BUFFALO, N.Y. — Mingchen Gao is combining artificial intelligence with medical imaging in hopes of improving future health care. Her lab is doing so by developing algorithms that allow machine-learning models to analyze medical images.

"Machine learning is more about using the previous data, for example, the previous scans, CT scans and also some diagnoses from the doctors, some labeling from the physicians, something like that," said Gao, assistant professor of computer science and engineering for the University at Buffalo. "So we use that information to predict future events and say, if I have a new patient, also having medical images, and how can we predict a similar diagnosis as in the previous data?"

She says sometimes details of scans and such can be overlooked by doctors, that’s where AI can come in to help.

"For example, some kinds of early-stage cancers, they are not really observable on CT scans, but we are using those machine learning models to pick up those subtle signals, subtle information from CT scans," said Gao.

She says it could also be helpful for second opinions. However, there are challenges in combining AI and imaging. One of those hurdles is creating AI tools that can work in the real world instead of lab-controlled experiments. 

"That's much more difficult because of the quality of the images and sometimes because of the previous data they had there only for a certain type of people, like people from a certain race, but not from like a broad population. So how can we adapt the model to a broader range of patients?"

Another issue often brought up in discussions of AI is privacy issues. 

"So in many cases, we don't want the patient data to leave the hospitals because of the privacy concern," Gao said. "But the machine learning model, they need to use a huge amount of patient data."

Those are just a few of the challenges Mingchen is trying to solve through her research lab. She stresses that AI can’t replace doctors.

"I think AI is a tool to assist radiologists and doctors and also patients eventually probably," she said. "If the patient has questions about their radiology images in the report, we can also use AI to help explain some of that like medical jargon."

She says with all the different types of illnesses and diseases out there, there’s still so much work to do when combining the medical and artificial intelligence fields and in her lab as well.