Research On Classification Of Multi-feature Fusion Based On Brain Function Hyper-network Model

The brain function Hyper-network model has been widely used in the application research of brain disease diagnosis.Because the Hyper-network model can represent the high-level relationships of multiple brain regions,this method has shown strong vitality in the field of brain disease diagnosis.However,most of the existing researchers mainly focus on studying the pros and cons of different brain function Hyper-network construction methods.Both use a single attribute(such as clustering coefficient)or simple linear splicing to characterize the topology of the brain network.However,different topological attributes represent different topological relationships in the Hyper-network.Although the clustering coefficients also show good classification performance in the brain disease diagnosis system,the role of other topological attributes in the Hyper-network may be ignored.Moreover,there is no gold standard in the existing research that stipulates which indicators are more conducive to quantifying the topology of the brain function Hyper-network,and it is impossible to distinguish the pros and cons of the classification performance of different topology indicators.At the same time,the information contained in the feature set composed of a single topological attribute or simple linear splicing may also be too one-sided,unable to fully characterize the topological structure of the Hyper-network,and ultimately affect the accuracy of the diagnostic model.In addition,traditional classifiers may not be able to perform well in the fusion feature set during the classification process.Therefore,this paper proposes a brain disease diagnosis model based on multi-feature fusion.The model can fuse information of different topological attributes and significantly improve the accuracy of the model.It provides a better choice for future researchers in the field of brain function Hyper-network.The main innovations of this paper are as follows:First,introduce a variety of different Hyper-network topology indicators for classification research,and design detailed experiments to compare the pros and cons of the classification performance of different attribute features,mining and analyzing the potential differences in the classification process using different features,and filter out It is more conducive to the topological properties of brain disease classification.The experimental results show that among the 11 common Hyper-network topological attributes,there are 3 single-node clustering coefficients,2 double-node clustering coefficients,and 6 features of the average shortest path,which clearly show better classification performance,which is relevant in this field.When researchers choose topological attributes,they provide a more comprehensive reference basis.Second,construct a fusion feature set that contains more complete information and is more suitable for brain disease classification.In order to make up for the shortcomings that the feature set contains information that may be too one-sided in the past research,this experiment uses the most commonly used linear splicing method in this field to compare the fusion feature set constructed by the 6 excellent features selected in the experiment.And through the design of multiple sets of control experiments,it is proved that the classification accuracy,Relif-F weight,correlation,redundancy,and comparison of previous methods of the fusion feature set under P<0.05 and the same number of features are all better than a single feature.It is superior to other fusion feature sets.Third,establish a complete multi-feature fusion classification model.Traditional classifiers may not be able to perform well in the fusion of features in the classification process,so it is necessary to further introduce advanced fusion methods.Although the weighted linear kernel in multi-core learning is widely used in the field of brain diseases,this method cannot well retain the local information in the original feature set.Therefore,on the basis of the constructed fusion feature set,this paper further combines the multi-core extension method in multi-core learning to construct a complete multi-feature fusion classification model.On the one hand,the classification model uses six excellent features to form a fusion feature set,which contains more comprehensive information.On the other hand,it also uses a feature fusion method based on multi-core expansion to further improve the classification performance of the model.And the superiority of the classification performance of this classification model is proved in many aspects such as P<0.05 and Relif-F.This paper also carried out a repeatability verification on the ADNI public data set,which proved the repeatability of the performance of the fusion feature set and the multi-feature fusion classification model in this paper.In summary,on the one hand,this paper analyzes the classification performance of 11 common topological attributes through comparative experiments,which provides a reference for the extraction of topological attributes in disease diagnosis.On the other hand,a fusion feature set containing more comprehensive information is also proposed,and a multi-feature fusion classification model with better classification performance is further proposed.It can provide a better choice for the future development of the brain function Hyper-network diagnostic system,avoiding the problem of missing information that may be caused by the use of a single feature and the blind use of feature splicing.