| Alzheimer’s disease(AD)is the disease with the highest prevalence in the elderly over the age of 60.Mild cognitive impairment(MCI)which can be divided into converted mild cognitive impairment(cMCI)and stable mild cognitive impairment(sMCI)is the early stage of Alzheimer’s disease and has a high risk of developing into Alzheimer’s disease.Support vector machine(SVM)can be used for classification recognition,however,the choice of model parameters has a greater impact on the classification results.Therefore,Combining structural magnetic resonance imaging(MRI)data and optimization theory,this paper proposes an early diagnosis algorithm for Alzheimer’s disease based on optimized support vector machines.The specific work is as follows:Firstly,a comprehensive diagnosis algorithm for Alzheimer’s disease based on nested cross-validation optimization support vector machines(CV-SVM)proposed to recognize the overall six classification tasks of Alzheimer’s disease and its early stage mild cognitive impairment(MCI).Combining a support vector machine(SVM)model with Gaussian radial basis kernel(RBF)and nested cross-validation(CV)which the inner nested select optimal parameters of model to establish a nonlinear support vector machine classifier.Preprocessing the MRI images and extract the gray matter volume(GMV)features of the region of interest(ROI)as biomarkers to identify each stage of Alzheimer’s disease.The results show that the six tasks achieved good classification results.The stable MCI and converted MCI groups achieved77.79% accuracy,which was 11.42% higher than the basic support vector machine.And compared with the Decision tree(DT)and K-Nearest Neighbor(KNN)classifiers,it also improved by 14.57% and 12.88%.Secondly,an early diagnosis algorithm for Alzheimer’s disease based on Adaptive mutation particle swarm optimization algorithm optimized support vector machine model(AMPSO-SVM)is proposed,mainly identifying cMCI patients from three groups of AD,s MCI patients and healthy controls(NC).Each iteration operation uses the fitness variance of the population and the current optimal fitness value to dynamically determine the mutation probability.When the probability is less than the random number,the mutation operation is performed to help the particle jump out of the current search position through reinitialization,avoiding the local optimal problem of the traditional particle swarm optimization(PSO)algorithm.Then the improved adaptive mutation particle swarm optimization(AMPSO)is used to optimize the SVM to solve the problem of selecting its regularization parameters and kernel parameters.The experimental results show that the AD and NC groups have the best classification effect,followed by the cMC and NC groups.The sMCI and cMCI groups achieved 81.3282% accuracy,which were 14.96%,3.54%,and 3.04% higher than the basic SVM,CV-SVM,and PSO-SVM,respectively;compared to the basic SVM,the accuracy of cMCI-NC,cMCI-AD,and AD-NC increased by 12.11%,13.94%,and 10.87%.Finally,an mild cognitive impairment diagnosis algorithm based on support vector machine with hybrid optimization of sine cosine algorithm and particle swarm optimization(HSCAPSO-SVM)model is proposed,which mainly recognizes mild cognitive impairment converters and non-converters.Making full use of the global search performance of the sine and cosine algorithm(SCA)and the advantages of fast local convergence of particle swarm optimization(PSO)for collaborative search to make up for the poor global optimization ability of particle swarm optimization in the late stage and the slow local convergence of sine cosine algorithm.At each iteration,the best top N2 particles are selected for sine cosine algorithm and particle swarm optimization,respectively,and then the two populations are merged and executed repeatedly,which greatly improves the possibility of finding the global optimal solution.A hybrid sine cosine algorithm and particle swarm optimization algorithm(HSCAPSO)was used to optimize SVM model parameters and predict MCI patients on the ADNI dataset.The experimental results show that the prediction results of the sMCI and cMCI groups have obtained an accuracy of 83.61% and an AUC value of 0.8695.By comparing with the basic SVM,LapSVM,LinSVM,CV-SVM,and PSO-SVM,it is found that the method in this paper achieved excellent performance in various performance indicators and improved the classification performance of SVM in MCI conversion prediction. |
