Mild Cognitive Impairment Classification Based On Scale Combination

With the development of times,the healthcare has been gradually improved,which cause dincreasingly obvious phenomenon of aging population.An aging population brings about increment of elderly disease,and Alzheimer disease(AD)is one of elderly disease with high incidence,additionally,an effective treatment for which has not been found yet.Mild cognitive impairment(MCI)as middle state from normal to AD is given great attention in medical field and early discovery and treatment of MCI can help reduce the incidence of AD.Currently,there are some principal detection methods,including psychological scale test,body fluid detection and imaging detection,furthermore,owing to easy performance,psychological scale test was used to evaluate disease severity clinically.However,some shortcomings of psychological scale test are not ignored,such as low accuracy,and un-consideration for effect of different tests evaluating illnesses.This paper uses data mining algorithms to detect and classify MCI.The data set used in this article includes 15 AD patients,12 patients with MCI,and 14 persons as normal controls.For each test group,three different test scales were used for testing: Mini-mental State Sxamization(MMSE),Alzheimer’s disease assessment scale-cognitive(ADAS-cog)and Chinese Character Writing Scale(CCWS).In this paper, CCWS is a part of the Chinese Aphasia Test(ABC)in the Neuropsychology Research Department of the Affiliated Hospital of Beijing Medical University.This article uses it to verify whether the writing of Chinese characters has significant significance for MCI.In this paper,support vector machine-recursive feature elimination(SVM-RFE)algorithm was used for feature selection,principal component analysis(PCA)algorithm is used to eliminate correlation among features and perform dimension reduction for features,and SVM algorithm is used to train classifier models.Ten-fold cross validation was selected to evaluate the generalization of models,and accuracy rate was chosen as evaluation criterion of models precision.The experiment is divided into two parts.In the first part,classification ability of a single scale was evaluated.It indicated that the ability of MMSE classifying AD is greater than that of ADAS scale,while ADAS’s ability for classifying MCI is greater than that of MMSE scale.According to the model evaluation of CCWS scale,Chinese writing has the value of detection for MCI,and classification accuracy for MMSE scale,ADAS scale and CCWS scale is 0.88,0.87,0.73 respectively.In the second part,according to merits and drawbacks of various scale,the classification performance of MCI is improved by feature fusion.Finally,the accuracy of three classifications is 0.93 through the form of Scale combination.The multi-classification model of the Scale combination used a total of 12 features,which was the same as the MMSE scale feature quantity and 7 feature levels less than the ADAS scale feature,and improved the classification accuracy without increasing the number of features.Therefore,the data mining method can effectively detect MCI,and the feature selection method can select a better combination of test items of the scale and improve the classification accuracy.Finally,the combination of scales improves the accuracy of classification without increasing the number of features.