Dynamic Systems Approach to Deep Learning with Different Types of Data Sets and Its Application to Prediction of Alzheimer’s Disease

• Activity Number: 328 - Integrative Approaches for Statistical Analysis of Data from Multiple Sources
• Type: Topic Contributed
• Date/Time: Tuesday, July 30, 2019 : 10:30 AM to 12:20 PM
• Sponsor: ENAR
• Author(s): Momiao Xiong* and Helen Engle and Yuanyuan Liu and Zhouxuan Li and Qiyang Ge and Shudi Li and Shan Liu
• Companies: University of Texas School of Public Health and University of Texas School of Public Health and University of Texas School of Public Health and University of Texas School of Public Health and University of Texas School of Public Health and University of Texas School of Public Health and University of Texas School of Public Health
• Keywords: Omics; MRI; Deep learning; dynamic systems; big data; Alzheimer’s disease

Abstract

Multiple omics and Imaging datasets provide complementary spatio-temporal information about phenotypes and disease. A novel deep convolutional and recurrent neural network that uses both cross-sectional and longitudinal omics and imaging dataset will be developed for prediction of phenotypes and disease to improve accuracy. A novel of dynamic systems approach to deep learning with different types of datasets in which estimation of parameters in neural networks with multiple datasets and disease prediction will be mathematically formulated as a nonlinear optimal control problem will be proposed. Pontryagin’s maximum principle will be used to derive a set of neural ordinary differential equations. A number of numerical methods for solving neural ordinary differential equations will be explored. The proposed methods will be applied to prediction of Alzheimer’s disease (AD) in Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset. The preliminary results are very encouraging.