Depression analysis using visual and textual cues

Depression is the major cause of mental health illness around the world and lack of accurate depression assessment methods makes this one of the serious diseases. In this paper, an AI-based architecture is proposed that can act as a robust method for premature detection of depression. The user is made to attend the Physical Health Questionnaire of depression scale and the video of the people’s face is recorded and used for diagnosis. The raw videos are used to extract various facial features like Pose, Gaze, Facial Action Units and used to detect the level of depression using a stacked stateful LSTM neural network architecture. The dataset is trained using custom training steps to extract the features over every frame of the whole video and classify based on that. A text classification module is designed that uses bidirectional LSTM architecture to classify the user utterance which can be used to diagnose the person prior to the video diagnosis method using a talking interview or using a bot.