Machine Learning Classifier Using Abnormal Resting State Functional Brain Network Topological Metrics In Major Depressive Disorder

The human brain is one of the most complex system in the world. Its complexity is not only reflected in the number of neurons and connections, but also in how the brain is wired on different scales and how such patterns of their connections produce cognitive functions, thoughts, feelings, and behaviors. Recently, a number of researchers have applied complex network methods in cognitive neuroscience, especially for the study of psychiatric disorders. Using complex network principles and statistical physics methods, researchers can analyze and discover basic network properties and the potential topological relationships between nodes. Complex network theory provides a new perspective and method for brain research.The current study explores the resting state functional brain network, compares methods based on complex network theory, and develops a software platform. On the basis of these observations, different network metrics, including global, local, and modular properties, are compared to explore the differences between groups. We then confirm the clinical availability of major depressive disorder (MDD) as a disease model. The aim is to determine the imageology symbol for the clinical early diagnosis and treatment of MDD at the network level. Finally, referring to the network metrics, the current study constructs an aided diagnosis model of MDD using a machine learning algorithm that could be helpful in clinical applications.The current study includes the following research outcomes:(1) Construction, analysis and classification of a resting state functional brain network in MDDGlobal and local metrics are calculated using graph theory-based approaches. Non-parametric permutation tests are then used for group comparisons of topological metrics, which are used as classified features in different algorithms. A sensitivity analysis is used to calculate the change in variance of each feature in the target category.(2) Research of the different community structures of resting state functional brain networks in MDDA greedy algorithm is used to divide the community structures. In terms of the modularity of the brain network, differences in the modularity metrics of normal control subjects and MDD patients are found, including the modular components, modular roles, and connections between modules. These differences are used as classification features in the machine learning method, the results of which exhibit a highest accuracy of90.50%.(3) Gene effects for the resting state functional brain network in MDDConvergent evidence from multimodal imaging studies has demonstrated that brain networks are both structurally and functionally heritable. A general linear2×2analysis of variance test is performed to investigate the main effects of genotype and prevalence, as well as their interaction. The metrics with significant differences are selected as features in the classification research.(4) Feature selection and classifier using regional homogeneity method in MDDThe current study explores the regional homogeneity of brain regions in the resting state to test the abnormality hypothesis in MDD patients. Classification and sensitivity analysis are also used to determine changes in the variance of each feature in the target category.The current study is the main component of the National Natural Science Foundation Project "The study of fMRI data analysis methods and diagnosis treatment models in major depressive disorder (No.61170136)", and is also supported by the University Science Research and Development Project of Shanxi Province (No.20121003) and the Special/Youth Foundation of Taiyuan University of Technology (No.2012L014).In brief, the current study focuses on resting state funcaional brain network construction and analysis techniques, the development of a software platform, and topological network changes in the condition of brain disease. On this basis, the current study explores the imageology symbols of early diagnosis and prognostic evaluation of major brain diseases, and constructs a diagnostic model. This is not only at the frontier of international scientific issues, but is also a major national requirement.