AI and your health
Researchers are using artificial intelligence technology to improve treatment outcomes and population health
By Jill Pease
Developing precision doses for a non-pharmaceutical treatment for brain health. Real-time surveillance of antimicrobial resistance. Predicting the effectiveness of new cancer nanomedicines. Forecasting hotspots of infectious disease transmission. These are just a few of the projects by College of Public Health and Health Professions researchers that use artificial intelligence technology in studies designed to improve treatment interventions and population health.
PHHP scientists join faculty across campus in the University of Florida’s drive to become a national leader in AI. UF’s AI initiative is fueled by a $100 million public-private partnership with technology corporation NVIDIA that has enhanced the capabilities of UF’s HiPerGator, one of higher education’s most powerful supercomputers. The university is also hiring 100 new faculty members with expertise in AI. The goal is to address some of the world’s greatest challenges, create access to AI training and tools for underrepresented communities, and build momentum for transforming the future of the workforce.
AI technology is already being used in health care to reduce medical errors, develop new medications, improve health system efficiencies and patient experience, diagnose illness through the analysis of medical imaging scans, and guide robots used in minimally invasive procedures. In public health, AI is being applied to disease surveillance and health promotion interventions.
But for all that feels new about AI, Mattia Prosperi, Ph.D., a professor in the UF department of epidemiology and PHHP’s coordinator of artificial intelligence, thinks of it as a continuum of the work scientists have been doing for decades.
“A neural network (a form of machine learning) is nothing but an advanced statistical model that is able to analyze very, very complex scenarios that 20 years ago with a single statistical model you could not,” Prosperi said. “Not because you couldn’t imagine it, but because you didn’t have the computational power to calculate all the possibilities. From a theoretical point of view, a ‘deep’ neural network is a collection of basic statistical models we have been using forever. It’s just that instead of having one of them, you have a thousand of them and they’re all talking to each other.”
Solving the problem of the right dose
PHHP researchers studying the use of a noninvasive brain stimulation treatment paired with cognitive training have found the therapy holds promise as an effective, drug-free approach for someday warding off Alzheimer’s disease and other dementias.
Yet determining optimal dosing for the treatment known as transcranial direct current stimulation, or tDCS, which is delivered by a safe and weak electrical current passed through electrodes placed on a person’s head, has been a challenge because of individual differences in anatomy.
“In the field of tDCS, a fixed dosing approach is the standard convention,” said Adam Woods, Ph.D., an associate professor of clinical and health psychology and associate director of the Center for Cognitive Aging and Memory at UF’s Evelyn F. and William L. McKnight Brain Institute. “All people receive exactly the same dose of tDCS, even though we know that individual differences in head and brain anatomy significantly alter how much current enters the brain and where it goes.”
Woods and his colleagues, including fellow principal investigator Ruogu Fang, Ph.D., an assistant professor of biomedical engineering at the UF College of Engineering, have received a $2.9 million grant from the National Institute on Aging to use AI technology to evaluate 16 million data points captured from research participants. The aim is to develop a precision dose for each individual who receives the treatment.
“AI methods allow analyses across multiple data types and sources. For example, analyzing multi-modal neuroimaging data such as MRI and fMRI data, and combining them with behavioral outcome data,” said team member Aprinda Indahlastari, Ph.D., a research assistant professor of clinical and health psychology. “Simply put, AI methods allow us to analyze big data sourced from different types of acquisition methods and enable us to find important characteristics across data sources that are unique to each person.”
Improving cancer nanomedicine outcomes
Nanomedicines may offer clinicians a way to deliver precise, targeted therapy directly to tumors without damaging surrounding tissue. But progress in the development of new drugs that treat cancer at the nanoparticle level has been frustratingly slow. Good results in animal models haven’t necessarily translated to clinical success in humans, in part because of low delivery efficiency of nanoparticles to tumors.
With the support of a new $1.3 million grant from the National Institute of Biomedical Imaging and Bioengineering, researchers led by Zhoumeng Lin, B.Med., Ph.D., DABT, CPH, are building a tool that can offer drug researchers insight into how well a new nanoparticle-based cancer therapy will work, even before a drug enters animal testing.
“This project will provide a tangible tool to improve the design of nanoparticles to accelerate clinical translation of cancer nanomedicines from animals to humans in order to benefit cancer patients,” said Lin, an associate professor of environmental and global health and a member of UF’s Center for Environmental and Human Toxicology and Center for Pharmacometrics and Systems Pharmacology.
To build their predictive model, researchers will use an artificial neural network and train it with hundreds of data sets from physiologically-based pharmacokinetic (PBPK) computer models. PBPK models describe the absorption, distribution, metabolism and excretion of a drug in the body using mathematical equations, and they can be used to predict the concentration of a drug following different therapies. Pharmacokinetic lab experiments using nanoparticles will be carried out to validate and/or optimize the new AI-PBPK model.
For the project’s final outcome, the team will convert the smart model into a publicly available web-based interface for use by nanomedicine researchers.
Using AI ethically
Prosperi and members of his Data Intelligence Systems Lab are applying AI technology to numerous studies, including real-time surveillance of antimicrobial resistance, tracking epidemics using molecular sequencing, and predicting where future HIV/AIDS hot spots may emerge , a study that is supported by a $2.8 million grant from the National Institutes of Health.
But the work he is most enthusiastic about is examining how bias can be removed from AI models so that findings are accurate, fair and do not cause unintended harm. Well-known examples exist of AI models producing findings that could lead to health disparities because the model was trained on biased data, Prosperi said. These could include an AI diagnostic tool for detecting melanomas that was only trained with images of lesions on lighter skin.
