MIT professor’s AI tools to detect breast and lung cancer get her recognized in TIME

Dr. Regina Barzilay’s world turned upside down when she was diagnosed with breast cancer in 2014.

It seemed to come out of nowhere. She didn’t know anyone in her family who had BRCA genes — the genes normally associated with higher risks of developing breast and ovarian cancer — and she was living a healthy lifestyle.

Barzilay said she wishes that she had had access to an artificial intelligence tool she later developed with Massachusetts Institute of Technology students called Mirai. The tool takes a mammogram and determines the likelihood of developing the disease in five years.

Mirai, released in 2019, is among several AI tools Barzilay has helped to lead and develop that landed her in TIME Magazine’s 2025 list of artificial intelligence leaders, innovators, shapers and thinkers, called TIME100 AI 2025.

Her artificial intelligence tool would’ve predicted that she was high risk — and not only that, but would’ve picked it up two years earlier than her doctors did, she said.

Read more: Trump says higher ed is too liberal. Tufts launched a center to tackle that issue

“You can see in 2013 and 2012, it was already there if you blow it up,” Barzilay said, showing a MassLive reporter her previous mammograms.

The most common cancer in women worldwide is breast cancer, and half of breast cancers develop in those that don’t have identifiable risk factors, other than gender and age, according to the World Health Organization.

“It’s really not a unique story. It happens to many, many women whose disease — even they go and do their mammograms quite regularly — it’s just not diagnosed on time. So I think the technology can do a much better job than using our eyes,” said Barzilay, who is a professor at MIT and the AI Faculty Lead at MIT Jameel Clinic.

Diagnosing breast cancer early not only makes it so treatment is less “brutal” with fewer side effects like the loss of hair, but also increases a person’s survival rate, she said.

What are the other tools she’s developed?

Beyond Mirai, Barzilay has also worked on an AI model called Sybil, which predicts 6-year lung cancer risk from a patient’s low-dose CT scan.

Lung cancer is the world’s deadliest cancer. In 2023, 131,584 people in the U.S. died from lung cancer, according to the Centers for Disease Control and Prevention.

Created in collaboration with Massachusetts General Hospital, the tool was inspired by Barzilay’s sister’s friend, who was diagnosed with and died from lung cancer in her 50s despite being healthy and never having smoked.

“She started coughing [but] her doctor didn’t identify that she may be at risk. So it took a very long time until they even sent her to do a simple X-ray,” she said. “It was too late.”

“We tested it extensively on different populations and different races and ages and genders and are looking at safe clinical pipelines to bring it to the patients,” she said.

Sybil is being moved forward beyond just the U.S. — it is undergoing Taiwanese FDA approval due to the high rates of lung cancer there, Barzilay said.

She has also joined efforts with other educators and researchers, such as MIT professor Dr. Jim Collins, to create an AI model that screened over 100 million molecules to identify the first AI-discovered antibiotic in 2020.

Read more: Huge gift to UMass Amherst a ‘gamechanger’ amid federal funding uncertainty

Much of the work that Barzilay has done has been focused on reducing uncertainty in the medical profession.

“I think that whenever you’re going to a doctor and if you really have a question, concern, during a checkup, we very soon realize how much uncertainty we have because the doctor can only predict what’s going happen. But most of the time they would say we think this would happen, but we don’t really know,” she said.

“What AI is good at is actually taking whatever data we have and giving you likelihoods with probability, which gives you a way to reduce your uncertainty,” she said.

Addressing distrust in artificial intelligence

As AI has become more of a topic of discussion over the past few years with the proliferation of tools like ChatGPT, Barzilay said work needs to be done to improve people’s trust in AI tools.

AI is not just ChatGPT, Barzilay said. AI tools in clinical care, in particular, are much more narrow — and they need to be clear and validated.

“I am not saying that we should be blindly trusting to the technology, but I think that if technology is properly validated, FDA approved, properly incorporated into the clinical pipeline, then I think you should be able to trust it,” she said.

For example, Mirai has been validated in over 2 million mammograms in over 70 hospitals and 22 countries and Sybil in 25 hospitals and 11 countries.

While medical professionals’ errors can be common, many people still go to the doctor when something is wrong, Barzilay said.

“Do you really trust that your blood sample, when you get blood was not contaminated? It would not even come to your mind. Why? Because you ensure all the processes are done correctly. So the question is, can we bring AI to the point, both informing the public and internally validating it properly so we start thinking about AI the same way as we think about any other medical test or device that we’re using in patient care,” she said.

She said the use of AI in collaboration with doctors in the medical field, even if scary for some people, is the future.

“To me, the concern is not the fact whether it’s going to make a mistake. To me, the concern is why in all the areas of our life we have AI and we trust it — to a degree, of course — but in the area which maybe matters the most, our health, we still don’t have it. And to me this is the main concern,” she said.

Read more: A judge ruled Harvard should get its federal funding back. What happens next?

For Barzilay, the next step is to take the tools she’s helped develop, including Mirai and Sybil, and bring them to the U.S. clinical setting.

However, the biggest challenge in getting these tools to patients isn’t getting FDA approval — which still needs to be done — but in developing clinically approved protocols to determine what should happen after someone is determined to be high risk.

“It’s not enough just to predict the future. You should want to improve the future,” Barzilay said.