Deep Learning Model Helps Tailor Screening Protocol for Lung Cancer Risk
NCI-funded researchers collaborated with scientists from NCI’s Division of Cancer Epidemiology and Genetics and Division of Cancer Prevention on a recent study using a deep learning algorithm that could help personalize lung cancer screening for high-risk patients (i.e., individuals who either currently or formerly smoked).
Read the full report in JAMA Network Open.
The researchers applied a deep learning algorithm to low dose computed tomography (LDCT) scans from 10,831 people. They classified the patients according to risk: those with suspicious nodules who needed aggressive follow up and those at lower risk who could receive screening less frequently.
Oncologists recommend annual LDCT screening for patients at high risk for lung cancer, because it can detect disease at the earliest stage. However, LDCT does have risks, including radiation exposure and the possibility of “false positive” results.
According to Dr. Rebecca Landy, primary author on the study, “The deep learning algorithm predicted risk more accurately than existing statistical models. By using risk to inform who we screen and when, we can reduce harm while also making screening more cost-effective.”
As noted by Dr. Landy, “Clinical decision support tools (like prediction models and deep learning algorithms) have the potential to enhance how we care for patients over their lifetimes, enabling us to make adjustments as needed to better inform their care.”