Classification Aided Diagnostic Model For Alzheimer’s Disease Based On Structural MRI And Ensemble Learning Algorithm

Objective:The aims of this study were to use different feature selection methods and ensemble learning algorithms to establish Alzheimer’s Disease(AD)classification aided diagnosis models based on the data set of structural magnetic resonance image features and clinical information,and to identify diagnostic markers in the progression of AD.This study may provide statistical decision support for realizing early warning of disease risk and automated clinical diagnosis of AD.Methods:The data we used were from the Alzheimer’s Disease Neuroimaging Initiative(ADNI)with a total of 493 subjects,including 125 normal control(NC),121 early mild cognitive impairment(EMCI),109 late mild cognitive impairment(LMCI),and 138 patients with AD.The information of Structural Magnetic Resonance Imaging(sMRI)and clinical information(including age,gender,years of education,marital status,and cognitive assessment scales)were collected.Voxel based morphometry(VBM)was used to extract the features of MRI neural images.Support vector machine recursive feature elimination(SVM-RFE),L1 regularized logistic model,and feature selection algorithm based on gradient boosting decision tree(GBDT)were used for preprocessing to eliminate some redundant features to simplify the combined classification model.Based on the data set of sMRI combined with clinical information,support vector machine(SVM),random forest(RF),Ada Boost,GBDT,and Stacking integrated strategy were used to establish AD classification aided diagnosis model,using 10-fold cross validation.Stacking integrated algorithm belongs to the multi-classifier combination method.In this study,SVM,RF,Adaboost,GBDT stability learning algorithms were used as the primary classifier for combination learning,and the secondary classifier used logistic regression algorithm to build AD classification aided diagnosis model.The indicators of the evaluation model included accuracy,sensitivity,specificity,F1 score and Area Under ROC Curve(AUC).Results:In this study,sMRI features were used to establish AD classification aided diagnosis models.The Stacking ensemble model performed better than the four single classification models of SVM,RF,Ada Boost,and GBDT.The accuracy ranges of NC-EMCI,NC-LMCI,NC-AD,EMCI-LMCI,EMCI-AD and LMCI-AD in a single classification model were66.23%～66.72%,67.48%～71.36%,83.68%～87.07%,66.96%～68.70%,79.54%～81.82%,69.62%～73.33%,AUC ranges were 0.6898～0.7098,0.7065～0.7790,0.9255～0.9412,0.7051～0.7558,0.8637～0.8380,0.7757～0.8001,respectively;for Stacking model,accuracy rate were 74.32%,77.46%,88.61%,75.36%,88.46%,74.67%,and AUC were0.8393,0.8149,0.9591,0.7943,0.9382,0.8415,respectively.Then data set of sMRI features and clinical information were used to establish AD classification aided diagnosis models.The Stacking ensemble model still performed better than the four single classification models of SVM,RF,Ada Boost and GBDT.The accuracy ranges of NC-EMCI,NC-LMCI,NC-AD,EMCI-LMCI,EMCI-AD and LMCI-AD in a single classification model were 71.15%～75.97%,79.47%～82.86%,95.04%～96.96%,70.43%～73.91%,94.57%～95.77%,83.78%～87.83%,AUC ranges were 0.7799～0.8208,0.8761～0.9112,0.9872～0.9905,0.7750～0.8083,0.9857～0.9886,0.9320～0.9548,respectively;for Stacking models,accuracy rates were 81.08%,85.91%,97.47%,78.26%,97.44%,86.67%,and AUC were 0.8724,0.9254,0.9987,0.8467,0.9967,0.9681,respectively.In the classification of NC-EMCI,NC-LMCI,NC-AD,EMCI-LMCI,EMCI-AD,LMCI-AD,the accuracy of the models based on combined features were improved by 6.76%,8.45%,8.86%,2.90%,8.98% and 9.34% respectively.The accuracy of stacking models combined with features from high to low were NC-AD,EMCI-AD,LMCI-AD,NC-LMCI,NC-EMCI,EMCI-LMCI.Conclusion:In this study,the stacking ensemble algorithm was used to establish multi classification auxiliary diagnosis model for AD,to realize the early screening,recognition and early warning of high-risk individuals of AD in the elderly population,and to form the diagnosis mode of AD combining cognitive impairment and abnormal neuroimaging markers,so as to provide reference for early monitoring the progression of MCI to AD,identifying the early stage of AD and slowing down the onset of AD.Moreover,it may provide methodological reference for researches on early precise prevention and treatment of other brain diseases.