Sybil: A“ Hunting Hound” For Spotting Future Lung Cancer
Sybil: A“ Hunting Hound” For Spotting Future Lung Cancer
“ A Deep Learning Image Analysis Tool for Identifying Individuals at Risk for Lung Cancer” – Allison Chang, MD, PhD, 2024 Conquer Cancer Young Investigator Award
The current criteria in the U. S. for annual lung cancer screening using a CT scan is quite extensive. It can also exclude many at‐risk individuals, as only 50 % of people diagnosed with lung cancer would have qualified for screening prior to their diagnosis. 3 Any solution to this must address the dual goals of expanding the ability to screen more individuals and accurately identify who qualifies for routine screening.
“ We currently recommend an annual low‐dose CT scan for people between the ages of 50 and 80 if they have at least a 20 pack‐year smoking history and if they are either currently smoking or have quit within the last 15 years. When you think about it, that is fairly restrictive because if you are somebody who smoked a pack of cigarettes a day from ages 20 to 40 and you are 60 now, you don’ t actually qualify for lung cancer screening anymore,” said Allison Chang, MD, PhD, an attending physician in thoracic cancers at Massachusetts General Hospital.
To make improvements in this area, Dr. Chang and researchers at Massachusetts General Hospital have developed an AI model called Sybil that could help clinicians use CT scans to not only detect lung cancer but also determine future risk. Data presented over the last 3 years have shown that Sybil accurately predicted future lung cancer risk among screen‐eligible populations with significant tobacco use in the National Lung Screening Trial. Because the United States only has CT scans from smokers, researchers also found that Sybil was able to accomplish the same feat with data from individuals in Taiwan and South Korea who have and have not used tobacco.
To show the value of this on the patient level, Dr. Chang shared two of the scans that Sybil was trained and tested on during her presentation at the International Association for the Study of Lung Cancer’ s 2025 World Conference on Lung Cancer. In the first one, the clinician found no abnormalities that would cause concern for lung cancer, but Sybil spotted high‐risk activity. A biopsy a year later found 2.2 cm of poorly differentiated squamous cell lung cancer.