Research On Earlv Diagnosis And Prediction Method Of Alzheimer’s Disease Based On Multi-feature Fusion

Alzheimer’s disease(AD)is a common type of neurological dementia,which is serious and causes a huge burden on the patient’s family and social medical care.At present,there is no cure for AD,and the real hope to overcome it lies in early recognition and intervention.Mild cognitive impairment(MCI)is the prodrome stage of AD.In some cases,effective treatment of MCI patients may delay symptoms or even recovery.Studies have shown that about 10%～15%of MCI patients convert to AD every year,and MCI patients who don’t convert within a certain period of time may be stable in this state.Therefore,early identification of MCI patients and prediction of their further conversion to AD will help patients receive treatment as soon as possible,and can improve the therapeutic effect.Magnetic resonance imaging(MRI)can give an expression of the tissue structure of the brain,and it has been used in the clinical diagnosis of AD.In this paper,we put forward the idea of using the fusion of various morphological and radiological features of MRI,combining the brain tissue morphology and image texture information to maximize the presentation of image features,and then researched on early diagnosis and prediction of AD.The specific contents of this paper are as follows:Firstly,based on the data of AD,MCI,normal control(NC),progressional MCI(pMCI)and stable MCI(sMCI),Freesurfer was used to achieve the pre-processing of MRI,and then the techniques of extracting various of morphological features and radiological features were studied.Then,in order to avoid adverse effects on model performance due to too high feature dimensions,this paper designed a hybrid feature selection algorithm based on Max-Relevance and Min-Redundancy(mRMR)and Support vector machine-Recursive feature elimination(SVM-RFE)to obtain the low-dimensional optimal feature subset with maximum correlation,minimum redundancy and best discriminative ability.Then,SVM were used to construct the AD/NC,AD/MCI,MCI/NC,AD/MCI/NC and pMCI/sMCI classification models,so the early diagnosis and prediction of AD based on mixed feature selection were completed.At last,the advantage of this method was proved through experiments,the accuracys of AD/NC,AD/MCI,MCI/NC,AD/MCI/NC and pMCI/sMCI classification were 93.5%、90.1%、87.1%、82.6%和 81.3%.What’s more,the sensitive brain regions closely related to AD were analyzed and according to the results of feature selection,so as to explore the diagnostic markers and transformation predictors of AD.Next,considering that the tandem fusion strategy cannot describe the complex interrelationships between different types of features,this paper proposed a multi-feature fusion algorithm by combining Linear discriminant analysis(LDA)and Canonical correlation analysis(CCA).This algorithm uses CCA to achieve the goal of maximizing the correlation between two groups of features,and meantime introduces the category information of samples by LDA to to achieve that the samples of different categories have the maximum distance,and the samples within the same category have the minimum distance,that is,the samples with the best separability can be obtained.Then,each classification model was constructed by combining the kernel SVM classifier,so as to realize the early diagnosis and prediction of AD based on the discriminant canonical correlation analysis.Finally,the accuracys of AD/NC,AD/MCI,MCI/NC,AD/MCI/NC and pMCI/sMCI classification were 95.9%,92.4%,90.0%,85.7% and 84.5%.