The finding, announced in late January by a team of researchers at Harvard’s Massachusetts General Hospital and the Massachusetts Institute of Technology, is part of a growing medical trend of using algorithms to predict everything from breast cancer and prostate cancer to the likelihood of tumors regrowing. Though research is increasing, scientists say more testing needs to be done before fully unleashing these products into clinical settings.
The tool is called Sybil, named after the prophet in ancient Greek literature. It is a deep-learning model, meaning computers parse through huge data sets to identify and categorize patterns. Sybil was trained on six years of lung scans of patients in the United States and Taiwan, researchers said.
The study results showed Sybil achieved scores scientifically considered “good” and “strong” in predicting lung cancer over six years. It was stronger with its one-year prediction rates, the study scientists noted.
Lung cancer is “the biggest cancer killer because it’s relatively common and relatively hard to treat,” said Florian Fintelmann, an interventional radiologist at the Massachusetts General Cancer Center and study co-author. “If you detect lung cancer early, the long-term outcome is significantly better.”
Cancer is the second-leading cause of deaths globally, and as advances in artificial intelligence software and computing power have increased, it has become a ripe area for researchers to apply the technology in the hope of helping doctors with diagnoses.
Much of the technology involves analyzing large troves of medical scans, data sets or images, then feeding them into complex artificial intelligence software. From there, computers are trained to spot images of tumors or other abnormalities that researchers claim can be more accurate and quicker than the human eye.
In recent years, there’s been a surge in new therapies to combat lung cancer, study researchers said, but many patients still die of the disease due to barriers.
Those who are old and poor might not get screenings because of limited federal funding. Many patients diagnosed with lung cancer either never smoked or are former smokers who quit more than 15 years ago, MIT researchers said, making them ineligible for screenings in the United States.
For those who can get screened, the most common way is using low-dose computed tomography scans, called LDCT. Researchers created Sybil to turbocharge the screening process, allowing software to analyze LDCT images without the assistance of radiologists to predict cancer risk up to six years in advance.
But building Sybil was a challenge, study authors said. Peter Mikhael, a researcher and affiliate of MIT’s Jameel Clinic and its computer science and artificial intelligence laboratory, described it as “trying to find a needle in the haystack.”
Most of the imaging data to train Sybil didn’t contain overt signs of cancer, since early-stage lung cancer is in small portions of the lung and can be hard for the naked eye to spot. To ensure the software could assess cancer risk, the study team “labeled hundreds of CT scans with visible cancerous tumors” and fed them into Sybil before unleashing the software on CT scans with limited signs of cancer, researchers said.
The team used data sets from the National Lung Screening Trial, Massachusetts General Hospital and Chang Gung Memorial Hospital in Taiwan. Some of the data skewed overwhelmingly from White people, according to the study.
Medical experts caution that cancer software requires more study before being put to clinical use, according to government scientists and research studies.
Researchers from Harvard and the Netherlands have said that the skills to translate information generated by AI algorithms remains in the “nascent stage.” Moreover, the benefits AI may provide medicine is currently quite narrow. Even with these detection tools, doctors still need to make diagnoses, design treatment plans and manage overall care.
Other medical experts point out that more testing needs to be done to see how well the software works on a variety of patients, using different scanners and tools. There also needs to be more work done to show the software actually benefits people, either by helping them live longer, preventing cancer or saving time and money. How algorithms work must be transparent, not a “black box,” they said.
MIT researchers said they will continue their work.
“An exciting next step in the research will be testing Sybil prospectively on people at risk for lung cancer who have not smoked or who quit decades ago,” said Lecia Sequist, director of the Center for Innovation in Early Cancer Detection at Massachusetts General Hospital.