Study shows AI beats radiologists in detecting lung cancer

Register now

Testing has shown that a deep learning algorithm developed by scientists at Google and Northwestern Medicine appears better at spotting lung cancer than experienced radiologists.

Lung cancer is the most common cause of cancer death in the United States, with about 160,000 deaths occurring in 2018. Lung cancer screening using low-dose computed tomography (CT) can reduce mortality by 20 percent to 43 percent and is now included in U.S. screening guidelines. However, screening error rates can be high, resulting in delayed or missed diagnoses and unnecessary biopsies and follow up procedures, all of which come at high clinical and financial costs.

In a recent study, scientists at Google developed a large-scale deep learning algorithm that uses a patient’s current and prior CT volumes to predict a patient’s risk of lung cancer. They applied the model to 2,763 de-identified CT scan sets provided by Northwestern Medicine to validate the accuracy of the new system.

The deep learning system provides an automated image evaluation system to enhance the accuracy of early lung cancer diagnosis. The study was published in Nature Medicine.

“Radiologists generally examine hundreds of two-dimensional images or ‘slices’ in a single CT scan but this new machine learning system views the lungs in a huge, single three-dimensional image,” says study co-author Mozziyar Etemadi, MD, a research assistant professor of anesthesiology at Northwestern University Feinberg School of Medicine and of engineering at McCormick School of Engineering.

“AI in 3D can be much more sensitive in its ability to detect early lung cancer than the human eye looking at 2D images. This is technically ‘4D,' because it is not only looking at one CT scan, but two (the current and prior scan) over time,” he says.

The researchers conducted a two-part retrospective study with six U.S. board certified radiologists. When prior CT images were not available, the model outperformed all six radiologists, with absolute reductions of 11 percent in false positives and 5 percent in false negatives. Where prior CT imaging was available, the model’s performance was on par with the radiologists.

"Our work examines ways AI can be used to improve the accuracy and optimize the screening process, in ways that could help with the implementation of screening programs,” says Shravya Shetty, technical lead at Google. The results are promising, and we look forward to continuing our work with partners and peers."

For reprint and licensing requests for this article, click here.