You need to be screened for a disease – would you rather have your picture taken or undergo extensive, expensive and invasive genetic tests?
It sounds too good to be true, but it’s a real question thanks to advances in artificial intelligence and the work of Wael Abd Almagedresearch leader at USC’s Information Sciences Institute (ISI), who is using AI and facial recognition analysis to accurately predict congenital adrenal hyperplasia, a disease that causes mild facial changes.
This is just one example of the work being done by the researchers at ISI, who are joining forces to form the Center for Artificial Intelligence Research for Health (AI4Health).
The centre, headed by director Michael Pazzaniprincipal scientist at ISI, will focus on research that enables breakthroughs in ethical artificial intelligence algorithms and systems to improve healthcare, fight misinformation and analyze big data.
Finding the intersection of AI and medicine
Pazzani said: “ISI has already used AI for health research, one of the goals of AI4Health is to make it more systematic to make it easier for medical school researchers to find people with expertise in AI.”
With that goal in mind, AI4Health will hold a number of events in collaboration with Keck School of Medicine at USC. The first event is set for Thursday, December 1, 2022, at 11 a.m. to 1 p.m. on the USC Health Sciences Campus. At this event, six researchers from ISI and six researchers from Keck will each give a five-minute talk about their work. Pazzani explained that these events will seek to “find intersections between Keck and ISI and increase the number of collaborations.” Register at AI4Health.isi.edu.
“Health data has become much more abundant in recent years,” Pazzani said. Electronic health records, genomic data, information from sensors and wearables, and medical images – all these data are ripe for analysis by AI. Information can also be obtained from scientific journal publications and social media, both of which continue to increase rapidly in volume.
And this level of big data is where AI and machine learning work best: looking for patterns in the data, extracting information from text (ie journals and social media) and making predictions based on data analysis.
AI4Health will use AI to exploit the increasing amounts of health data, as well as find solutions to the challenges that come with big data.
AI4Health research areas
For data to be useful, researchers must be able to find it; it is useful if it is composed, organized and annotated; and it must be available or distributed to interested parties. Making all that happen is what is known as data managementand several ISI researchers have been active in this area as it applies to health.
Carl Kesselman, ISI Fellow ad director of the Informatics Systems Research Division, created the pipelines and workflows that enable FaceBase 3 Data Management and Integration Hub to collect and curate huge datasets on craniofacial and dental development in humans and animal models. All of which are available to the wider craniofacial research community with the aim of promote research into craniofacial development and malformations.
Yigal ArensThe ISI Senior Administrative Director and Interim AI Division Director and his team have worked for years with the National Institutes of Health and the National Institute of Mental Health to create NIMH Repository and Genomic Resource (NRGR). NRGR is a collection of biospecimens and data from people diagnosed with mental health problems and their relatives. Datasets from the repository are made available to researchers with the aim of stimulating research and development by providing timely access to primary data and biomaterials.
Important work like this—work that facilitates the use of the abundance of health data out there—will continue as part of AI4Health.
Knowledge discovery and data analysis
Thanks to the abundance of health data, researchers are able to use artificial intelligence to tease out the patterns that can lead to breakthroughs. This often means analyzing electronic health records, medical images or data from wearable sensors to discover new relationships.
How does it look in practice? The work of ISI senior research manager Greg Ver Steeg who have found predictive factors for Alzheimer’s disease among the patient’s medical data.
Or ISI research manager Abigail Horn‘s work to understand behaviors that lead to diet-related diseases. Horn has tied enormous amounts of mobile phone mobility data and health data to show that the food environment is highly correlated with diet-related diseases. The research also analyzes digital restaurant menus to determine the quality of food available to local communities, hopefully occupancy the way for more effective public health policies or interventions for demographic groups affected by poor diet.
But there are some health data that may not seem like “health data” at first glance. Posts on social media, e.g. Emile Ferrara, head of the ISI research team, has worked to counter social media manipulation and misinformation on a range of public health issues: COVID-19 conspiracies; anti-vax campaigns; promotion of tobacco; and the conflation of politics and public health policies online.
Another dataset ripe for knowledge discovery and analysis is the ever-growing volume of electronic journal publications. With AI, these can be analyzed to create databases of health information.
“Knowledge discovery refers to the research of how to use machine learning to find patterns in the data,” said Pazzani, who followed up by explaining that precision health refers to “finding the disease risks and treatments that will work best for each person.”
A priority for Keck School of Medicine at USC, precision health uses the identification of genomic data or other factors to improve the health of a subset of the population. This could mean tailoring treatments to a group of patients, looking at a virus with a specific genome and much more.
Pazzani gave an example: “There are a number of drugs for Parkinson’s disease that are unfortunately only about 25 percent effective, but for a certain group of patients they are 90 percent effective.”
This is where AI comes into play. He continued: “So if you can understand the relationship between a patient’s genetic background and the drug, then you can tailor a drug to a specific patient or a specific group of patients.”
And this type of analysis can have tangible consequences: “It is difficult to get something that is 25 percent effective approved by the FDA. Getting something approved that’s 90 percent effective for people with a particular genome is a lot easier.”
Machine Learning for Health
AI and machine learning (ML) can also be used for clinical decision making by suggesting diagnoses or recommending interventions to clinicians. AbdAlmageed’s work using facial recognition analysis predicting congenital adrenal hyperplasia is one example, and many researchers at ISI are already heavily involved in this field.
Pazzani, who has an extensive background in machine learning, has worked on using ML to detect cognitive impairment, recommend treatment for HIV patients, analyze chest X-rays, diagnose glaucoma and more. AI4Health researchers, including Pazzani, will continue their work on ML for health while looking for new opportunities and applications for ML in healthcare, with the goal of creating both better patient experiences and improved health outcomes.
Improving the patient experience can also be done through telehealth using AI systems to assist in remote healthcare. AI can analyze chat text, voice and images to provide quick feedback to clinicians or patients. Again, the analysis of these facial changes by AbdAlmageed is a good example of this, but there is a wide range of ways telehealth can be improved with AI.
Pazzani said: “This could be a chat doctor visit to decide what type of doctor to see. Or maybe we can provide reassurance and say ‘take two aspirin and call me in the morning’ for some people. And others, we can see that it’s an emergency and we’ll get them to the right specialist.”
Catalyzing research and looking for breakthroughs
More than a dozen researchers already working on AI research as it applies to health will join the AI4Health initiative. Alongside Pazzani as director, the center will be led by ISI’s Wael Abd Almaged, Jose-Luis Ambite, Abigail Horn and Greg Ver Steeg as co-directors. This team along with researchers from ISI and across USC will work to catalyze research, look for breakthroughs and most importantly work to improve health outcomes for patients.
Published on November 22, 2022
Last updated on November 22, 2022