Voxel-based Morphometry Analysis And Deep Learning Approach For Diagnosis Of Early Alzheimer’s Disease By Cerebral Grey Matter Changes

Objective The voxel-based morphometry(VBM)was used to analyze cerebral grey matter atrophy changes in mild cognitive impairment(MCI),and deep learning methods based on convolutional neural networks was used to diagnose early Alzheimer’s disease in cerebral grey matter changes.Methods17 MCI subjects and 13 normal cognitive subjects were recruited in the 2nd Affiliated Hospital,Medical College of Shantou University.Two groups underwent 3T structural magnetic resonance imaging(MRI)scanning.In addition,94 MCI subjects and 60 normal cognitive subjects with the same parameter structural magnetic resonance scan were selected from the Alzheimer’s Disease Neuroimaging Initiative(ADNI)database.Statistical parameter map(SPM8)software was used to analyze the obtained structural MRI data by VBM method,including the extraction of cerebral grey matter,statistical differences analysis,and the correlation analysis between cerebral grey matter of MCI and clinical cognitive scale scores.Then we selected the subjects from ADNI database as the training set,and 30 subjects collected from in our hospital as the test set.The deep learning method based on convolutional neural network(CNN)was used to extract features in grey matter images of the brain.ResultsCompared with normal cognitive subjects,MCI subjects had grey matter atrophy in the left parahippocampal gyrus,left central posterior gyrus,left superior parietal gyrus,and bilateral frontotemporal lobes(p < 0.0001)by the analysis of VBM method.The left parahippocampal gyrus,left inferior frontal gyrus,left parietal inferior horn gyrus,right putamen,and bilateral temporal lobe in the brain regions of MCI subjects showed a significant correlation with the decline of MMSE scores(p < 0.01).The accuracy,sensitivity,and specificity of the CNN-based deep learning method for identifying MCI subjects were 80.9%,88.9%,and 75%,respectively.And the area under the receiver operating characteristic curve(AUC)was 0.891.ConclusionThe VBM method suggests that grey matter atrophy in early AD is mainly concentrated in the entorhinal cortex,frontal and temporal cortex,which may be the brain structure basis for memory and cognitive impairment.And the deep learning method based on CNN has a good effect in identifying early AD by the changes of cerebral grey matter,providing a potential auxiliary tool for clinical diagnosis of early AD.