Research On Classification And Prediction Of Alzheimer’s Disease Utilizing Machine Learning

Alzheimer’s Disease(AD)is a kind of progressive neurodegenerative disease,which occurs most frequently in aged population.The common symptoms of AD include memory decline,language difficulties,cognitive impairment,etc.AD patients often suffer from loss of the ability to take care of themselves,which impose heavy burden on the patients’ families.With the accelerated population aging process in the world,AD has been one of major diseases in elderly people.Currently,the primary causes of AD are not clear and there is no specific cure for AD.AD classification and prediction is very helpful for timely treatment and intervention in its early stage so as to reduce the incidence rate of AD and delay AD progression in elderly population.The AD classification task is to classify the NC(normal cognitive)and AD groups,and the AD prediction task is to predict whether MCI(mild cognitive impairment)converts to AD or not in the future.In this paper,the MRI(magnetic resonance imaging)data are applied to research on AD classification and prediction utilizing machine learning.The methods on AD classification and prediction were studied.The deep feedforward network(DFN)was improved utilizing densely connected method to build the densely connected deep feedforward network(DCDFN)model.Then the method on AD classification and prediction utilizing DCDFN was proposed.Combined with age and gender,the volumes of density map and subcortical tissue segmented using FSL were taken as features for AD classification and prediction.The 5-fold cross-validation method is utilized to evaluate model performance on ADNI dataset.The experimental results show that the proposed method on AD classification and prediction is effective and better than some traditional machine learning models.Furthermore,the accuracy of AD classification and prediction is effectively increased by adding the features of age and gender.The feature selection methods on AD classification were also studied.A multi-criteria fusion feature selection method by cumulative weighted was proposed,which normalized evaluation results of 10 feature selection methods base on filter,wrapper and embedded respectively.The importance of features was evaluated utilizing cumulative weighted feature importance score and the features for AD classification were selected in accordance with the threshold of feature importance.The feature selection methods were evaluated on the TADPOLE dataset.The experimental results show that the proposed feature selection method is effective and the optimal thresholds of feature importance of various classifiers are usually different.Using the proposed feature selection method,the DCDFN model is better than other traditional classification model on AD classification.