Artificial Intelligence (AI) is an increasingly popular topic in the news and conversation. The tools are rapidly changing the way most of us live and work. Many people are concerned about the dangers and risks, calling for caution and regulation. On the other hand, AI has enormous potential to boost human productivity and solve complex problems. The PBS science series, NOVA recently featured an episode titled “A.I. Revolution,” highlighting several ways AI is “quietly revolutionizing medicine,” including projects at Mass General Brigham using AI to predict cancer.

Mass General Cancer Center oncologist Lecia Sequist, MD, MPH, talked with the show about Sybil, an AI tool that can correctly forecast lung cancer much earlier than expert human radiologists.

An AI model for lung cancer screening

Lung cancer is the leading cause of cancer in both men and women in the United States. It’s also the leading cause of cancer death worldwide. The disease is much easier to treat when it’s caught at earlier stages. Although screening can detect early-stage lung cancer, humans are not yet able to predict who will develop lung cancer in the future.

But Sybil, an AI computer model, can.

Using AI in oncology to forecast disease

Regina Barzilay, PhD was working as a computer scientist, using AI to interpret dead languages, when she was diagnosed with breast cancer. She and Mass General Brigham radiologist Constance Lehman, MD, PhD, decided to explore how AI can predict the disease.

They collected data on 128,000 mammograms, including 3,800 that had led to cancer diagnosis within five years. They used the data to develop a model to predict the likelihood of a person developing breast cancer in the next five years. The software they developed, called MIRAI, was 75% to 84% accurate in predicting future cancer diagnoses.

And then, a friend of Barzilay was diagnosed with lung cancer and received treatment from Dr. Sequist. The team wondered: Could a similar AI model be developed for lung cancer?

Dr. Sequist’s team at the Mass General Cancer Center’s Early Detection and Diagnostics Clinic worked with Dr. Barzilay’s team to develop a model called Sybil to examine low-dose chest computed tomography scans.

“We taught the model to recognize the patterns of risk that indicate a future lung cancer by using thousands of CT scans from patients who were participating in a clinical trial,” Dr. Sequist says. “We had a lot of information about them, including who was diagnosed with cancer and when, demographic information, health information, and outcomes information. But importantly, once the model was trained, all that is needed is the CT scan itself–no other information about the patient is needed to determine future lung cancer risk.”

Sybil has been able to forecast lung cancer correctly about 80% to 95% of the time in the populations tested—even before expert human eyes can see any changes or signs of cancer. The model is now in further clinical trials at Mass General Cancer Center and collaborating sites around the country. Research results have been published in the Journal of Clinical Oncology and featured on NBC News.

Learn about Sybil and other early-detection research at the Mass General Cancer Center’s Early Detection and Diagnostics Clinic.