Approved research

Application of machine learning for depression classification and outcome prediction

Lay summary

Currently, the diagnosis of depression is mainly based on clinical presentations and there is a lack of reliable markers to differentiate patients with depression from those without depression. Furthermore, the neurobiological mechanism of depression has been rarely understood. The machine learning approach can train a classifier for early diagnosis and outcome prediction, and propose a network hypothesis for disease. In this project, we applied the machine learning algorithms into depression for this purpose. First, we use these algorithms to differentiate patients with depression from those without depression for early diagnosis; Second, we used the MRI data at baseline to predict the outcome endpoints of the depression in the longitudinal data at follow-up; Third, we will identify whether patients with depression have hippocampal volume decrease because this area is very important for cognitive function. Third, the Mendelian randomization analysis will be used to investigate the possible causal relationship among the genes, hippocampal volume, brain functional alterations, and various outcomes (depression score, anxiety score, sleep parameters). These findings could help us early diagnosis and outcome prediction for the depression, and tell us whether depression subjects have hippocampal atrophy, and their relationships with genes, sleep, depression symptom or other variables. Furthermore, these finding may broaden and expand our understanding of the neurobiological mechanism of depression. In view of huge works on data cleansing and data analyses, this project will last for 36 months.