UI group wins $1 million to work on medical artificial intelligence
A team of researchers at the University of Iowa are leading a multi-university project to work on the advancements of medical artificial intelligence.
Entrance to the University of Iowa’s Seamans Center for the Engineering Arts and Sciences at 103 South Capitol Street, Iowa City, IA on Friday, Oct. 2nd, 2020. A one million dollar grant from the National Science Foundation was rewarded to a University of Iowa collaboration between engineering and medical students and faculty to work on advancing medical AI.
October 5, 2020
As Computers and Artificial Intelligence (AI) play a key role in improving medical fields, experts in medical and engineering research at the University of Iowa are merging disciplines to work towards the advancement of medical AI with the help of a $1 million grant from the National Science Foundation.
Assistant Professor of Industrial and Systems Engineering Stephen Baek, the main person behind this research, said he is working alongside a large team of medical and engineering experts at the UI and across the world.
Baek said this research will also be sent to medical institutions at other universities to form a network to test the model, including Harvard, Yale, Stanford, the University of Chicago, and Seoul National University in South Korea.
“My research is basically about creating an informatics system that can support human experts making more informed decisions. So I believe that artificial intelligence agents can help human experts make better decisions” Baek said. “Treating a cancer patient, for example, is a highly demanding job. The doctors, physicians must understand all the charts, images, and different information, and then have to finally reach to a conclusion in terms of how you’re going to treat a patient.”
Baek said his hypothesis is that AI algorithms should be able to support physicians in making challenging decisions in a way that could revolutionize the medical field.
“In medicine, you want to have a bullet proof solution,” he said. “99 percent accuracy is not enough, because there’s a 1 percent chance you mess up with a patient, which means the patient might die. So we want to make sure everything is perfect and reliable.”
The main way they can assure everything is reliable is to collect a lot of data, Baek said, as AI systems are hungry for data to work properly. This data needs to also be accurate for all people, he said, which can be a problem if a particular hospital does not see a lot of diversity, as different hospitals have different demographics and populations.
UI Professor of Electrical and Computer Engineering and Radiation Oncology Xiaodong Wu said huge data sets are needed to create effective medical AI models.
“Currently, in this era of precision medicine, in order to allow medical imaging AI models to offer effective clinical decision support, large amounts of image and clinical data are required by most of the current medical AI models,” Wu said. “[AI models] make use of data, just make use of data from a single or relatively few institutions, also maybe from just a few geographic regions or patient demographics.”
RELATED: Combining arts & engineering: NEXUS hosts student open house
UI Associate Dean for Research and Ph. D programs at the Tippie College of Business Nick Street said he has been working alongside Baek on this project, as well.
Street said what hospitals really need is the records of every patient on Earth collected into one spot, but for security reasons, medical records cannot leave the hospital they are in.
“The problem with that is sharing patient data is not something that we can just send over email. It requires an agreement – the consent from the patient. It is private information so there’s an ethics concern, there’s a regulatory concern, there’s administrator concern,” Street said, “So even if people like me who are doing data science, wants to develop an AI model and then test it against multiple different institutions, there’s always a barrier in terms of patient privacy and patient data sharing.”
The solution, he says, is instead of having the AI sit in one location and collect data sets, or sending a residency person to other hospitals, they can send the AI agent to the location where the knowledge data exists instead. This gives other institutions the opportunity to improve the data.
This collaborative research is made possible through a competition through the National Science Foundation (NSF). Over the next nine months, 28 other teams will work on constructing prototypes and pitches to present to the NSF.
The NSF will then select teams that are able to move on to the next phase, in which the institutions are rewarded a $5 million grant to make the proposals come to life.
Baek, Street, and Wu all said although this will be a tough competition, they are hopeful and optimistic that they will move on to the next stage.
Street said to him, the most exciting part of the competition is the various research expertise Baek connected them all with during this project.
“This to me is the exciting part of working at a place like this, is that you’ve got so many people with so many different backgrounds,” he said. “In this case, we found a way to come together to do something that no one of us could have done on our own.”
