Recognition Of Mild Cognition Impairment Based On Dictionary Pair Learning

Mild cognitive impairment(MCI)is an excessive phase of normal aging and Alzheimer’s disease(AD),and MCI has a high risk of being converted to Alzheimer’s disease(AD).With the advancement of aging,the prevalence of Alzheimer’s disease has gradually increased,but it is incurable at this stage.Therefore,MCI,as a high-risk transformation stage of Alzheimer disease,has become the focus of AD prevention and Research hotspots.This paper mainly introduces the dictionary learning to introduce mild cognitive impairment recognition,analyzes classical dictionary learning algorithm,finds its advantages and disadvantages,uses the sparsity of coding coefficients,and introduces graph regularization and semi-supervised learning.The supervised and semi-supervised graph regularization dictionary pair learning algorithms identifies and classifies of mild cognitive impairment.The specific research work is as follows:1.For image classification,it is not guaranteed that all the features are conducive to classification,there may be some unrelated features,resulting in the problem of overfitting imagination.Through the analysis of the dictionary learning algorithm,we found that the sparsity of coding coefficients can reduce the irrelevant features.The impact of the dictionary learning algorithm applied to identify the status of mild cognitive impairment was compared.2.For the dictionary learning algorithm,only the reconstruction error and the sparseness of coefficients on different classes are considered,and the problem of neighboring geometric topological relations of the image is not considered.This paper proposes a graph regularization dictionary pair learning algorithm.The graph regularization terms are used to maintain the geometric topological relations of the images and increase the distinguishability between the samples of different classes.In the MRI images on the ADNI1 dataset,the MCI was subdivided into Progressive MCI(PMCI)and Stable MCI(SMCI).Experiments show that the graph regularization dictionary pair learning has a better prediction than the learning algorithm for mild cognitive impairment.Directly using the extracted features plus classifiers works well.3.For the problem that there are few samples of MRI images marked by doctors and there are many unlabeled samples,the missing information of the samples cannot be marked.Based on the regularized dictionary learning algorithm,this paper proposes asemi-supervised graph regularized dictionary pair learning.The algorithm will be applied to the recognition of mild cognitive impairment.In order to make full use of the existing image resources,the unlabeled samples are added to the tag samples for dictionary learning,and unlabeled samples are tagged by the probability of reconstructing the errors on the sum of all class reconstruction errors.On the ADNI1 dataset,PMCI and SMCI classifications,and ADNI datasets downloaded in addition,were tested on the three groups of AD and NL,AD and MCI,MCI and NL to verify the validity and feasibility of the algorithm.