13 November 2023

In the Netherlands and many other European countries, no nationwide lung cancer screening program exists yet. Consequently, early-stage lung cancer is generally diagnosed after identifying an abnormal spot on a chest CT scan ordered for other medical reasons. These spots or lung nodules are difficult to find and can potentially be overlooked, as they are not necessarily the focus of an examination and can be as small as a few millimeters. Software using AI has proven to be successful for aiding radiologists in this task, but its performance is understudied outside lung cancer screening where the scans are very diverse.

Researcher Ward Hendrix from the Diagnostic Image Analysis Group (Department of Medical Imaging) therefore developed and validated AI software for the detection of lung nodules in non-screening CT scans at the Radboudumc and Jeroen Bosch Hospital. This research is part of a collaboration between the hospitals and initiated by Colin Jacobs, Matthieu Rutten, Mathias Prokop, and Bram van Ginneken. Their research was published on the 27th of October in the new journal Communications Medicine from the Nature portfolio.

The researchers retrospectively collected CT scans at both hospitals and asked a panel of five expert (thoracic) radiologists to independently label all lung nodules. Two additional radiologists verified the malignancy status of each nodule and searched for any missed cancers by using data from the national Netherlands Cancer Registry. Based on this data, the researchers found that the AI software could accurately locate cancerous nodules as well as non-cancerous nodules that would need attention. A comparison with the panel of the radiologists showed that the AI software was also able to locate nodules in common blind spots of radiologists.

These findings suggest that AI is ‘mature’ enough to detect lung nodules in heterogenous CT scans from our hospitals and has the potential to aid radiologists in this setting. In future work, the research team will investigate whether the software can be optimized for rare forms of lung cancer. The algorithm is publicly available on the web platform Grand Challenge.

Read the study here:

Hendrix W, Hendrix N, Scholten ET, Mourits M, Trap-de Jong J, Schalekamp S, Korst M, Van Leuken M, Van Ginneken B, Prokop M, Rutten M, Jacobs C. Deep learning for the detection of benign and malignant pulmonary nodules in non-screening chest CT scans. Communications Medicine. 2023 Oct; 3(1):156. Doi: 10.1038/s43856-023-00388-5. PMID: 37891360; PMCID: PMC10611755.

Related news items