Thinking Big
The Precision Health Through Artificial Intelligence Initiative uses leading-edge big-data approaches to advance the frontiers of medicine
When UCI announced its Big Ideas Challenge in February 2018, more than 260 individuals submitted their biggest, boldest and most audacious multidisciplinary ideas that could have a truly transformational impact. That summer, two proposals, both centered on the same concept, were chosen as the joint winner. Their mutual inspiration and aspiration: the integration of artificial intelligence in providing next-level healthcare.
Two years later, that winning initiative, Precision Health Through Artificial Intelligence, is making strides in revolutionizing healthcare delivery and research at UCI. From COVID-19 to stroke care, breast cancer to Huntington’s disease, PHAI’s leadership team is not only creating cutting-edge AI tools but empowering others across specialties to do so as well, expanding the capabilities of medicine for the benefit of patients regionally and globally.
According to IBM Research Healthcare and Life Sciences, the average person will generate more than 1 million gigabytes of health-related data in a lifetime. This colossal compilation of information from electronic medical records, radiological and pathological imaging, fitness trackers, implants, advanced screening techniques and more is the equivalent of 300 million books. And it’s growing at blazing speeds. As of 2020, the amount of health data is estimated to double every 73 days.
How can UCI researchers sift through these vast quantities to extract and develop not just useful but targeted treatments for patients? Enter PHAI, a merging of the minds of two relatively new institutes on campus: the Center for Precision Health and the Center for Artificial Intelligence in Diagnostic Medicine. “PHAI is bringing together the clinical and basic scientists on campus, the experts who understand where the biggest problems and opportunities are, with the machine-learning tools,” says Dr. Peter Chang, a member of PHAI’s leadership team and co-director of CAIDM.
Chang is uniquely qualified to bridge the medical and AI worlds. He’s a radiologist by training but a self-taught software engineer who, throughout medical school, wrote algorithms for machine learning as a hobby. He has created several AI startups and worked closely with such Silicon Valley stalwarts as Amazon and Nvidia. Chang was on the verge of pivoting completely from medicine to industry when UCI Health offered him an opportunity he couldn’t refuse. He would join UCI radiologist Daniel Chow – a longtime friend – in launching CAIDM, a multi-specialty initiative to develop and integrate AI technology across UCI’s healthcare system.
“What makes CAIDM unique is that we’re not focused on just the really cool technology and seeing where we can apply it,” says Chow, CAIDM’s co-director and also a member of the PHAI leadership team. “Instead, we’re asking, ‘What are the clinical challenges?’ and seeing where we can apply our knowledge of AI to fix these specific problems. It’s the opposite of that idiom about too many hammers looking for a nail. We figure out what the nail is and then design an appropriate tool for it.”
Picture(s) of Health
Where machine learning excels – and leaves our mere human minds in the dust – is in finding patterns. By sorting through thousands, if not millions, of data points, AI can detect the finest of distinctions that might influence a patient’s course of treatment, its speed and how well he or she responds.
Some of the earliest AI applications in medicine have been developed in the radiological field, due to the wealth of images and data available. Because Chang and Chow are both radiologists, it’s natural that CAIDM’s initial focus has been on imaging-related research. For example, one AI tool under development would use imaging to more precisely stratify early breast cancers into those that need monitoring versus treatment. Another would evaluate radiological images for renal cell carcinoma, a disease often incidentally detected when an individual has a scan for a different reason.
One of CAIDM’s tools is now helping to save the lives of stroke patients in UCI Medical Center’s emergency department. “Stroke care is very time-sensitive,” Chow says. “You need to administer therapy for strokes within three to four and a half hours of onset. But it can take two to four hours for a radiologist to review a head CT before therapy can begin. The challenge, or nail, that we saw was: How can AI expedite triage of stroke patients?” CAIDM’s tool can analyze a CT scan and detect in about 20 seconds the existence and extent of any cerebral hemorrhages, which directly influences the course of treatment.
Collaborating on COVID-19
One of PHAI’s primary goals is to expand the application of AI across all specialties on the UCI campus. The coronavirus crisis has jump-started that plan, as PHAI leaders collaborate with colleagues in numerous departments – ranging from computer science to pathology – to develop and apply AI tools specifically for COVID-19. “It’s through entities like PHAI that critical connections are made among researchers across disciplines, and it’s exciting to see the administration so supportive and aligned behind this initiative,” says Suzanne Sandmeyer, the Grace Beekhuis Bell Chair in Biological Chemistry and vice dean of research in UCI’s School of Medicine and a member of the PHAI leadership team.
“PHAI is bringing together the
clinical and basic scientists
on campus, the experts who
understand where the biggest
problems and opportunities are,
with the machine-learning tools.”
For the COVID-19 AI tools, Chow and radiologist Jennifer Soun – with the help of a large group of students – are recording data from the electronic medical records of all UCI Health patients with a confirmed diagnosis of COVID-19. Chang is using that data to create predictive models to better estimate the disease’s progression and identify individuals who might need escalated care.
The focus of the project’s second stage is to establish a public web portal so that physicians and caretakers worldwide can employ the models to run risk predictions for their own patients. Those results, in turn, will enter a huge real-time database of COVID-19 patients that can be used to continuously improve the models. “Over time, with enough aggregated data,” Chang says, “the machine-learning algorithm has the potential to uncover disease signatures and personalized recommendations for care.”
Analyzing Biomarkers with AI Tools
Facilitating such personalized care recommendations is another primary goal of PHAI. A key tool in the fast evolving field of precision health is MultiOmyx, a big-data approach that can peer into a single cell and analyze dozens of proteins and DNA biomarkers at once as potential targets for existing or new therapies.
As UCI’s Center for Precision Health started building out the technical infrastructure required for MultiOmyx, its leaders – Sandmeyer and Leslie Thompson, Donald Bren Professor of psychiatry & human behavior as well as neurobiology & behavior – recognized the tremendous synergies between their work and that of CAIDM. Those synergies resulted in their shared win of the Big Ideas Challenge and the creation of PHAI.
“As investigators generate data through MultiOmyx and other methods, PHAI can help them use that data more effectively by applying AI tools,” says Thompson, also a member of the PHAI leadership team. “Those tools can inform the investigator on which samples to select for further screening, such as at UCI’s Genomics High-Throughput Facility.”
An expert on Huntington’s disease, Thompson is putting her own data through the AI paces. “Huntington’s disease is caused by a single genetic mutation, but the symptoms of this incurable neurodegenerative disease can vary across individuals and even within families,” she explains. “If we could identify meaningful biomarkers to stratify patients, that could help identify which patients might, and might not, respond to intervention and help improve the design of clinical trials.”
Chang and Chow are currently using AI to analyze hundreds of CT scans of Huntington’s patients’ heads for distinctive radiological patterns that could indicate the onset and progression of the disease. Once those results are in, Thompson anticipates putting AI to work on her MultiOmyx data.
“These single-cell approaches are so new that we’re just now getting to the point of having enough data to apply AI,” she says. “So even though AI has been around a while, the opportunities for its application in health are increasing. That’s what drives PHAI.”
Indeed, the short list of potential PHAI projects is thrilling. But the one most representative of PHAI’s mission is a crowdsourcing challenge that will invite everyone affliated with UCI to brainstorm how AI can analyze pathology images to detect and classify brain plaques in Alzheimer’s patients.
“PHAI is meant to be a resource for the whole UCI community, to bring together people with any domain expertise and leverage that knowledge together,” Chow says. “Sometimes you don’t know there’s someone else on campus doing the same thing or who has a similar interest as you. Those are the kind of connections that kind of connections that.