CT-Based Deep Learning Model May Reduce False Positives with Indeterminate Lung Nodules by Nearly 40 Percent

New research suggests that a computed tomography (CT)-based deep learning model may offer comparable detection of lung cancer while offering a significant reduction of false positive findings in contrast to the Pan-Canadian Early Detection of Lung Cancer (PanCan) risk stratification model.

For the retrospective study, recently published in Radiology, researchers assessed the deep learning model in a cohort of 4,146 participants with a median smoking history of 38 pack years. The cohort included 7,614 benign nodules and 180 malignant nodules, according to the study authors.

The researchers found that the deep learning model offered an equivalent AUC for predicting cancer at one year (98 percent) and a slightly higher AUC at two years (96 percent vs. 94 percent) in comparison to the PanCan model. For cases involving indeterminate nodules on CT, the study authors noted the deep learning model correctly identified 68.1 percent of benign lesions as low risk at one year in comparison to 47.4 percent with the PanCan model.

“We saw that we could reduce the false positive rate relative to (the) PanCan (model) by 39.4 percent, which is substantial, without missing any of the cancers that required urgent care … cancers that were diagnosed within one year during the initial screening trials,” noted lead study author Noa Antonissen, M.D., who is affiliated with the Department of Medical Imaging at Radboud University Medical Center in Nijmegen, the Netherlands.

“… Having an AI tool that can do an accurate risk stratification at the level of an expert radiologist can really help to manage these screenings better, so that we can reduce the number of false positive screens in the end to the minimum that we need to still detect the cancers that we find in the screening program,” added study co-author Colin Jacobs, Ph.D., an associate professor of artificial intelligence in thoracic oncology at the Radboud University Medical Center in Nijmegen, the Netherlands. “I think that is the ultimate goal that we have with this research: that we can support screening programs with AI tools that improve the effectiveness and the efficiency of our screening programs.”

(Editor’s note: For related content, see “Can Deep Learning Enhance Low-Dose Chest CT Assessment of Lung Cancer Risk?,” “Can CT-Based Deep Learning Bolster Prognostic Assessments of Ground-Glass Nodules?” and “Olympus Launches CT-Based AI Software for Emphysema Screening.”)

For more insights from Drs. Antonissen and Jacobs, watch the video below.