Research On The Classification Model Of Alzheimer's Disease Based On Deep Learning

The diagnosis and pathogenesis of Alzheimer's disease(AD)has been a research hotspot in the field of biomedical diseases.Accurate classification of AD and finding out the key parameters that have an important impact on AD are the prerequisites for finding a treatment for AD.At present,when domestic and foreign researchers and doctors perform AD analysis,they usually only use the Mini Mental State Examination(MMSE)to score subjects,analyze their disease severity,or perform diagnosis of AD through only use Magnetic Resonance Image(MRI).For MMSE,there are differences in the scoring standards for different ethnic groups,and there is no guarantee that the same standard applies to everyone.For MRI,information can only be obtained through images,and other related information cannot be obtained at the same time.This study proposes to add MRS,age,and gender data based on TI-weighted MRI image information of the brain area,use deep learning techniques to model and classify,and explore the impact of the addition of Magnetic Resonance Spectroscopy(MRS)data on the classification of AD.The T-TEST and the “Neural Network Node Stimulation Method” proposed by the innovation screen the key data for modeling and improve the classification accuracy of the model for AD.Correlation analysis was performed between the data screened by the two methods and MMSE.And the relationship between the screened data and AD was explored to provide a reference for the study of key brain structure and MRS data for AD.The main research contents of this paper include:(1)A method of AD classification modeling by combining the MRI multi-modal information and MRS information of the brain area is proposed.The multi-atlas brain segmentation technology the MRI image are used for multiple Structural voxels.And MEGA-PRESS technology of MRS are used for extraction of metabolite concentrations.Using support vector machine for modeling,the experimental results show that the addition of MRS data improves the model's AUC value from 0.7607 to0.8,which effectively improves the performance of the model,and the classification accuracy is 80%.(2)Design a deep neural network based on stacked auto-encoders for modeling.Experimental results show that the addition of MRS data effectively improves the classification accuracy of the model for AD.The predicted value and label value of the model are reduced from the previous 0.0580 to 0.0111,and the AUC value is0.9750,which has a high classification effect.(3)Innovatively proposed "Neural Network Node Stimulation Method" to conduct dimensionality reduction analysis on the model's MRI,MRS,age and gender data,screen out the data that has a greater correlation with AD,and perform modeling to further improve the classification accuracy of the model.(4)Correlation analysis was performed between the data selected by the tow methods(T-TEST and "Neural Network Node Stimulation Method")and MMSE.That is,linear regression analysis,to obtain the correlation coefficient R value.After analysis and comparison,it was proved that the structure of the brain regions including the hippocampus and amygdala,as well as the spectral data ?-aminobutyric acid(GABA?5)located in the parietal lobe are true data which is large correlation with AD.