Hyper-network Functional Magnetic Resonance Imaging Classification Method Based On Multiple Features Fusion In Alzheimer’s Disease

Exploring functional interactions among different brain regions have contribute to understanding the pathological underpinnings of neurological disorders.Brain networks,as an important representation of those functional interactions,and thus are widely applied in the diagnosis and classification of neurodegenerative diseases.However,a conventional functional connectivity network is generally constructed based on the pairwise correlations among various brain regions,which ignores their higher-order relationships.Exist neuroscience studies have demonstrated that high-order relationships are crucial to brain network analysis.Hyper-network method has been proposed and it can effectively describe the interactions among multiple brain regions.However,this method extracts the local properties of brain regions as features,which ignore the global network topology information.This limitation can affect the evaluation of network topology and even reduce the performance of the classifier.The limitation can be solved by adopting subgraph feature-based method,but this method is not sensitive to change in a single brain region.Considering that above these feature extraction methods result in the loss of information,we propose a novel hyper-network functional magnetic resonance classification method based on multiple features fusion in Alzheimer’s disease.The proposed method combines the brain region features and subgraph features,and then adopts a multi-kernel SVM to classify.It can retain not only the global network topological information,but also the sensitivity to change in a single brain region.In order to certify this method,28 normal control subjects and 38 Alzheimer’s disease patients were selected to participate in the experiment.This method obtained satisfactory classification accuracy,with an average of 91.60%.The abnormal brain regions involved the bilateral precuneus,right parahippocampal gyrus\hippocampus,right posterior cingulate gyrus,and other crucial brain regions in Alzheimer’s disease.Finally,this paper retested the proposed method on the published ADNI data set to verify the stability of it.The experimental results demonstrated that,hyper-network functional magnetic resonance classification method based on multiple features fusion in Alzheimer’s disease can achieve better classification performance.The main innovations of this paper are as follows:First,based on the hyper-network model,subgraph features were adopted to classify.After constructing hyper-network,the hyper-edges were regarded as subgraph features,and the subgraph features were selected by the frequently scoring feature selection algorithm.Then the classifier based on the graph kernel was used to classify.The advantage of subgraph features to classify is that it can compensate for the loss of topology information of brain region features.The results also proved that the subgraph features are good at detecting the connection patterns among abnormal brain regions.Second,based on the hypernetwork model,the multi-feature method was adopted to classify.Two different types of features were combined and multikernel SVM was used for classification.The advantage of multi-feature is that it included two different types of advantages,which can guarantee the sensitivity of the individual brain region without losing the global topology property of the network.The experimental results also showed that multi-feature method is superior to single feature for classification in terms of sensitivity,specificity,accuracy,AUC value and Reflief weight.Thirdly,for Alzheimer’s disease,different feature extraction methods were used to classify.And then we compared the effects of different feature selection methods on classification results.This paper compared that the classification results under the three different feature extraction methods of brain regions features,subgraph features,multi features.The results proved the superiority of multi features to classify.In order to validate the stability and robustness this method,we tested on public ADNI data sets,and compared the common abnormal brain regions on Alzheimer’s disease under the two different data sets.