Classification Study Of Alzheimer’s Disease Based On Complex Brain Networks

Alzheimer’s disease is a neurodegenerative disease that destroys brain cells,leading to abnormalities in memory,cognition,thinking,and behavior,and seriously affecting people’s work and life.Alzheimer’s disease mainly occurs in the elderly,and the increase in life expectancy has led to a rapid increase in the number of dementia patients.Research has shown that early diagnosis can effectively slow down the development of AD.With the rise of machine learning,the use of machine learning for early diagnosis of Alzheimer’s disease has become a hot topic of research.However,traditional machine learning methods are not stable in classification performance when faced with complex samples.Although deep learning has good applicability,it does not have interpretability and cannot provide us with effective information.Based on the above content,this article conducts the following research:(1)We proposed a subclass weighted logistic regression(SWLR)model based on logistic regression.The SWLR model introduced the concept of subcategories on the basis of the logistic regression model,and constructed a subcategory coefficient matrix for each subcategory through adaptive weighting of different subcategories.This enabled the model to better handle linearly indivisible samples compared to logistic regression,and also improved the problem of reduced classification accuracy caused by individual differences in classification problems.In addition,we also introduced L2 regularization to solve the problem that traditional logistic regression is prone to overfitting.(2)Effectively improving the accuracy of AD classification by combining SWLR with HCPMMP.We combined SWLR with HCPMMP,the most sophisticated brain partitioning method currently available,and conducted classification research based on complex brain network indicators such as degree,efficiency,and eigenvector centrality.We used the ADNI dataset to evaluate the model.The results of cross validation showed that the proposed model had good performance,with HC vs.AD accuracy of 95.8%,an improvement of8.3% compared to LR,HC vs.EMCI accuracy of 91.6%,an improvement of 10.4%compared to LR,HC vs.LMCI accuracy of 93.7%,an improvement of 8.3% compared to LR,EMCI vs.LMCI accuracy of 89.5%,an improvement of 6.3% compared to LR,and LMCI vs.AD accuracy of 91.6%,an improvement of 8.4% compared to LR.(3)We extracted and analyzed core brain regions based on SWLR,determined the distribution of core brain regions,and derived the migration pattern of core brain region distribution based on the distribution of core brain regions.We analyzed the coefficient matrix based on the SWLR classification results,screened out the core brain regions of different stages of Alzheimer’s disease,and used HCP workbench to locate the core brain regions.We found that according to the disease development process of HC-EMCI-LMCI-AD,the distribution of the core brain regions showed a clockwise migration trend on the two-dimensional image.In order to prove this discovery,we conducted two rounds of experiments.In the first round,we only used core brain regions to construct a feature matrix and used SWLR to classify.The binary classification results of the 20 core brain regions of HC vs.EMCI,EMCI vs.LMCI,and LMCI vs.AD were 83.3%,81.3%,and 85.4%,respectively.The binary classification results of the 30 core brain regions of HC vs.EMCI,EMCI vs.LMCI,and LMCI vs.AD were 87.5%,85.4%,and85.4%,respectively.which is not significantly different from the classification accuracy using all brain regions.This proved that the brain regions extracted by using the coefficient matrix are indeed core brain regions.In the second round of experiments,we randomly selected an equal number of brain regions from non-core regions for five sets of comparative experiments.The results showed a significant difference in classification accuracy between core and non-core brain regions,proving that the division of core regions is correct.The combination of two rounds of experiments has proven the rationality of our inference about the regular migration of core brain regions.Finally,we compared it with existing research and obtained corresponding theoretical support.