Design on Modeling of Multimodal Depression Aided Diagnosis from Psychological Perspective

The pervasive hopelessness and high risk of suicide in depressive patients suggested that the self might be abnormal among them. Current state-of-the-art depression diagnostic methods include on the basis of an interview style assessment between a clinician and a patient, patient self-reporting, and typical rating scales. However, these tests are subjective and single in nature, low diagnostic accuracy and lack an objective predictor of depression. Therefore, in order to enhance current diagnostic methods, an objective screening mechanism based on physiological and behavioral signals is needed. Learning from previous relevant research and preliminary experimental result, a range of acoustic features have already been identified for applying to the classification of depression. The current project will systematically address above issue, by focusing on abnormalities of self-related processing in patients with unipolar depressive disorder and healthy controls. On the basis of multimodal data acquisition, we will utilize accurate and sophisticated ways of feature extraction and fusion to develop a depression classification model. We present a method of combining the deep learning algorithm with classical psychological experiment technology, which improves the accuracy recognition and provides an objective basis for clinical diagnosis. The result will not only provide new evidence and support for studies on abnormalities of self-related processing in depression via multimodal analysis, but also lay a solid foundation for enriching the diagnosis of depression.