Tensor-Based Multi-Modal Multi-Target Regression for Alzheimer’s Disease Prediction

The assessment of Alzheimer’s Disease (AD) progression via the analysis of physical changes within the brain has attracted great interest from the fields of healthcare, computational medicine, and machine learning alike. Recent studies have demonstrated that using both multi-modal data and multiple AD assessment scores in a predictive model can better reflect pathological characteristics and enhance prediction performance. However, using such high-dimensional structure information to model inter-correlation between multiple targets remains a challenging task. In this paper, we propose a Tensor-based Multi-modal Multi-Target Regression (TMMTR) method for AD detection and prediction, which enables simultaneously modeling multilinear structure information as well as intrinsic inter-target correlations in a general learning framework. We also investigate the tensor-structured sparsity that supports the interpretability of our prediction. Experiments conducted on the ADNI dataset validate the superior performance of our method when compared to other state-of-the-art methods.