Research And Application Of Brain Network Construction And Feature Learning Algorithms Based On Structural Information

Analytical diagnostic methods based on brain networks are widely used in the diagnosis of brain diseases such as schizophrenia.There are three core steps in brain network analysis methods,i.e.,the construction of brain network,feature learning and classification diagnosis.Among them,the construction of the brain network is the basic step.Advanced brain network construction methods can reflect important structural and functional characteristics of the brain,thereby further providing a data basis for feature learning and classification diagnosis.Feature learning of the brain network is a key step in the intelligent diagnosis and analysis of brain diseases,which can contribute to improve the performance of classification diagnosis.This paper is based on some researches in the field of machine learning,and proposes to combines the structural information issues involved in the process of brain network construction and feature learning.In summarize,the main work of the paper is as follows:In terms of brain network construction,most existing methods tend to consider only first-order structural information of paired brain regions while ignoring useful higher-order structural information.Some of the early brain disease patients have subtle changes in brain function networks,which cannot be detected in conventional brain network construction methods.It is well known that the high-order method is usually more sensitive to the subtle changes in signal than the low-order method.For exploiting the high-order structure information among brain regions,we define the triplet correlation among three brain regions,and derive the second-order brain network based on the connectivity difference and ordinal information in each triplet.For making full use of the complementary information in different brain networks,we proposed a hybrid approach to fuse the first-order and second-order brain networks.The proposed method is applied to identifying the biomarkers of schizophrenia.The experimental results on six schizophrenia datasets show that the proposed method outperforms the existing brain network methods in the diagnosis of schizophrenia.In terms of brain network feature learning,this paper proposes a multi-view low-rank learning framework that can learn first-and second-order brain network structure information together and apply it to the diagnosis of schizophrenia.The proposed method not only embeds the first-and second-order structural information of the brain network into the learning model,but also enhances the cooperation between the first-and second-order brain networks by combining the ideal representation term.The experimental results of the three schizophrenia datasets(totally including 168 patients and 163 normal controls)show that our proposed method achieves promising classification results in the diagnosis of schizophrenia.In terms of real application scenarios,currently,incomplete multi-modality brain imaging data learning is still a challenge.In clinical applications,it is difficult to obtain all the modalities of each sample due to cost or equipment limitation.Therefore,processing incomplete multi-modality data has wide application prospects.The traditional methods usually discard incomplete subjects from the data.However,in most of the brain imaging analysis problems,the number of subjects is usually very limited,so discarding incomplete samples usually leads to inefficient training and poor robustness.Therefore,this paper proposes an incomplete multi-modality data learning method based on low-rank representations to learn the low-rank representation of the complete modality part,and then recover the incomplete modality by the consistency between the modalities.Extensive evaluation of the proposed method on 89 temporal lobe epilepsy(TLE)patients,103 frontal lobe epilepsy(FLE)patients,and 114 healthy individuals with incomplete DTI and f MRI data showed that our method can achieve promising classification results in identifying patients with TLE or FLE.