Early-Stage Non-Severe Depression Detection Using a Novel Convolutional Neural Network Approach Based on Resting-State EEG Data

Over 300 million people worldwide are affected by depression, with symptoms that have a major impact on patients and, in the worst cases, can lead to suicide. As the severity of the disease increases over time, early detection can save a patient's life. The disease is diagnosed by professionals using questionnaires that might be influenced by biases, and of which the accuracy and reliability are not guaranteed. For this reason, an increasing number of studies are looking at physiological ways of detecting the disease, with electroencephalogram-based machine learning prediction models having been successful in recent years. However, the focus is not on the early detection of mild depression, which can be the entry point to major depression. In this work, we developed a deep learning based model using a 1D convolutional neural network to detect mild depression in resting-state electroencephalogram data. We evaluated the model using a realistic world-like dataset and were able to achieve a balanced accuracy of 69.21%. With this result, we are setting a new benchmark for resting-state-based early detection. Due to the low level of preprocessing and the associated fast computing time and low computational intensity, our innovative approach can serve as a basis for applications in the real world. This enables patients with suitable hardware to recognize the disease themselves at an early stage and thus receive timely treatment to prevent further development.