Group Guided Sparse Group Lasso Multi-task Learning for Cognitive Performance Prediction of Alzheimer's Disease

Alzheimers disease (AD), the most common form of dementia, causes progressive impairment of cognitive functions of patients. There is thus an urgent need to (1) accurately predict the cognitive performance of the disease, and (2) identify potential MRI (Magnetic Resonance Imaging)-related biomarkers most predictive of the estimation of cognitive outcomes. The main objective of this work is to build a multi-task learning based on MRI in the presence of structure in the features. In this paper, we simultaneously exploit the interrelated structures within the MRI features and among the tasks and present a novel Group guided Sparse group lasso (GSGL) regularized multi-task learning approach, to effectively incorporate both the relatedness among multiple cognitive score prediction tasks and useful inherent group structure in features. An Alternating Direction Method of Multipliers (ADMM) based optimization is developed to efficiently solve the non-smooth formulation. We demonstrate the performance of the proposed method using the Alzheimer’s Disease Neuroimaging Initiative (ADNI) datasets and show that our proposed methods achieve not only clearly improved prediction performance for cognitive measurements, but also finds a compact set of highly suggestive biomarkers relevant to AD.