Beyond just nodules—model uses all low-dose CT data to make long-term lung cancer risk predictions
Including global lung features in risk assessments based on low-dose CT (LDCT) lung screenings could improve long-term cancer risk predictions.
That’s according to new data published in RSNA’s Radiology this week. The study details the utility of a prediction model that integrates global lung features alongside standard data relative to lung nodules to make more accurate estimations of long-term cancer risk. In testing, the model improveD these risk assessments, even in individuals whose initial LDCT scans were deemed negative.
Experts involved in the research suggested their findings highlight the shortcomings of using nodule characteristics and patient history alone to predict an individual’s true lung cancer risk.
“Although several models have demonstrated excellent performance in nodule diagnosis, risk assessment based solely on visible nodules has inherent limitations. For example, the NELSON trial reported a 12.8% rate of interval lung cancers and 40.1% rate of post-screening lung cancers, which often lack visible nodules on baseline scans,” Chengting Lin, MD, with the Department of Radiology at Zhejiang Cancer Hospital in China, and colleagues explained. “There is accumulating evidence of an association among chronic inflammation, emphysema, pulmonary fibrosis, and lung cancer risk. CT imaging can be used to assess these global lung abnormalities, providing additional value for risk prediction.”
Using LDCT data from the Wenling 2019–2020 screening cohort (Zhejiang Province, China) and clinical CT data from Zhejiang Cancer Hospital (2017–2023), the team developed a three-year risk prediction model they refer to as ScreenLungNet. Multiple nodule and global lung features were integrated into the model’s framework. The group developed three additional models as well—one that makes predictions based on a single high-risk nodule, one that does so using multiple nodules and one that assesses risk based on global lung features alone—to compare the utility of a model that uses both nodule and lung feature data together.
The model was tested internally and externally on data from the Wenling 2021 and National Lung Screening Trial cohorts. On these test sets, ScreenLungNet outperformed all other models for 3-year risk predictions, achieving an area under the curve of 0.94. On the National Lung Screening Trial cohort, the model yielded an AUC of 0.93, accuracy of 94.8%, sensitivity of 84.9%, specificity of 95.2%, positive predictive value of 44% and a negative predictive value of 99.3%, respectively. This performance was maintained in the National Lung Screening cohort with baseline negative LDCT scans as well (AUC, 0.87; accuracy, 97%; sensitivity, 72.1%; specificity, 97.5%).
“Global lung features proved critical for long-term prediction, particularly in screening-negative participants for whom radiologic signs were unclear,” the group noted. “Additionally, the global lung model outperformed the multiple-nodule model in terms of AUC, reversing the trend observed in the full [National Lung Screening Trial] cohort. Moreover, models incorporating global lung features also showed higher sensitivity than nodule-only models did, with a more pronounced improvement in the screening-negative group.”
The authors believe their model could help address the limitations of relying on nodules alone, though additional validation is needed.
Read more here.
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.