Research On Gray Matter Atrophy Of Alzheimer’s Disease

Alzheimer’s disease(AD) is a chronic neurodegenerative disease that usually happened in the elderly. The most common symptom is memory loss and other cognitive function decline. The AD patients are usually accepted treatments until they have become moderate dementia. Therefore, it has limitted effect for preventing its progress and early intervening. Currently, diagnosis of AD is mainly based on neuropsychological tests. The magnetic resonance imaging(MRI) can reveal gray matter atrophy and has been widely used for early diagnosis of AD. Based on structural magnetic resonance image, we analyzed the difference of gray matter volume and atrophy pattern between different stage of Alzheimer’s disease. The gray matter volume was also used for automatic classification and cognitive score prediction model for Alzheimer’s disease.We applied voxel-based morphometry(VBM) to compare gray matter volume among normal controls(NC), stable mild cognitive impairment(s MCI), convert mild cognitive impairment(c MCI) and Alzheimer’s disease group, and investigate atrophy pattern using longitudinal imaging data of subjects. We obtained the voxels with significant differences among groups, and these voxels were further reduced by support vector machine recursive feature elimination. Then selected voxels were used to create classification model. Moreover, a prediction model for Mini-Mental State Examination(MMSE) score is adopted according to the voxels changing in longitudinal anaalysis.The VBM analysis showed gray matter atrophy of Alzheimer’s disease is a gradual process.The atrophy began from the temporal lobe, insula, basal ganglia, then extended to parietal and occipital lobes and cingulate cortex, and ended with frontal lobe. c MCI and AD had similar gray matter atrophy pattern,with fast speed and wide range. Whereas,s MCI and NC had similar gray matter atrophy pattern, with slow speed and narrow range. Small differences in gray matter between s MCI and c MCI existed at baseline, but it increased gradually with time. The classification accuracy of NC/MCI, NC/AD and MCI/AD, s MCI/c MCI could reach 84.9%, 100%, 81.4% and 84.9% respectively. Lateral and medial temporal lobe were important regions of feature voxels in all four classification comparisons. Region of interest analysis and feature voxels suggested that posterior cingulate gyrus, supramarginal gyrus and precuneus are crucial to distinguish s MCI and c MCI in early stage. In cross-validation test of MMSE score prediction model, mean square error was 2.5460, correlation coefficient of real and estimated score was 0.8191.