6533b826fe1ef96bd1283ade

RESEARCH PRODUCT

Deep Learning Techniques for Depression Assessment

subject

description

Depression is a typical mood disorder, which affects a significant number of individuals worldwide at an increasing rate. Objective measures for early detection of signs related to depression could be beneficial for clinicians with regards to a decision support system. In this paper, assessment of depression is done by applying three deep learning techniques of Convolutional Neural Network (CNN). These techniques are transfer learning using AlexNet, fine-tuning using AlexNet and building an end to end CNN. The inputs of the CNNs are a combination of Motion History Image, Landmark Motion History Image and Gabor Motion History Image, and have been generated on a depression dataset. Accuracy of the three deep learning techniques are computed. As of now, transfer learning technique achieved a result comparable to the state of the art, of 83 % accuracy.