The objective of this project is to develop a data-driven smart monitoring methodology of Alzheimer's disease (AD). AD follows an accelerated degradation trajectory as compared to normal aging. Accurate monitoring and prognosis of the disease trajectory is critical for the success of many preventative interventions. Currently, no first-line screening system for monitoring the fast-growing preclinical population is available. While emerging personalized health screening systems provide the infrastructure to routinely screen massive numbers of individuals, it is an essential challenge to transform the role of these systems from passive information collection into smart monitoring to proactively characterize the underlying complex time-varying disease trajectory shaped by an individual's risk factors. This project aims at developing such a "smart monitoring" approach that will equip nowadays cyber infrastructure with powerful data-driven decision-making capabilities for better management of the preclinical individuals, leading to more efficient targeted screening and affordable care, better treatment planning and management, and improved quality of life for both patients and caregivers. Successful implementation will provide a substantial boost for the detection of the 4.5 million preclinical individuals anticipated in the next 20 years. Its generic nature will also impact monitoring of other progressive medical conditions, given the rapid adoption of personalized screening systems in other areas. The interdisciplinary nature of this research across data-driven monitoring, prognostics, optimization, and health care will prepare students a diversified education background. Broader impacts will be also generated through new curriculum modules, online software toolkits for implementation, and involving underrepresented undergraduate and graduate students in research experience programs.
The success of the project will significantly advance the state of the art in data-driven monitoring, prognostics, and selective sensing, and contribute to the science base of the emerging personalized screening systems. Specifically, to model and quantify the disease trajectory, a health index (HI) model will be constructed by synthesizing the degradation information from multiple biomarkers via the development of non-parametric and semi-parametric data fusion schemes. Then, to predict the personalized disease trajectory, personalized prognostics methodologies will be developed that can offline predict and online update the personalized HI model via the development of multi-level degradation models and Bayesian updating approaches. Capitalizing on the personalized prognostics methodologies, selective sensing methodologies will be developed to adaptively identify the screening tests that are most informative for the statistical estimation of the HI via a seamlessly integration of a novel Bayesian network model with robust optimization techniques. A team of five PIs with diverse but complementary research backgrounds will be working closely with two leading AD research institutes in the U.S. to develop, test, and validate the methodologies.
