The Feature Analysis And Recognition Of Magnetic Resonance Diffusion Kurtosis Imaging For Alzheimer’s Disease

Alzheimer’s disease (AD) is a typical kind of neurodegenerative disease, withsevere memory cognitive function decline or obstacles, and there is no effectivetreatment and greatly affect the health of the elderly life quality. Diffusion KurtosisImaging (DKI) is a new developped diffusion MRI technique and its introduction ofadvanced kurtosis statisticsdescribing the nonGaussian diffusion of water moleculesin biological tissue on the basis of diffusion tensor imaging (DTI), which can be moresensitive to the microstructure characteristics and has been expected to play animportant role in early diagnosis of AD.Based on the tensor eigenvalue decomposition of high-order kurtosis, this thesisfirstly proposed the parametric mapping methods from the high-ordered kurtosistensor. Through the analysis of scatter plot and correlation between the parametersfrom the area-of-interest (ROI), kurtosis information performed outstanding todistinguish the different brain tissue, and the conjoint analysis of multiparameters ishelpful to the automatic image segmentation of different organizations. In correlationanalysis, the discovery of significant positive correlation between the anisotropicparameters and average kurtosis coefficient provides an important reference forfurther explore the biological significance and the diagnosis value of kurtosis.The MRI clinical imaging data of17AD and26healthy elderly (age, gendermatched) were acquired as the experimental group and the control group respectly,and its feature sets from DKI and DTI were optimized with the recursive featureselection based on support vector machine(SVM-RFE). The results showed that theaccuracy of AD detection of the optimal feature subset was significantly increasedcompared to that bifore the feature optimization. The significance of optimizedfeatures, ROC curve analysis and correlation analysis were also conducted then, and itwas found that firstly the pattern recognition results could be used to analysis theconjoint form of characteristics in diagnosis and that the machine learning resultscould reflect the real biology changes, which has a great clinical significance;secondly kurtosis indexes had high sensitivity to the mico changes of AD, and alsorevealed that the diffusivity and kurtosis in lesion detection of white mattermicroscopic myelin damage had complementary effect.This research, from the view of MRI diffusion kurtosis imaging feature set combined with pattern recognition in this thesis, was an exploratory researchcombining artificial intelligence and DKI. All the results were hoped to be helpful tothe DKI application in early diagnosis and biological analysis of Alzheimer’s disease.