Researchers from MIT and the Massachusetts General Cancer Center have created a new AI tool capable of predicting the probability of an individual developing lung cancer within a six-year period. The study on the new device, dubbed Sybil in reference to the ancient Greek oracle, was published in the Journal of Clinical Oncology in January.
The software in play was specifically designed to bypass two sticking points in what has become the standard for lung cancer screening guidelines. Firstly, the current methodology directs the majority of its attention to current or past smokers despite an alarming trend of nonsmoker diagnoses being on the rise. Accordingly, researchers made sure to apply Sybil to those without a history of smoking. Secondly, convenience of screenings was a priority for current or former smokers, as the requirements of the suggested low-dose computed tomography (LDCT) chest scans for those over age 50 result in fewer than 10% of those in that group actually following through with their scans.
Sybil was trained with a data set comprising more than 8,800 LDCTs from Mass General as well as more than 12,000 scans from the Chang Gung Memorial Hospital in Taiwan. The latter accounts for a wide range of diverse smoking histories, and importantly, brings in data from nonsmokers. The tool was tested with those collective 20,000+ scans in addition to previously unseen scans from 6,000 patients in the National Lung Screening Trial.