Researchers Develop a Model to Predict Lung Cancer Risk Using CT Scans

Using a single, low-dose, chest computed tomography (CT) scan, researchers in the Journal of Clinical Oncology detail a model to predict lung cancer risk. According to the American Cancer Society (ACS), over 200,000 people are diagnosed with new lung cancer each year in the United States, contributing to over 130,000 lung cancer-related deaths annually.

According to a study presented at the Radiological Society of North America, only 16% of lung cancers are diagnosed early on in disease progression. That contributes to the low one-year survival rate of less than 50% and even lower five-year survival rate of 18.6%.

Authors of the study in the Journal of Clinical Oncology note that, despite being an effective and useful screening tool, low-dose computed tomography (LDCT) is often not utilized by those eligible for screening. Researchers set out to develop a deep-learning model to assess LDCT data and predict individual lung cancer risk.

The model, called Sybil, was built using LDCT data from the National Lung Screening Trial. According to the publication, the model only requires one scan and is self-sufficient, meaning additional clinical data and radiologist annotations are unnecessary.

To validate the model, researchers used three data sets, including 6,282 LDCTs from the National Lung Screening Trial (NLST), 8,821 LDCTS from Massachusetts General Hospital (MGH), and 12,280 scans from Chang Gung Memorial Hospital (CGMH).

Across all three data sets, the model was relatively accurate at predicting lung cancer within the next year. In the statistical analysis, researchers found that the area under the receiver-operator curves was 0.92, 0.86, and 0.94 for the NLST, MGH, and CGMH groups, respectively. The six-year incidences yielded areas of 0.75, 0.81, and 0.80 for each respective category.

Based on the results, investigators in the study concluded that “Sybil can accurately predict an individual's future lung cancer risk from a single LDCT scan to further enable personalized screening. Future study is required to understand Sybil's clinical applications. Our model and annotations are publicly available.”