Analysis On Brain Structural MR Images And Classification Of The Patients With Mild Cognitive Impairment

The Alzheimer’s disease(AD)is a neurodegenerative disease that usually occurred in the elderly,and the mild cognitive impairment(MCI),whick includes the stable MCI(s MCI)and the converted MCI(c MCI),has the high risk to convert to the AD.It is a crucial task to investigate the similarities and differences in brain structure between the s MCI and c MCI,and explore the differences between their normal elderly and patients with AD,which provides valuable information to reveal the changes in the brain and convertion of the MCI,control and intervention in the disease progression.MR images of brain structures from 219 subjects were selected,including 72 subjects of the normal elderly as the normal control(NC),46 subjects with s MCI,40 subjects with c MCI and 67 subjects with AD.The MR images of each subject were acquired in three time points,that is,the baseline(frist scanning),12 months and 24 months.First,the images were pre-processed and four types of the cortex feature indluding cortical thickness,gray matter volume,surface area and the mean curvature,were calculated.Second,the two-sample t-test was used to analysis the differences between the NC,s MCI,c MCI and AD group.Third,according to the baseline and longitudinal variation of cortical features,recursive feature elimination(RFE)and sparse constraints dimensionality reduction(SCDR)were applied to reduce the dimension.Fourth,the the support vector machine(SVM)classifier was performed for discrimination between the four compared groups.We can get SVM-RFE model by combining SVM and RFE,and get SVM-SCDR model by combining SVM and SCDR.The dimensionality reduction method was combined with support vector machine,Finally,the performance of different cortical features and the two models were compared and analyzed.The experimental results showed that the changes of brain cortex were a gradual process in the AD,especially,the regions of the temporal lobe suffered the extensive atrophies.For the s MCI,the changes of the cortical thikness and the mean curvature were mainly in the temporal lobe,whereas the variation of gray mater volume became remarkable for the c MCI,and the differences of the surface area are unsignificant until the MCI developed to AD.In terms of classification performance,the SVM obtained better results using the cortical thickness and gray matter volume,followed by the surface area and the mean curvature,in particular,the combine cortical features,as well as the baseline features and the longitudinal changes,were used to significantly improve the classification performance,and the SVM-RFE outperfomed the SCDR.This work demonstrated there were significant differences in brains of the NC,s MCI,c MCI and AD cohorts,and the atrophies in cortex were a gradual process from the NC to the MCI and the AD.The cortical features had potentials to distinguish different stages of AD,evaluate the risk of the MCI conversion.