The Study Of Identify Progressive Mild Cognitive Impairment Based On Functional Neuroimaging

Mild cognitive impairment(MCI) is the best stage to prevent and treat Alzheimer's disease(AD). But not all patients with MCI will progress to AD. So it is necessary to distinguish patients with non-progressive MCI(MCI-NC) from progressive MCI(MCI-C)before we can provide effective intervention in MCI period. Therefore, this paper has been investigated the classification problem between MCI-C and MCI-NC based on the information of the magnetic resonance image(MRI) and positron emission tomography(PET). Finally 64 subjects, where 32 of them are patients with MCI-C and 32 with MCI-NC,are selected from the ADNI database for inducting our study in order to establish the classification model who have a good performance.1) Image preprocessing. The basic information of 64 MCI patients from the ADNI database has been selected using the method of control variables to make MCI-C and MCI-NC matching by age. We strictly preprocessed the MRI and PET images at each step,i.e., head movement correction, image segmentation and standardization, and smooth.2) Getting brain gray matter voxel value. 90 brain regions templates of AAL brain partition has been created and the values of the corresponding brain gray matter has been gotten by SPM8, WFU, Get-totals and other tools. finally a dataset with a 64×180 dimension has been obtained for our study.3) Mild cognitive impairment recognition using multi-fusion method. Based on our study, PET modal information has been gotten better performance than MRI information in classifying MCI-C and MCI-NC. To consider supplementary effect of different modal information, this work combined these two biomarks into a strategy of information fusion,and applied genetic algorithm to find the best weights of them. The classifier is constructed using support vector machine algorithm. The results shows that the accuracy, sensitivity and specificity with the multi-fusion classification method based on genetic algorithm are better than single-mode method.4) Mild cognitive impairment recognition based on ensemble learning method. We used a random projection method to reduce the dimension of the original data which we have been obtained, then we applied it into a two-stage ensemble classifier. Finally, the results we achieved shows that our two-stage ensemble classifier based on random projection can obtain performance with an accuracy of 74.22%, a sensitivity of 66.25% and a specificity of82.19%. Compared with the other works, our work improved the classification performance between MCI-C and MCI-NC.