The Research Of Brain Structures Of MCI Based On Surface Morphology

Brain is the most important and complex organ in our bodies. It has great means to study its structures and mechanism to better protect it. Mild cognitive impairment (MCI) is thought to be a transition stage between normal control (NC) and Alzheimer’s disease (AD). MCI subjects has mild cognitive decline and unstable condition. As AD is an irreversible disease, so it’s important to research MCI to reduce the death rate of AD.Surface based morphometry is a new analysis method and has been proved to be sensitive in the brain structure measure especially in some gyrus. However, most studies focused on the horizontal comparisons between MCI and AD, other aspects are rarely to be researched. Our study hope to detect the abnormalities in the MCI based on the magnetic resonance imaging (MRI) from different aspects to better understand the pathology of MCI and prevent or treat AD. Our main contributions are as follows:1. As no studies research the longitudinal changes of MCI, we explored the longitudinal cortical changes of MCI based on the surface based morphometry. After the steps of preprocessing, some surface parameters were obtained. We compared the intra-group differences and inter-group differences of MCI and NC in two years, analyzed the atrophy rate difference and the correlation between atrophy and mini-mental state examination (MMSE) scores. Furthermore, we researched the atrophy trend of regions in MCI group at four time points. We found some abnormal brain areas in MCI such as the left hemisphere temporal gyrus, insular gyrus, parahippocampal gyrus and fusiform gyrus. These regions atrophied linearly and faster than NC and is responsible for the clinical features.2. We first used the cortical thickness to construct the cortical network of MCI and NC and found the differences in the network properties. The cortical thickness was partitioned into78brain areas using the automated anatomical labeling (AAL). Then the thickness of brain areas was computed to get the partial correlation coefficient between regions and the cortical network was built. We used the permutation test, the Fisher’s Z transformation to compare the differences between MCI and NC in average clustering coefficient, average absolute path length, betweenness centrality and brain region’s correlations. We found MCI showed greater clustering coefficient, longer average absolute path length, changes in hub nodes and interregional correlations. The results were consistent with previous functional or VBM studies and proved the facts that MCI changed the message processing of brain in some ways.3. To combine the medical imaging technology and clinical diagnosis, we proposed a feature selection method based on the cortical thickness to classify the converted MCI (CMCI) group and stable MCI (SMCI) group. Firstly, we compared the cortical differences at baseline time between NC, CMCI and SMCI. We found the cortical thickness relationship is NC>SMCI>CMCI. Then we partitioned the brain into78areas and computed the regional cortical thickness as the feature vectors. One feature selection method is according to the significant levels in thickness between CMCI and SMCI. The other one is according to the support vector machine (SVM) recursive feature elimination (RFE) algorithm. And we also did the successive grid search technique to find the optimal model parameter values for radial basis function (RBF) kernel. The accuracy under the leave-one-out cross validation reached76.77%and showed it’s possible to achieve the early prediction of MCI using the medical imaging technology.