Research On The Abnormal Patterns Of The Hippocampus In Alzheimer's Disease Based On Multi-center Structure MRI

Alzheimer's disease(AD)is a neurodegenerative disease that occurs mainly in the elderly.The early stage of AD is generally manifested as memory decline.With the aggravation of AD and the change of brain structure,it shows the decline or even loss of language,calculation,emotion and other functions.Alzheimer's disease brings a burden to the families and the society.Mild cognitive impairment is high risk group for AD,compared with normal controls,MCI has a greater probability of converting to AD.If we can accurately identify the high-risk population in the early stage,we can delay the progress of the disease to a great extent.But unfortunately,up to now,AD still lacks effective and stable biomarkers.The pathogenesis of AD is not clear,and there is no effective medicine that cures this disease.It is undoubtedly of great significance for the early prevention and treatment of AD if a robust biomarker was found.The medial temporal atrophy is believed to be the best significant MRI marker at a prodromal stage of further progression to AD dementia,and hippocampal atrophy is the most robust.For later stage of AD,large-scale alterations of the brain in terms of localized atrophies in structures.which are detectable in MRI images.The volume and/or shape,however,only a crude measurement for the complex anatomical change that occurs in AD and often ignore the fact that atrophy is not uniform across different disease phase.Thus,in order to find more information about the pathogenesis of AD and more subtle changes in hippocampus,a new method which can get more detail information from MRI should be used.Radiomics,a method to analyze anatomy of regions of interest based on structural imaging data,including intensity,shape and texture,has previously been successfully applied to evaluate imaging biomarkers for various tumors and neurological disorders.However,it remains unclear whether hippocampal radiomic features,including intensity,shape and texture,are reproducible and robust as predictors of progression from mild cognitive impairment(MCI)to AD dementia and neurobiological foundation.The primary aim of this study was to verify whether hippocampal radiomic features are robust MRI markers for AD.Multivariate classifier based support vector machine(SVM)provided individual-level prediction of AD(N=261)from normal controls(NCs,N=231)with accuracy(ACC)=88.21%(SPE=89.49%,SEN=86.81%)with inter-site cross validation,and the intra-site cross validation results are slightly better as ACC=89.19±0.04%(SPE=90.14±0.05%,SEN=88.10±0.06%)based on the data from 6 sites.Further analysis on an large independent ADNI data(N=1228)as well as in-house data(N=715)reinforced that the hippocampal radiomic features are robust imaging markers of AD,more importantly,the classification output were highly correlated with the cognitive ability.The second key goal was to determine whether these radiomic features relate with neurobiological basis.A systemic analysis demonstrated that informative features identified with high consistency were significantly associated with the clinical information,including APOE,polygenic risk scores,CSF A(3,CSF Tau,longitude changes of cognition.These results suggest that hippocampal radiomic features can serve as a robust biomarker for clinical application in AD.The second task of this paper is to explore whether PET amyloid image radiomics features can also be used as a new biomarker for AD.The multi-variable machine learning model constructed by support vector machine can be used to classify AD and NC with ACC=0.89,SPE=0.90,SEN=0.86,AUC=0.95.Only using the radiomics features of hippocampus of structural magnetic resonance,the results showed that ACC=0.82,SPE=0.87,SEN=0.76,AUC=0.90.If the radiomics features of hippocampus and PET were combined,ACC=0.92,SPE=0.93,SEN=0.91,AUC=0.97.Furthermore,NC and MCI were classified as ACC=0.66,SPE=0.43,SEN=0.79,AUC=0.64 when combined PET and MRI.MCI and AD were classified as ACC=0.78,SPE=0.86,SEN=0.62,AUC=0.84.The results showed that PET radiomics features can also be used as an effective biomarker for AD.The recognition effect of AD using PET and MRI is better than that of single modal.Radiomics based on multi-modal image is more helpful for the diagnosis of early Alzheimer's disease.In this study,we found that radiomics features is a robust biomarker for AD.Combining multi-modal radiomics features may provide us with a good method and tool to identify early AD/MCI,and provide a new direction for us to better understand the etiology and pathogenesis of AD.