Research On Multi-modality Data Fusion For Classification Of Schizophrenia

With the development of cross-disciplines in various fields such as imaging genomics and machine learning methods,the use of different types of multi-modal data fusion analysis to assist in the diagnosis and prognosis of some complex diseases provides an important theoretical basis and reality value.Schizophrenia is a complex psychiatric disease with hereditary nature.Because its etiology is affected by different factors,the classification and learning task of schizophrenia has attracted more and more researchers’ attention.Combining the related algorithms of machine learning and the multi-modality data fusion analysis of omics data in biomedicine provides reliable ideas and methods for the classification and research of schizophrenia.However,most of the existing methods are to perform feature selection on the data on a single modality,without forming a systematic association between the data,and ignoring the inherent structural information between the modality and the modality.To this end,this paper proposes a multi-modality data fusion algorithm,which makes full use of the interrelationships in multi-modality data,and designs a data fusion method by analyzing the multiple relations of homogeneous networks and the interaction relations of heterogeneous networks to further improve the spirit of the accuracy rate of schizophrenia classification.Finally,the classification accuracy rate of the model proposed in this paper reached 83.21%.The main research contents and tasks are as follows:(1)The multi-modality data fusion method is used to integrate relevant information from multiple modality related data sets,which overcomes the basic limitations of a single modality.Traditional single-modality data analysis will lose important information of the original data,thereby affecting the subsequent feature extraction process.Based on the joint feature learning model,this paper designs a new multi-modality data fusion model to achieve cross-modality feature extraction,make full use of the structural relationship between multiple sets of data,and then improve the classification performance.(2)The similarity relationship between multiple modality data is represented by the establishment of a similarity matrix,and the internal information of the modality and the structural information between the modalities are fully utilized.A method based on a Gaussian radial basis function is used to construct a similarity matrix intra-modality and inter-modality,to integrate similarity networks of multiple modalities,to complement the missing network structures,and to strengthen the network structure shared by multiple data.The multi-modality joint learning model is designed by using the multiple relations of the homogeneous network and the interaction relations of the heterogeneous network existing in the multi-modality data network.(3)The simulation data set and the real data set are verified by the proposed method on the established model.The results show that in the classification of schizophrenia,the multi-modality data fusion algorithm proposed in this paper is superior to other competitive methods.Improve the classification accuracy of schizophrenia.At the same time,this paper also uses the model to detect schizophrenia-related markers,revealing the significant interactions between risk genes closely related to schizophrenia,environmental factors,and abnormal brain regions,providing for mechanism research and clinical diagnosis of mental illness important theoretical support.