Research On AD Classification Method Based On Optimized Sparse Representation

Alzheimer’s Disease(AD)is a serious neurological disease that mainly appears in the elderly.It is mainly manifested as progressive perception and cognitive deficits,and ultimately complete loss of daily behavior.Due to its irreversibility the early diagnosis of the disease has important theoretical and clinical significance.This paper has proposed two new AD diagnosis methods that classification method based on the brain network construction algorithm of Weighted Graph Sparse Group Representation(WGraph SGR)and an AD diagnosis method based on ensemble learning of optimized Sparse Representation(SR).The conduct experiments has done on it,and compare with other methods to verify the effectiveness.The main work of this paper as follows.First,a classification model of Alzheimer’s disease based on WGraph SGR network construction algorithm is proposed.This method is based on WGraph SGR,which can construct the best brain functional network based on Rest State Functional Magnetic Resonance Imaging(RS-f MRI)data,and further provide classification algorithms.Pearson Correlation(PC)and Sparse Representation(SR)are two most commonly used brain network modeling methods.We take the advantages of both to ensure the construction of a more biologically meaningful brain network by integrating a unified framework of connection strength,group structure and sparsity.Our proposed method has been validated in AD and Health Control(HC)classification tasks,and has obtained better results compared with other brain network construction methods.Second,an AD classification algorithm based on optimized SR ensemble learning is proposed.Since the classification results of multiple different SR methods based on Support Vector Product(SVM)are integrated,the generalization ability and classification effect of the method can be improved.We mainly use the maximum voting method to integrate the classification results of different methods.First,adopting more accurate and biologically meaningful brain function network modeling methods to guide the construction of brain function networks.Then,a combination of multiple brain network modeling methods is used to construct SVM classifiers respectively.Finally,we use ensemble learning to take advantage of the above-mentioned classifiers while suppressing classification errors,thereby synthesizing the results of multiple classifiers to achieve the effect of reducing errors and improving classification accuracy.The experimental results proved that the classification accuracy obtained by our proposed method is nearly 5 percentage points higher than that obtained by the traditional SR method.