Multi-task Feature Selection Algorithm And Its Application For Multimodal Neuro Image

Alzheimer's Disease is one of the most common types of senile dementia,and potential disease pathology may precede the onset of cognitive symptoms.Although some effective treatments can be used for Alzheimer's Disease to delay its onset cycle,but there is no existence treatment solutions for it,so many countries in the world attach great importance to the research of early diagnosis or treatment programs for Alzheimer's Disease.Multimodal neuro image provides a rich source of data for the study of Alzheimer's Disease,but how to extract specific features from these three-dimensional neuro imag is a fundamental problem of computer-aided decision-making in Alzheimer's Disease.The multimodal multi-task feature selection method studied in this paper is of great importance to the computer aided decision-making of Alzheimer's Disease,and it also has some reference for multimodal data analysis in other fields.In this paper,multi-task feature selection algorithm is used to analyze and study the multimodal image data.The experimental results show that the proposed algorithm can help to find the biological markers which is sensitive to brain diseases,strengthen the understanding of pathology of brain diseases and provide reference to it.The main innovations as follows:1.A density clustering-based multitask feature selection algorithm is proposed.The algorithm uses the similarity matrix based on the density clustering to measure the structural relation after the linear mapping of the samples.That is,the same class samples in the original space after linear mapping are close with each others,different class samples with close distance in the original space after linear mapping will be far,this structural relationship constraints embedded in the multi-task feature selection framework will be achieve the effective selection of features.2.A minimum interclass variance-based multitask feature selection algorithm is proposed.This algorithm is combined intra-class variance minimization with multi-task feature selection algorithm,a interclass variance minimization term is constructed in multitask feature selection model,which will be achieve the effective selection of features.3.We proposed a multi-task feature selection algorithm of two-stage strategy.Firstly,the Laplacian Score feature selection method based on the effective distance is introduced,used to pre-reduce feature dimension of label data,and then the feature after dimension reduction is taken as the input of minimum interclass variance-based multitask feature selection algorithm,achieving the robust select of features.In our experiment,10-fold cross-validation strategy and grid search strategy to achieve optimal parameter in the unsupervised and supervised feature selection method.