Research On Weakly Supervised Multi-task Matrix Completion For Prediction Of AD Disease Progression

Alzheimer’s disease is one of the most famous brain diseases.In the medical artificial intelligence field,the prediction of AD disease progression has become a research focus.However,medical imaging data usually has a high dimensionality and a scarce sample size.Scholars from home and abroad have proposed and improved many models to improve the prediction accuracy with the smallsample and high-dimensional medical imaging data,and achieved certain research results.However,some of these models cannot effectively use samples with missing features or labels,or do not consider the complementary information between multi-view data,and cannot achieve better results in practical applications.In the face of the above challenges,this paper introduces the matrix completion theory to model the AD disease progression prediction problem as a multi-task matrix completion model,and propose two models for the two hot research directions of the AD disease progression of brain feature selection and multi-view data fusion.The main research contents and innovations of this article are as follows:1)In view of the small-sample-size problem in the research of AD disease progression,the problem of incomplete score and data noise,and in order to select the features of brain region,a weakly supervised sequential multi-task matrix completion model is proposed.Combined with a novel transductive feature selection strategy,the model can select the task-share features of multiple time points and the task-specific features of different time points at the same time,so as to find the best ROIs most related to AD.In addition,the model can not only effectively use the samples with incomplete score,but also improve the generalization performance of the model by mining the feature distribution of training data and test data at the same time.Meanwhile,regularization term constraint is used to ensure the smoothness of the clinical scores at adjacent time points.Finally,experiments are carried out on a real data set—Alzheimer’s disease neuroimaging Initiative(ADNI),and the results show that the model has good performance.2)In view of the problem of incomplete view in the research of AD disease process,in order to fully mine the complementary information of multi-view data,this paper proposes a weak supervised multi-task matrix completion model based on incomplete multi-view data.The model introduces the theory of latent space representation,and assumes that there is a latent representation among multiple view data,it describes the data and reveals the common potential structural information among different view data,and fully utilizes the complementary information of multi-view data.Morever,l2,1 regularization term is used to denoise outliers,and time series smoothing regularization constraint is used to prevent large deviation between predicted clinical scores at adjacent time points.Finally,experiments on real data set ADNI show that the model has good performance.