AI tool predicts lung cancer without radiologists or clinical histories
A new artificial intelligence tool can predict the risk of lung cancer in patients using only a low dose CT scan and without radiologist annotation [1].
The tool, known as Sybil, was validated on three datasets of more than 20,000 scans. It was able to accurately predict which of these patients were most at risk of developing lung cancer in the years following their LDCT scan.
“Sybil requires only one LDCT and does not depend on clinical data or radiologist annotations,” explained study co-author Florian Fintelmann, MD, of the Department of Radiology at Massachusetts General Hospital. “It was designed to run in real-time in the background of a standard radiology reading station which enables point-of care clinical decision support.”
Sybil is a result of the combined efforts of Mass General Cancer Center and the Massachusetts Institute of Technology (MIT). The deep learning model was developed and trained to predict risk of lung cancer in the one to six years following a patient’s LDCT scan, and it does not require clinical information relative to risk factors to do so, which is important since lung cancer rates are growing among individuals who do not have a history of persistent smoking, researchers explained.
“Instead of assessing individual environmental or genetic risk factors, we’ve developed a tool that can use images to look at collective biology and make predictions about cancer risk,” the experts noted.
Sybil was validated using data from the National Lung Screening Trial (NLST), in addition to 8,821 LDCTs from Massachusetts General Hospital and 12,280 scans from Chang Gung Memorial Hospital in Taiwan. The LDCTs from Taiwan were completed on patients with a variety of smoking backgrounds, some of whom had no history of smoking at all.
The AI tool achieved AUCs of 0.92 (NLST participants), 0.86 (MGH dataset) and 0.94 (Taiwan dataset) when predicting who would develop lung cancer within one year of their scan, and 0.75, 0.81 and 0.80 for assessing risks at the six-year mark. In many cases, Sybil identified patterns that were not visible to radiologists.
In the future, researchers plan to further validate the tool on more diverse datasets. This study was retrospective, but a prospective study that tests Sybil’s real-world utility is in the works.
Study data is published in the Journal of Clinical Oncology.
In addition to her background in journalism, Hannah also has patient-facing experience in clinical settings, having spent more than 12 years working as a registered rad tech. She began covering the medical imaging industry for Innovate Healthcare in 2021.