Lab Manager | Run Your Lab Like a Business
3D render of cancer in the lung

MIT Researchers Develop a Model for Predicting Lung Cancer

AI tool can analyze CT scans and accurately predict if a patient will develop cancer within six years

Holden Galusha

Holden Galusha is the associate editor for Lab Manager. He was a freelance contributing writer for Lab Manager before being invited to join the team full-time. Previously, he was the...

ViewFull Profile.
Learn about ourEditorial Policies.
Register for free to listen to this article
Listen with Speechify

Researchers from MIT have announced a new artificial intelligence (AI) tool that can analyze lung scans and predict if the patient will develop lung cancer within six years. The software, dubbed Sybil, has predicted patient outcomes with exceptional accuracy.

Lung cancer is the deadliest cancer on the planet. In 2020, it resulted in 1.7 million deaths worldwide—more people than the next three deadliest cancers (colon, pancreatic, and breast) combined. Because of its aggression, the key to successfully treating lung cancer is early detection. According to Florian Fintelmann, thoracic interventional radiologist at MIT’s Abdul Latif Jameel Clinic for Machine Learning in Health, Mass General Cancer Center (MGCC) and one of the paper’s co-authors, “...if you detect lung cancer early, the long-term outcome is significantly better. Your five-year survival rate is closer to 70 percent, whereas if you detect it when it’s advanced, the five-year survival rate is just short of 10 percent.”

Get training in Lab Crisis Preparation and earn CEUs.One of over 25 IACET-accredited courses in the Academy.
Lab Crisis Preparation Course

Researchers hope that Sybil will help make it possible to accurately predict if a patient will develop lung cancer within six years, before signs of it show up on scans. Sybil works by analyzing three-dimensional low-dose computed tomography (LDCT) scans, which are the most common method of lung cancer screening. While humans and previous cancer detection algorithms rely on visual features like pulmonary nodules to predict lung cancer, Sybil can pick up on imaging “predictive of future lung cancer risk beyond currently identifiable features such as lung nodules.” The paper detailing Sybil’s creation, published in the Journal of Clinical Oncology, indicates that Sybil achieved exceptionally high C-indices over the six years it was tested. A C-index, or concordance index, is a metric that evaluates how good a prediction model is. C-index scores below 0.5 indicate the model is poor, a score of 0.5 means the model is equivalent to random guess, and values over 0.7 indicate a good model, with scores approaching 1 being a very strong model. Sybil, trained on data from the National Lung Screening Trial with over 50,000 participants, has achieved C-indices ranging from 0.75 to 0.94 over the course of its testing.

Because Sybil was initially trained only on scans from smokers, as they are the only ones eligible for CT screening in the US, the researchers are now aiming to test the model with data from non-smokers or those who quit decades ago. The nonsmoker data will be from Taiwanese patients, as Taiwan does offer LDCT screenings for non-smokers. Over 50 percent of women diagnosed with lung cancer are nonsmokers, as well as 15 to 20 percent of men. Confirming that Sybil’s predictive abilities are accurate with nonsmokers may open the door to helping them detect cancer early enough for effective treatment.