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

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.

[1]  Jiangtao Cui,et al.  Estimating Interactions of Functional Brain Connectivity by Hidden Markov Models , 2018, ADMA.

[2]  Xin Yao,et al.  MWMOTE--Majority Weighted Minority Oversampling Technique for Imbalanced Data Set Learning , 2014 .

[3]  D. Hu,et al.  Identifying major depression using whole-brain functional connectivity: a multivariate pattern analysis. , 2012, Brain : a journal of neurology.

[4]  Ben Glocker,et al.  Disease prediction using graph convolutional networks: Application to Autism Spectrum Disorder and Alzheimer's disease , 2018, Medical Image Anal..

[5]  Xiang Wang,et al.  Evidence of a Dissociation Pattern in Resting-State Default Mode Network Connectivity in First-Episode, Treatment-Naive Major Depression Patients , 2012, Biological Psychiatry.

[6]  Dinggang Shen,et al.  Deep Chronnectome Learning via Full Bidirectional Long Short-Term Memory Networks for MCI Diagnosis , 2018, MICCAI.

[7]  Ludovica Griffanti,et al.  Image processing and Quality Control for the first 10,000 brain imaging datasets from UK Biobank , 2017, NeuroImage.

[8]  Danielle S. Bassett,et al.  Modeling and interpreting mesoscale network dynamics , 2017, NeuroImage.

[9]  Andrew T. Drysdale,et al.  Resting-state connectivity biomarkers define neurophysiological subtypes of depression , 2016, Nature Medicine.

[10]  Daoqiang Zhang,et al.  Hybrid Functional Brain Network With First-Order and Second-Order Information for Computer-Aided Diagnosis of Schizophrenia , 2019, Front. Neurosci..

[11]  G. Deco,et al.  Dynamic functional connectivity reveals altered variability in functional connectivity among patients with major depressive disorder , 2016, Human brain mapping.

[12]  Yoshua Bengio,et al.  Empirical Evaluation of Gated Recurrent Neural Networks on Sequence Modeling , 2014, ArXiv.

[13]  Stuart J. Ritchie,et al.  Resting-State Connectivity and Its Association With Cognitive Performance, Educational Attainment, and Household Income in the UK Biobank , 2018, Biological psychiatry. Cognitive neuroscience and neuroimaging.

[14]  A. Young,et al.  High heterogeneity and low reliability in the diagnosis of major depression will impair the development of new drugs , 2015, BJPsych Open.