David Samuel Smith - US grants

DESCRIPTION (provided by applicant): The goal of this proposal is to develop and validate novel compressed sensing (CS) approaches to dramatically improve the spatial and temporal resolution of quantitative dynamic contrast enhanced magnetic resonance imaging (DCE-MRI). CS exploits prior information (assumptions) about MR images to infer missing data and produce high-quality images from significantly less data than previously thought possible. CS has already proven extremely successful in MR angiography and cardiac MRI, where it has accelerated some acquisitions by up to 10- to 100-fold, but is relatively unexplored in cancer imaging. DCE-MRI involves the serial acquisition of heavily T1-weighted images before and after the injection of a contrast agent to increase water relaxation rates in tissues. The resultin data can then be analyzed with appropriate pharmacokinetic models to extract quantitative parameters reporting on, for example, vessel perfusion and permeability, and tissue volume fractions. DCE-MRI has been applied to predict the early response to neoadjuvant chemotherapy in breast cancer, but the technique is not yet robust and accurate enough for the clinic. A fundamental practical limitation of DCE-MRI is the necessity to simultaneously acquire high temporal resolution, to adequately sample the contrast time course, and high spatial resolution, which is required for clinical morphological assessment and accurate tumor delineation. In traditional Cartesian MRI acquisitions, one must choose between high spatial or high temporal resolution before the scan. With a golden ratio acquisition, the tradeoff between spatial and temporal resolution is eliminated. A single DCE-MRI scan may then be used for both accurate kinetic modeling by slicing the data at high temporal cadence, while also allowing a high spatial resolution image to be formed by taking the data as a whole. Thus, a golden ratio acquisition coupled with CS has great potential to enable a clinically relevant DCE-MRI protocol that provides adequate temporal resolution for kinetic modeling without sacrificing the spatial resolution required for morphological evaluation. This project has three aims: (1) to develop a compressed sensing based high temporal resolution protocol for quantitative DCE-MRI, (2) to develop a compressed sensing based high spatial resolution T1-weighted anatomical imaging protocol for morphological evaluation, and (3) to apply the developed CS-based protocols in vivo for validation and evaluation. If this project is successful, it will significantly improve te ability to predict response to neoadjuvant chemotherapy, provide new CS methods for the community to apply to other in vivo applications, and validate CS in an important cancer imaging application.

The NSF Convergence Accelerator supports use-inspired, team-based, multidisciplinary efforts that address challenges of national importance and will produce deliverables of value to society in the near future. Today there is a huge gap between the global need to manage our ecosystems, protect our societies, and discover new therapeutics ? and the global capacity to deliver the data and models needed to solve the most pressing challenges of our time. This project is intended to bridge that gap by connecting researchers, policy makers, and industries to the scalable biome monitoring networks of the future ? by developing the unified biome datasets, cross-cutting models, and policy paradigms that will empower these disciplines to accelerate, innovate, and converge. If successful, this would lead to a fundamental paradigm shift in how disciplines study and manage the planet. It will contribute to the advent of a new generation of scientists developing predictive AI models of the biome, and to developing science-based methods and tools for shaping policies and delivering policy-aware tools to solve societal-scale challenges. We expect that deep monitoring of biome and the new science and technology ecosystem emerging from it will have wide impact on human health, agriculture, national security, and ecology.

The technical goals of this project have been carefully instantiated so that progress towards convergence makes a lasting impact on a range of scientific problems. First, the life sciences, engineering, and policy domains continually face the challenge of managing and unifying disparate biome and ecological datasets. These issues are addressed head on by bringing together uniquely deep and state-of-the-art biome and ecological data sets, identifying the hard unification problems, and providing a reference solution to unification. Second, there is a focus is on new unified agent-based models for predicting mosquito populations, as mosquito-borne diseases already account for over 600 million cases of human disease per year, with a disproportionately large impact on disadvantaged communities in sub-Saharan-Africa. By accelerating the development of new predictive mosquito models ? especially by generalizing them to additional species ? this project will provide long lasting contributions to human health and pandemic preparedness. Third, as deep biome data exponentially scales, the life sciences will become overwhelmed with genomic information. Convergence must lead to new methods to efficiently harness these data and autonomously derive insights.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

Individuals, industries, societies, and governments want to stay healthy. They need cost-effective systems to detect biological threats and predict future disease outbreaks as early as possible. COVID-19 acutely and painfully demonstrated the impacts of the unpredicted. The goals of this program, Computing the Biome, are twofold: (1) demonstrate an extensible data and AI platform that continuously monitors and predicts biothreats in a major U.S. city, and (2) create a framework for economic sustainability and global scalability of these results, by empowering businesses and advanced science missions to consume predictions and produce valuable consumer apps and breakthroughs.

This team will produce and interconnect novel data streams ranging from kilometer-scale hyper-local weather, to autonomously identified disease transmitting insects (only millimeters in size), to genomically recognized known and novel viruses (only nanometers in size) – demonstrating that cross-cutting continuous data streams for biothreat detection and prediction can be rapidly unlocked. By combining their expertise in ecology, epidemiology, and virology, the team will design new predictive models and anomaly detectors. This project will develop the first of these high-impact AIs focused on predicting mosquito-borne diseases, which are difficult to control and impact over 600 million people per year. More broadly, the resulting data platform will empower development of new foundational methods for use by the AI community – based on real-world data and grounded in the societal challenges of our age.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.