Integrating Static and Dynamic Data for Improved Prediction of Cognitive Declines Using Augmented Genotype-Phenotype Representations

Hoon Seo, Lodewijk Brand, Hua Wang

AAAI - 2021

Alzheimer's Disease (AD) is a chronic neurodegenerative disease that severely causes problems on patients' thinking, memory, and behavior. As an early diagnosis is important to prevent AD progression, recent algorithmic approaches have been proposed to predict cognitive decline. However, these models often fail to integrate heterogeneous genetic and neuroimaging biomarkers to improve diagnosis prediction and are not able to handle incomplete longitudinal data. In this work we propose a novel objective function to identify cognitive decline related to AD. Our approach is designed to incorporate dynamic neuroimaging data by way of a patient-specific augmentation combined with multimodal data integration aligned via a regression task. Furthermore, our approach, in order to incorporate additional side-information, utilizes structured regularization techniques popularized in recent AD literature. Finally, in order to reduce the impact of outliers, we propose an iterative method with tunable $p$-order norm variants of the side-information regularization terms. Our approach, can be used to learn a fixed-length vector representation from multimodal dynamic and static modalities which can be used in conventional machine learning algorithms. Our experimental results show that the proposed augmentation model can improve the prediction performance on the cognitive score prediction task for a collection of popular machine learning approaches. In addition, the results of our approach are interpreted to validate existing genetic and neuroimaging biomarkers that have been shown to be predictive of cognitive decline.