Deep-Learning Tool Can Predict Lung Cancer Risk Within Six Years

Researchers from Mass General Brigham and the Massachusetts Institute of Technology (MIT) have developed a deep-learning (DL) tool that can predict the risk of lung cancer for individuals with or without a significant smoking history within six years.

Statistics from the Centers for Disease Control and Prevention (CDC) indicate that lung cancer is the third most common cancer in the US, with more people dying from lung cancer than any other type. This trend is partially due to an increase in lung cancer rates among non-smokers, according to the study describing the tool published in the Journal of Clinical Oncology.

Low-dose chest computed tomography (LDCT) is recommended to screen people between 50 and 80 years of age with a significant history of smoking or who currently smoke, the researchers explained, noting that lung cancer screening with LDCT has been shown to reduce death from lung cancer by up to 24 percent.

However, non-smokers are not eligible for screening under these guidelines despite increased risk for this population, facilitating the need for broader screening methods. To bridge this gap, the researchers developed Sybil, a tool designed to analyze LDCT scans and predict the risk of lung cancer across populations with and without a significant history of smoking.

“Lung cancer rates continue to rise among people who have never smoked or who haven’t smoked in years, suggesting that there are many risk factors contributing to lung cancer risk, some of which are currently unknown,” said corresponding author Lecia Sequist, MD, director of the Center for Innovation in Early Cancer Detection and a lung cancer medical oncologist at the Mass General Cancer Center, in the press release. “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 press release states that the US Preventive Service Task Force recommends annual LDCTs for people over 50 with a history of 20 pack-years who either currently smoke or have quit smoking within the last 15 years. However, less than 10 percent of patients eligible under these criteria are screened each year. The research team aims to leverage Sybil to improve the efficiency of lung cancer screening and provide individualized assessments.

“Sybil requires only one LDCT and does not depend on clinical data or radiologist annotations,” said co-author Florian Fintelmann, MD, of the Department of Radiology, Division of Thoracic Imaging & Intervention at Massachusetts General Hospital, in the press release. “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.”

According to the press release, Sybil was developed using data from the National Lung Screening Trial (NLST) and validated using three independent datasets: scans from more than 6,000 NLST participants the tool had not previously seen; 8,821 LDCTs from Massachusetts General Hospital (MGH); and 12,280 LDCTs from Chang Gung Memorial Hospital in Taiwan. These data sampled people with a range of smoking histories, including those who never smoked, according to the study.

By measuring Sybil’s accuracy using area under the curve (AUC), a measure of how well a test can distinguish between disease and normal samples, the researchers found that the tool can predict lung cancer risk within one to six years.

The tool predicted cancer within one year with AUCs of 0.92 for the additional NLST participants, 0.86 for the MGH dataset, and 0.94 for the dataset from Taiwan. Further, Sybil predicted lung cancer within six years with AUCs of 0.75, 0.81, and 0.80 for those cohorts, respectively. For the AUC measure, 1.0 is a perfect score.

“Sybil can look at an image and predict the risk of a patient developing lung cancer within six years,” said co-author and Jameel Clinic faculty lead Regina Barzilay, PhD, a member of the Koch Institute for Integrative Cancer Research, in the press release. “I am excited about translational efforts led by the MGH team that are aiming to change outcomes for patients who would otherwise develop advanced disease.”

This is the latest example of researchers leveraging DL for lung cancer prediction.

Earlier this month, researchers validated a DL model to predict lung cancer risk using chest radiographs and EMR data, which may help identify individuals who could benefit from lung cancer screening, including those missed by Medicare screening eligibility criteria.

Lack of lung cancer screening uptake due to limitations in US Centers for Medicare & Medicaid Services (CMS) guidelines can perpetuate health disparities, the researchers explained. To combat this, they designed an automated prediction tool designed to complement CMS criteria, which accurately identified patients at high risk for lung cancer, including those missed by Medicare lung cancer screening eligibility criteria.