Research On Alzheimer's Disease Classification Algorithm Based On Sparse Logistic Regression

Alzheimer's Disease(AD)is an irreversible neurodegenerative disease.The clinical symptoms of this disease include the loss of cognitive and memory,which severely affects people's daily life.Until now,there exists no effective treatment for the disease.Mild cognitive impairment(MCI)is the early stage of AD.Compared with normal cognition,MCI has a higher probability to be converted into AD.Therefore,accurate identification of MCI patients with high conversion risk is of great significance in the early prevention of AD and can delay the deterioration of the disease.In recent years,with the development of neuroimaging technology,many researchers adopted machine learning algorithms to analyze medical image data to assist the diagnosis of AD.Therefore,based on s MRI data,in this paper,an AD classification algorithm on the basis of sparse logistic regression was proposed.The specific works are as follows:Firstly,an AD classification algorithm based on L_1/2 regularized sparse logistic regression was proposed.L_1/2 regularization makes the model have good sparseness and select important brain regions for classification,so as to deal with the characteristics of high-dimensional small samples of neuroimaging data,avoid overfitting and improve the computational efficiency of the model.What's more,it makes the model have high computational efficiency.The experiment was based on the s MRI of 197 subjects in the ADNI database.The gray matter volume of 90regions of interest was extracted from the s MRI of each subject as classification markers.The experimental results indicated that the proposed method not only improves the classification performance of AD and MCI,but also finds brain areas related to AD,thus providing certain neurobiological support for the model.Secondly,an AD classification algorithm based on generalized elastic net regularized sparse logistic regression was proposed.The generalized elastic net is composed of L_p regularization and L₂ regularization.Under the optimal parameters,L_p regularization can select different p values according to different classification tasks,so as to make the model produce desired sparseness.L₂regularization can enable brain regions with high correlation being selected or deleted at the same time.In addition,L₂ regularization can ensure that the model is as simple as possible,thereby improving the generalization performance of the model.Experimental results proved that the proposed method achieves superior classification performance in AD and MCI classification.In particular,our method identifies brain regions related to MCI conversion,which improves the accuracy of the model in MCI conversion prediction.Finally,an AD classification algorithm based on two-stage sparse logistic regression was proposed.In neuroimaging high-dimensional data,not all features can achieve good classification performance.The features with the strongest discrimination ability play an important role in improving the classification performance of AD.Therefore,we proposed a two-stage sparse logistic regression model to select disease-related biomarkers for the improvement of the classification performance of AD.The two-stage sparse logistic regression model combines particle swarm optimization algorithm and adaptive LASSO logistic regression together.Specifically,in the first stage,the particle swarm optimization algorithm was used for global search to remove redundant and irrelevant features,which can reduce the computational time for the later stage.In the second stage,adaptive LASSO performed a local search among the features selected in the first stage,and selected the optimal feature subset for classification.The experimental results displayed that the proposed method selects the discriminative brain regions related to the disease.Meanwhile,compared with other advanced classification methods in recent years,this method can significantly improve the classification performance of AD.