Classification Of Depression Magnetic Resonance Imaging Based On High--Order Network

Using resting-state functional magnetic resonance imaging technology to study the functional connectivity of the brain is one of the most important ways to study current brain diseases.This method can accurately diagnose a variety of brain disorders including depression.However,traditional functional connectivity networks have consistently ignored the dynamic changes in functional connectivity between brain regions.At the same time,the existing researches and methods also lack the research on the correlation degree of functional connectivity between brain regions,ignoring the important information which may be included in the disease diagnosis.Previous studies have proposed a method of constructing a high-order functional connectivity network,taking the time-varying characteristics into consideration in constructing an inherent functional connectivity network and using a clustering approach to reduce computational complexity.However,the randomness of the initial clustering centers and the number of clustering clusters have a great impact on the classification results.At the same time,the constructed networks lack the neurological interpretability.In order to solve these problems,this paper presents a method to construct a minimum spanning tree high-order connectivity network.The main innovations of this paper are as follows:First,aiming at the problem that Chen's high-order functional connectivity network can not effectively explain the physiology,this paper improves the construction method of high-order functional connectivity network.Second,a method to construct a minimum spanning tree high-order functional connectivity network is proposed.Through the minimum spanning tree,this unbiased method simplifies the structure of high-order functional connectivity networks,preserves the network core framework,At the same time,it ensures the neurological interpretability of the network,studies the deeper interaction information in the network,and ensures the accuracy of the classification.Thirdly,a Relief feature selection method based on two-to-two redundancy analysis is proposed.The redundant feature in the feature set obtained by the Relief feature selection algorithm is removed and the optimal feature subset is obtained.Fourth,this paper proposes a multi-parameter optimization framework,through which the optimal combination of parameters can be obtained,and over-fitting is prevented to improve the generalization performance of the classifier,so that the feature selection and classification results are more accurate and effective.In this paper,a method to construct a minimum spanning tree high-order functional connectivity network is proposed.By using the minimum spanning tree,which is an unbiased method,the structure of a high-order network is simplified and the network core framework is preserved.Without losing the dynamic characteristics of the time series,the network is guaranteed to be neurologically interpretable,the deeper mutual information in the network is studied,and the accuracy of the classification is guaranteed.At the same time,this paper proposes a multi-parameter optimization framework,select the important features from the minimum spanning tree high-order functional connectivity network and classify.Classification results show that the classification of resting-state functional magnetic resonance imaging based on minimum spanning tree and high-order connectivity network greatly improves the accuracy of the diagnosis of depression.