Depression Diagnosis using Deep Convolutional Neural Networks

Depression is a prevalent psychiatric disorder that impacts the quality of life of 300 million people around the world. The complex nature of depression manifestations in patients and the lack of technological advances in the diagnosis process has left a lot of room for improvement in this particular domain. At present, the diagnosis is mainly made by physicians during a conversation comprising the exploration of the symptoms and the diagnostic criteria for depression. Recently, the electroencephalography (EEG) has regained interest as a promising approach to provide bio-markers which are of clinical value in the diagnostic process and for response prediction to therapy. In the present landscape, even the addition of EEG data has resulted in a semi-automated process, where the expert still has to heavily modify the raw data. This adds an inherent bias to the process based on the expert and incurs costs as well as time to the process of diagnosis. In this paper, we present a fast, effective and automated method that is able to quickly determine if the patient has depression while still maintaining a high accuracy of diagnosis. Our approach is built on using raw EEG-data, performing frequency domain preprocessing in order to split the data into its different frequency domains and to create EEG ’images’. These images are then treated by a convolutional neural network, which is a novel approach in this area. Experimental results have shown to provide outstanding results and to work without the need for feature engineering or any human interaction, which is a core strength of the model we are proposing