High dimensional classification for spatially dependent data with application to neuroimaging (Q2209817)

In order to cope with brain imaging data (tensor-variate observations), the authors propose the classification of such spatially dependent data. The main result is the classification algorithm using the penalized maximum likelihood estimation based on linear discriminant analysis for high-dimensional data with spatially dependent features under assumption of multivariate normal distribution. The efficiency of the proposed methodology is illustrated on real neuroimaging datasets.