“AI systems can beat us at chess and video games, but when you look at some of the precision medicine models we’ve had so far that were AI powered, they’re struggling,” Prosperi said. “They give predictions that sometimes completely deviate from what one would expect. One of the reasons behind it is that models using biomedical data often use data that are not randomized. And therefore there is bias and bias will affect the learning of the model as well as its predictions.”
Prosperi and a growing number of experts argue that causal inference — an approach used for many years in epidemiology and statistics fields to help researchers consider the cause and effect of variables upon outcomes — could help AI models overcome blind spots that lead to bias.
“By fusing causal inference with AI we can get the best of both worlds, that is a strong theoretical approach for bias and a strong computational approach for answering research questions that sometimes you can’t answer with traditional statistics,” Prosperi said.
But bias may not only exist in the data, Prosperi said. Traditional gender and racial imbalances among scientists in computing and other fields who use AI tools may contribute, along with limited training on ethical use of health data available to researchers outside the health sciences.
AI across the curriculum
In addition to supporting cutting-edge AI research and building critical research infrastructure, UF is providing opportunities for all undergraduate students, regardless of their major, to learn about AI and how it impacts their fields of study.
At PHHP, three courses are being finalized that can be taken together as a certificate for the college’s undergraduate students. These classes will provide training on AI in health care and public health, ethics in AI, and data visualization in the health sciences. A certificate for the college’s graduate students that will help them apply AI tools to their research questions, is currently under development.
AI is already prevalent in health care and population health and it is important that students be prepared to use them effectively in their future careers, whether that is making treatment decisions for a patient or predicting the next pandemic, said George Hack, Ph.D., the college’s associate dean for educational affairs.
“Our students will need to understand how these tools work, where there are ethical concerns in their use, and understand how these tools generate findings and predictions,” Hack said. “That way they can assess what AI data is valuable for them to use as evidence in their decision-making, whether that be for the health of the patient or interventions in a population.”
Meet PHHP's newest AI hires
Aprinda Indahlastari, Ph.D., currently serves as a research assistant professor of clinical and health psychology. Her broader research interests are in optimizing and personalizing existing medical devices through the use of computational modeling, such as machine learning and finite element methods, with the goal of achieving precision medicine that is tailored to each person. Read more about her work in this issue.
“AI methods have big potential to help researchers and clinicians by identifying the key factors that drive age-related cognitive decline in older adults and finding an effective way to mitigate these factors,” she said. “AI methods can also help recognize important characteristics of successful interventions that can be used to optimize future interventions aimed at improving cognitive aging.”
Panayiotis (Takis) Benos, Ph.D. will join the department of epidemiology from the University of Pittsburgh. His group works on the intersection of machine learning, computational biology and systems medicine. The ultimate goal of the group is to identify risk factors and mechanisms affecting aging and contributing to the onset and progression of chronic diseases and cancer. They develop and use probabilistic graphical models and other machine learning methods to integrate and mine high-dimensional, multi-modal biomedical data and to investigate biological processes pertinent to health and disease. The disease focus of the lab includes chronic obstructive pulmonary disease, idiopathic pulmonary fibrosis, cardiovascular diseases and alcoholic hepatitis. Other ongoing projects are related to the identification of microbiome contributions to clinical outcomes in critically ill patients and the understanding of the mechanisms of cancer immunoprevention.
Joseph Gullett, Ph.D., is a research assistant professor in the department of clinical and health psychology. His research is focused on the use of machine learning methods to predict intervention outcomes and disease progression in older adults with mild cognitive impairment, and the relationship of white matter microstructure with clinical disorders and their associated neuropsychological function.
Read more about Dr. Gullett’s recent study examining AI’s potential to predict dementia.
Noah Hammarlund, Ph.D., joins the department of health services research, management and policy from the University of Washington. In his research, he merges health economics with innovations in artificial intelligence to investigate the role of social factors in the delivery of healthcare with the goal to better target policy solutions to disparities in health.
Muxuan Liang, Ph.D., will join the department of biostatistics next year from the Fred Hutchinson Cancer Research Center. In his research, he applies statistical and machine learning techniques to large databases like electronic health records, to help health care providers make decisions based on patient-level information. These may include decisions about treatment, tailored cancer surveillance strategy and individualized risk prediction.
“In many heterogeneous diseases, including cancer, it is uncommon to expect the same treatment effect for a single medication in each individual,” Liang said. “AI technology, such as deep neural networks, provides a great way to characterize the possible heterogeneity of the disease, accommodate interactions between multiple medications, and inform the practical use of medications and other interventions.”
Zhoumeng Lin, B.Med., Ph.D., DABT, CPH, joined the university last summer from Kansas State University as the first faculty member in PHHP hired under UF’s AI initiative. His research focuses on the development and application of computational technologies to address research questions related to nanomedicine, animal-derived food safety assessment, and environmental chemical risk assessment.
“The long-term goal is to develop AI-assisted computational approaches to support decision-making in human, animal and environmental health,” Lin said.
Read more about Lin’s latest project in this issue.
Feifei Xiao, Ph.D., arrives at UF next year from the University of South Carolina. She focuses on the development and application of powerful and efficient statistical methods for high throughput genetics and genomics data. Her work includes ongoing projects in cancer, aging and other public health related outcomes, with the goal of providing efficient statistical tools to integrate genetic and genomic data into the practice of precision medicine.
“Everyone has seen UF’s growth over the past years,” Xiao said. “UF Health is currently ranked as the top health center in Florida and one of the top hospitals in the U.S. I see so many collaboration opportunities in big data health sciences, including biomedical data, health electronic records data, and more. These resources provide a wonderful platform for me to conduct independent and collaborative research and to further establish my research team.”
