Automatic Identification of Depression Using Facial Images with Deep Convolutional Neural Network

Background: Depression is a common disease worldwide, with about 280 million people having depression. The unique facial features of depression provide a basis for automatic recognition of depression with deep convolutional neural networks. Material/Methods: In this study, we developed a depression recognition method based on facial images and a deep convolutional neural network. Based on 2-dimensional images, this method quantified the binary classification problem and distinguished patients with depression from healthy participants. Network training consisted of 2 steps: (1) 1020 pictures of depressed patients and 1100 pictures of healthy participants were used and divided into a training set, test set, and validation set at the ratio of 7: 2: 1; and (2) fully connected convolutional neural network (FCN), visual geometry group 11 (VGG11), visual geometry group 19 (VGG19), deep residual network 50 (ResNet50), and Inception version 3 convolutional neural network models were trained. Results: The FCN model achieved an accuracy of 98.23% and a precision of 98.11%. The Vgg11 model achieved an accuracy of 94.40% and a precision of 96.15%. The Vgg19 model achieved an accuracy of 97.35% and a precision of 98.13%. The ResNet50 model achieved an accuracy of 94.99% and a precision of 98.03%. The Inception version 3 model achieved an accuracy of 97.10% and a precision of 96.20%. Conclusions: The results show that deep convolution neural networks can support the rapid, accurate, and automatic identification of depression.