEG signals. We applied RNN to these features to classify depressed and non-depressed subjects. Our approach is based on empirical experiments to find the best parameters for our model. This may require some trial and error to achieve the optimal results, which could be seen as a potential drawback of our method. We plan to address this issue by using an optimization algorithm to automate the hyperparameter selection process in our future work.