Multi-modal Image Data Fusion Analysis For Alzheimer’s Disease

Alzheimer’s Disease(AD)is a neurodegenerative disease.Mild Cognitive Impairment(MCI)is the early stage of AD.The pathological information presented by patients at this stage is not obvious,it is easy to be mistaken for natural aging.Once the disease worsens,it is difficult to reverse it.Therefore,early detection and treatment of the disease is particularly important.Early studies of the disease used single modal data for analysis,and the results were not ideal.In recent years,with the development of medical imaging and computer technology,Multimodal data and data fusion methods have attracted extensive attention in the field of neuroimaging.Multimodal neuroimaging data usually have high dimensions and complexity.It is the focus of this paper to find efficient methods to extract valuable features from complex data sets.In order to solve the problems of insufficient biological significance,large error and low accuracy of disease diagnosis and classification in previous methods,three image data fusion techniques,namely pixel-level fusion,feature-level fusion and decision-level fusion,were applied to the classification of Alzheimer’s disease.The main work is divided into the following three points:(1)Summarize the research status of multimodal image data of Alzheimer’s disease at home and abroad as well as the mainstream data fusion methods.Introduce the commonly used image data and data sources in Alzheimer’s disease research,and the specific pretreatment process of brain image data was described in detail due to the unique structure and principle of brain image data.(2)In order to verify that compared with the single mode data,the multi-mode data has the advantages of complementarity of information and smaller error.The two kinds of preprocessed brain image data were fused with pixel-level data to obtain the new image data containing the two kinds of modal data information.The features of the two kinds of preprocessed brain image data and the single modal data were extracted and input into the classifier respectively to compare the classification effect.In addition,in order to prevent accidental results,three feature extraction methods and three commonly used classifier models were respectively used in the experiment,and the classification effects of multi-mode data and single-mode data obtained in the same feature extraction method and classifier were compared for many times.The experimental results further validate the superiority of multimodal data in the diagnosis classification of AD.In addition,the single classifier and the ensemble classifier are selected by the classifier,and the classification effect of the two classifiers is compared under the control of other variables.The experimental results verify that the ensemble classifier based on the decision level fusion strategy has better classification effect.(3)Study the application of feature level data fusion technology in the classification of Alzheimer’s disease.Canonical Correlation Analysis(CCA)is a commonly used method for feature-level fusion.However,the existing CCA-based fusion methods have problems such as high dimension,multicollinearity,single-mode feature selection,asymmetry and spatial information loss when reconstructing imaging data into vectors.This paper uses a new Structured and Sparse Canonical Correlation Analysis(SSCCA)technique to solve the above problems.Set three groups of experiments,and use different feature fusion methods under the same other conditions to verify the application effect of ss CCA feature fusion method.The results show that this method performs well in this project.