AI supports lung cancer screening: false positives decrease by forty percent Major step forward for lung cancer screening with AI

Researchers at Radboud university medical center have developed AI that reduces the number of false positives in lung cancer screening by forty percent. False positives often lead to unnecessary follow-up scans, higher costs and anxiety for patients. ‘We are convinced of the benefits of lung cancer screening, but we need to reduce the burdens. This is a major step forward.’

Lung cancer is the deadliest form of cancer worldwide, causing nearly 25 percent of all cancer-related deaths. Previous studies have shown that preventive screening of high-risk groups can reduce mortality by almost a quarter, because the disease is often detected at an earlier stage. The United States and several European countries have already implemented population-based lung cancer screening programs.

Reducing false positives is essential

A national screening program for lung cancer in the Netherlands will create a large workload for radiologists. Another challenge is the assessment of CT scans: lung nodules can look suspicious but later turn out to be benign. Such false positives lead to additional examinations, higher costs and unnecessary stress for patients. ‘When we introduce screening for lung cancer in the Netherlands, we need to reduce the number of false positives as much as we can,’ says Colin Jacobs, research group leader at the Imaging department of Radboud university medical center. ‘AI can help by estimating the cancer risk of each nodule more accurately, which means fewer unnecessary follow-up examinations.’

Training the algorithm with existing scans

That is why researcher Noa Antonissen validated an AI model developed at Radboudumc. ‘Our model was trained on U.S. lung cancer screening data with more than 16,000 lung nodules on CT scans, of which over 1,000 were malignant. The model creates a sort of 3D image of each nodule. Based on that, the AI calculates the likelihood that the nodule is malignant,’ Antonissen explains.

The researchers then tested the AI model on images from large international studies in the Netherlands, Belgium, Denmark and Italy. They compared the performance of the AI model with a widely used risk model (the PanCan model). Especially in the difficult group of nodules between 5 and 15 millimeters, AI clearly performed better: the number of false positives dropped by forty percent, while all cancer cases were still detected.

Important step toward practice

This study from Radboudumc shows that AI can play a major role and make screening much more efficient. Antonissen: ‘With this study, we show that AI can be an important tool to make lung cancer screening more effective. We keep the benefits and reduce the burdens for patients and healthcare. The next step is to prove this in clinical practice.’

About this publication

This research was published in Radiology: External Test of a Deep Learning Algorithm for Pulmonary Nodule Malignancy Risk Stratification Using European Screening Data. N. Antonissen, K. Vaidhya Venkadesh, R. Dinnessen, E. Scholten, Z. Saghir, M. Silva, MD, U. Pastorino, G. Sidorenkov, M. Heuvelmans, G. de Bock, F. Hoesein,  P. de Jong, H. Groen, R. Vliegenthart, H. Gietema, M. Prokop, C. Schaefer-Prokop, C. Jacobs, for the NELSON-POP consortium. DOI: 10.1148/radiol.250874.

• Want to know more about these subjects? Click on the buttons below for more news.

  News homepage news & media