Identification of Functional Connectivity Features in Depression Subtypes Using a Data-Driven Approach

Abstract

Biomarkers are not well understood in depression, partly because there is no golden rule of what is abnormal in which patients and how neurobiological information can be used to improve diagnosis. The heterogeneity of depression suggests that diverse circuit-level abnormalities in individuals lead to various symptoms. Investigating heterogeneous depression is crucial to understand disease mechanisms and provide personalised medicine. Dynamical functional connectivity (dFC), consisting of spatial-temporal characteristics of brain activity, has been shown to be effective in characterizing the circuit-level abnormalities in depression. However, most of the current studies on dFC are based on one-step mapping while ignoring hierarchical spatial-temporal information, which may leverage the power of diagnosis in depression subtypes. In this study, we propose Brain Network Gated Recurrent Units (BrainNetGRU) to discover hierarchical resting-state dFC features for the diagnosis of depression subtypes, using data from 770 depressive adults from the UK Biobank. Particularly, we devise diffusion convolutional filters and recurrent units to effectively learn distinctive dynamic brain connectivity for depression subtypes. Experimental results show that BrainNetGRU can identify three types of depression with an accuracy of 72.05%. In addition, BrainNetGRU shows that resting-state functional connections in default mode network (DMN), cingulo-opercular network (CON) and fronto-parietal network (FPN) are important in the diagnosis of depression subtypes.