Classification Analysis For A Lzheimer's Disease Based On Structural Magnetic Resonance Imaging

Alzheimer's disease(AD)is the most common kind of cognitive disorders among old people.The onset of AD is insidious and it would do harm to old people's health seriously with disease progress,so it is vital finding a useful method to diagnose and treat AD patients in time.Structural Magnetic Resonance Imaging(sMRI)has played an important role in clinical and scientific research of AD with the advantage of its non-invasive and non-radioactive.sMRI shows 3D structural information of human brains including volumes of brain tissues,cortical thickness and surface,providing effective evidences in the diagnosis of AD.At present,there are so much AD classification research based on sMRI.However,most of them collected sMRI data from only one site or Alzheimer's Disease Neuroimaging Initiative(ADNI)public database,without considering the influence of center.So the generalization ability on data from unseen centers of the derived results could not be guaranteed.In this paper,we proposed a simple feature extraction method based on the Human Brainnetome Atlas with 227 AD patients and 226 NC from ADNI database.Firstly,mean gray matter density of each brain region in the Brainnetome Atlas was obtained as feature.After that,an SVM(Support Vector Machine)model was introduced to classify AD from NC samples.The result showed that the mean accuracy was 85.2%with 10-fold cross validation.And post hoc analysis demonstrated that the mean gray matter density of several brain regions such as the hippocampus,amygdala and fusiform gyrus played important roles in classification and the atrophy of these regions had significant correlation with cognitive ability measured by the Mini-mental State Examination(MMSE)scores in AD patients.In addition,we utilized the Least Absolute Shrinkage and Selection Operator(LASSO)to predict individual MMSE score and the predicted values showed a very high positive correlation with true values(r=0.65,p<0.001).In conclusion,the present study demonstrated that the gray matter density of brain regions defined by the Brainnetome Atlas was meaningful for distinguishing AD patients from NC samples and could also be used to predict individual cognitive ability.In order to test the generalization ability of the proposed feature extraction method on data from unseen center,we chose sMRI data of NC and AD patients from 6 centers and used the same feature extraction method to get mean gray matter density of each brain region in the Brainnetome Atlas for each sample.After that,a Linear SVM model was introduced to classify AD from NC samples.The result showed an 85.84%mean accuracy with inter-site cross-validation.Meanwhile,we also got an 84.07%mean accuracy with cross validation among 3 datasets(our 6 center sMRI data,ADNI data and The European Diffusion Tensor Imaging Study on Dementia data).Finally,we analysed features which played important roles in distinguishing AD patients from normal controls during inter-site cross-validation of 6 center data and found the mean gray matter density of hippocampus,parahippocampal gyrus,fusiform gyrus,superior temporal gyrus,postcentral gyrus,basal ganglia and precuneus played important roles in AD classification.These results suggest that our feature extraction method is robust to the data heterogenicity introduced by different centers and could be helpful for AD diagnosis in practice,relevant features could be regarded as biomarkers in AD researches helping us understand the pathology of AD.