Research On Brain Network Analysis And Auxiliary Diagnostic Framework Of Major Depressive Disorder Based On Functional Magnetic Resonance Imaging

The cross-combination of magnetic resonance imaging(MRI)technology with complex network,graph theory,machine learning,and statistical analysis methods has brought good news to study brain network.Modern network science reveals that the human brain connectome is organized with a small-world network,free-scale mode,hierarchical modular structure,and central backbone of highly connected hubs,like rich club and diverse club.The development of these theories has advanced the research of abnormal mechanisms of brain diseases.The cognition of high order brain function is constantly improving,from the traditional point-to-point correlation to multi-region interaction,formed as hyper-network.Meanwhile,the construction perspective of the brain network is transformed from static to dynamic interaction.The topological properties of the hyper-network have achieved good results in the auxiliary diagnosis of brain diseases,but the performance of the hyper-edge weight attribute has not been explored.In addition,the dynamic hyper-network is a very attractive perspective.Extraction of absolute spatiotemporal variability based on dynamic hyper-network would develop new directions for the research of brain diseases.Depression is the leading cause of mental health-related disease burden in the world,which prevents people from reaching their full potential,and is associated with premature mortality from suicide and complication.Due to the heterogeneity of depression and the lack of clinical biomarkers,it is important to explore the pathological mechanism of depression and improve the quality of clinical auxiliary diagnosis.The main work of this research consists of three parts:Firstly,we constructed functional brain networks for patients with major depressive disorder(MDD)and healthy controls(HC),and originally explored the abnormality of the diverse club and rich club from three aspects,including the hierarchical connection mode(club connectivity,feeder connectivity and local connectivity),topological properties of each club members,club anti-aggression and stability.The results showed that rich club connectivity strength among superior frontal gyrus,middle temporal gyrus,fusiform gyrus,postcentral gyrus,and lateral occipital cortex significantly decreased in MDD patients,some of which also showed significant correlation with HAMD score in MDD patients.Meanwhile,regions with abnormal topological properties of diverse club and rich club were mainly located in the frontalsubcortical circuit,frontoparietal-limbic circuit and visual-limbic system.They are important emotional-attention circuity of the brain network that are highly related to emotional regulation.In the perspective of central club,our findings correspond to the current understanding of depression as a network-based disorder.Secondly,we constructed hyper-networks for MDD patients and HCs using sparse regression method.The conventional topology metrics and our proposed hyper-edge weight property of hyper-network were extracted as features to improve the autodiagnosis accuracy of MDD.We not only compared the classification performance of each hyper-network coefficients but also designed a multi-feature ensemble model to fuse all kinds of features.The hyper-edge weight combining the topology properties achieved 89.24% accuracy and area under the curve of 0.9571.The multi-feature ensemble model combining different hyper-network coefficients provides new insights into the automatic diagnosis with diverse information of MDD.Thirdly,we originally constructed functional dynamic hyper-network(DHN)for MDD patients and HCs.The improved Euclidean distance-based absolute spatiotemporal variability were calculated for each brain region to study the spatial and temporal abnormality of DHN in MDD patients.Compared with HCs,sub-regions in default mode network(DMN)showed significantly decreased temporal and spatial variability in MDD patients,which may be associated with depressive rumination and abnormal adversity in MDD patients.Meanwhile,the sub-regions located in the frontoparietal network showed significantly decreased spatial variability,which may be related to abnormal emotional regulation in MDD patients.It is worth noting that the sub-regions of the visual network in MDD patients showed significantly increased spatial variability.The spatial variability of c Cun G＿L was significantly positively correlated with the Hamilton Anxiety Scale(HAMA)score in MDD patients.Therefore,c Cun G＿L’s spatial variability has the potential to be an objective biomarker for clinical evaluation of MDD patients’ anxiety severity.It can also assist doctors to understand the patients’ condition,and then develop a personalized treatment plan to improve the effects of treatment.