Join and Deep Ensemble Regression of Clinical Scores for Alzheimer's Disease Using Longitudinal and Incomplete Data

Alzheimer's disease (AD) is a neurodegenerative disease with an irreversible and progressive process. Close monitoring of AD is essential for making adjustments in the treatment plan. Since clinical scores can indicate the disease status effectively, the prediction of the scores based on the magnetic resonance imaging (MRI data is highly desirable. Different from previous studies at a single time point, we propose to build a model to explore the relationship between MRI data and scores, thereby predicting longitudinal scores at future time points from the corresponding MRI data. The model incorporates three parts, correntropy regularized joint learning-based feature selection, deep polynomial network based feature encoding, and finally, support vector regression. The regression process is carried out for two scenarios. One is to use baseline data for predictions at future time points, and the other is to combine all the previous data for the prediction at the next time point. Meanwhile, the missing scores are filled in the second scenario to address the incompleteness presented in the data. The simulation results demonstrate that the proposed model accurately describes the relationship between MRI data and scores, and thus it can be effective in predicting longitudinal scores.