Diagnosis Of Alzheimer’s Disease Based On Structural Magnetic Resonance Imaging And Neural Network

Alzheimer’s disease(AD)is an irreversible,progressive neurodegenerative disease that destroys brain and memory function.Over time,Alzheimer’s disease progresses from brain atrophy to dementia.There is no cure for Alzheimer’s disease,so early detection and timely treatment of Alzheimer’s disease is needed.The sMRI has been widely used in the diagnosis of brain diseases.The main problem of previous studies is that the single characterization method lacks effective neuroimage markers,which leads to the bottleneck of diagnostic performance.This thesis studies the diagnostic methods of Alzheimer’s disease based on structural magnetic resonance imaging and deep neural network,mainly including two aspects:1.The Convolutional-Squeeze-Excitation-Softmax-NET(CSES-NET)deep neural network combined with multi-channel feature fusion for the early diagnosis of Alzheimer’s disease is proposed,which can extract multi-scale characteristic information in structural magnetic resonance imaging to enhance the diagnostic performance.Specifically,deep features based on voxel morphology,cortical features based on surface morphology and radiomics features based on original image,wavelet image and Gaussian filter image are extracted from structural Magnetic Resonance Imaging(sMRI).The deep features are extracted from the patch image by residual network CSES-NET,which rescales features in the residual structure and fits the correlation between channels.Next,the features of the three channels are fused and classified by a fully connected neural network.In addition,imaging is used to assess genetic variation,and genomewide association study is performed for radiomics features and cortical indicators.In association analysis,SNP loci and risk genes with significant phenotypic traits of the disease are identified.This method extracts the rich features of multi-channel,improves the classification performance of the model,and has a good classification effect in the experiment of all subjects’ diagnosis and classification.The genome-wide association study revealed several SNP loci significantly associated with traits,which were located on some gene segments such as MTRR,MME,FRMD4 A,etc.,and the functions of these genes were factors causing brain protein abnormalities.2.A diagnostic algorithm for Alzheimer’s disease based on feature fusion of regions of interest and genetic data is proposed.The region of interest based on structural Magnetic Resonance Imaging is the hippocampus brain region with high levels of amyloid deposition.After strict quality control,the gene loci with significant differences were screened out and the mutant polymorphic bases are encoded with gray values.Each base pair should have a gray value,and 3D gray image is generated as another important discrimination feature.The Treble Residual Net(TRNet)is used for feature extraction and selection of ROI-based and SNP-based information.The residual network greatly retains the feature heterogeneity and reduces gradient dissipation when extracting the feature maps.The disease probability of each subject is extracted from the output of the residual network,which serves as the feature input of the integrated classifier.The classifier of this method adopts the integrated learning model of structural magnetic resonance imaging and gene data.The basic learner takes the left hippocampus,right hippocampus and gene coding image as input respectively,and generates a model summarizing these image features.The probability obtained by each basic classifier completes the diagnostic classification through the integrated learning of random forest.In this chapter,experiments are carried out on the separate classification,integrated classification and different integration methods for each feature.The experimental results show that the random forest integrator has the best predictive performance when fusing genetic data in magnetic resonance imaging.The two research methods proposed in this thesis provide heterogeneous features for computeraided diagnosis from different perspectives,thus diagnostic accuracy is improved.In addition,the correlation of genetic imaging is also analyzed to identify the polymorphic genes that cause pathological features,which will contribute to the prevention and control of AD earlier.