Multiple-Response Prediction and Region Selection in Neuroimaging Analysis

• Activity Number: 360
• Type: Contributed
• Date/Time: Tuesday, August 11, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Imaging
• Author(s): Zhou Li* and Lexin Li
• Companies: North Carolina State University and UC Berkeley
• Keywords: brain imaging ; Alzheimer's disease ; multiple responses ; tensor regression ; multidimensional array ; Magnetic Resonance Imaging

Abstract

Many techniques have been developed for predicting continuous clinical variables (responses) related to Alzheimer's Disease (AD) from brain images. However, most of them do not consider the inherent correlations among different clinical variables. A few recently proposed methods take into consideration the intrinsic correlations between multiple responses, but use pre-selected brain regions as predictors and treat them as a vector. Such methods require prior knowledge and ignore the structural information of the imaging data as multidimensional arrays (tensors). In this paper, we propose a sparse multivariate tensor regression model which provides simultaneous prediction of multiple responses and selection of important brain regions related to AD using brain imaging data as predictors directly. The region selection is achieved by using a group-lasso type penalty on the coefficients. We develop a numerical algorithm to solve the penalized regression problem. In our simulation study and real data analysis with Magnetic Resonance Imaging (MRI) data, the proposed method is compared with several existing methods and exhibits competitive performance in prediction and region selection.