Research On Regression Analysis Of Clinical Variable Based On Functional Brain Images

In recent years,with the development of the global "brain project",the analysis and research of brain imaging techniques has attracted much attention to study brain disease.The regression of clini-cal variable based on functional brain image has become research hot spots.Functional brain imaging reflects the neural activity of the brain in a noninvasive manner,and clinical variable values can quan-titatively assess the stage at which a patient's brain disease is located.Through regression analysis,the correlation between brain activity and brain disease can be found,and the current stage of brain disease in the subject can be predicted to assist the doctor in diagnosis.Based on the latest research progress in machine learning,this dissertation is based on two common analytical data of functional brain imaging:brain function network and amplitude of low frequency fluctuation.The main work and innovation are as follows:Firstly,the current regression analysis of brain imaging is based on structural brain imaging,and many kinds of brain diseases(such as Alzheimer's disease)are caused by impaired brain function.The regression analysis based on functional brain imaging has great research significance.Therefore,a novel brain connectivity network based method is proposed to estimate the value of brain disease clinical variables.First of all,the functional connectivity network is extracted frorm the brain images.Then LASSO is adopted for feature selection eliminating redundant features and the clustering coefficients and edge weights of the network are fused as features.Finally,support vector regression machine(SVR)is involved to predict the value of the clinical variables.The proposed method is validated on ADNI dataset and the experimental results show that the proposed method can accurately predict the value of disease clinical variables and verify the effectiveness of the fusion of multiple features.Secondly,the brain network reflects the entire brain area rather than the anomalous area of the disease.Voxel-based methods(such as amplitude of low frequency fluctuation)can reflect the activity of the local brain.To help research the potential of brain diseases pathology.However,voxel-based data is usually high-order data,in order to solve the analysis problem of high-order brain image data,this paper proposes a tensor-based regularized multilinear regression algorithm,named multilinear LASSO(mLASSO),by reformulating the LASSO algorithm for tensor space.The proposed algorithm firstly decomposes tensor space to vector space by applying mode production and employing weighted vectors.Then,our algorithm iteratively uses LASSO to update the weighted vectors for converging the proposed model.Finally,the optimum weighted vectors are applied to all direction in the tensor space in order to generate the regression model.The contribution of this paper is twofold:1)The algorithm employs the whole of the structural information of the dataset for generating a regression model.2)Since the proposed method employs LASSO,it can significantly improve the performance of the generated model by using embedded feature selection.The proposed method is validated on real 3D image data and the experimental studies confirm that the proposed method achieves satisfactory performance on the multilinear data.