Google’s cancer-spotting AI outperforms radiologists in reading lung CT scans

When applied to over 45,800 deidentified chest CT screenings, the AI detected 5% more cancer cases and 11% fewer false positives compared to unassisted radiologists, according to Google. (Wikimedia Commons)

Google’s artificial intelligence development has reached a milestone in lung cancer imaging and prediction, with a CT scan model being able to diagnose cases as well as or better than a group of six radiologists.

Alongside researchers at Northwestern University, New York University-Langone Medical Center and Stanford Medicine, Google’s AI team was able to generate an overall prediction of malignancy from 3D volumetric scans, as well as identify subtle lung nodules.

Instead of viewing hundreds of individual 2D slices from a single low-dose CT scan as a radiologist would, the AI examines the lungs as a single 3D object—while also noting any changes over time by comparing differences with a patient’s previous scans in order to help measure the growth rates of any suspicious lesions.


Like this story? Subscribe to FierceBiotech!

Biopharma is a fast-growing world where big ideas come along every day. Our subscribers rely on FierceBiotech as their must-read source for the latest news, analysis and data in the world of biotech and pharma R&D. Sign up today to get biotech news and updates delivered to your inbox and read on the go.

RELATED: Google AI predicts hospital inpatient death risks with 95% accuracy

And when applied to over 45,800 de-identified chest CT screenings, gathered from research data sets from the National Institutes of Health and Northwestern University, the researchers were able to detect 5% more cancer cases while reducing false-positives by more than 11% compared to the findings of unassisted radiologists.

In one case, Google described how the AI detected potential lung cancer that had previously been ruled out in an asymptomatic patient with no history of cancer. In a company blog post, technical lead Shravya Shetty wrote that this early progress lays the groundwork for future clinical testing. The researchers’ results were published in Nature Medicine.

RELATED: Google to retake DeepMind’s AI healthcare unit as it moves toward a cohesive strategy

Going forward, Google’s AI team will be collaborating with the company’s cloud-based healthcare and life sciences divisions to prepare the model for further clinical validation and eventual deployment, according to Shetty.

Suggested Articles

Avidity Biosciences is on a roll—after inking an R&D deal with Eli Lilly and hiring a new CEO, the company is reeling in $100 million.

What the NASH field needs, says Genfit CEO Pascal Prigent, is something like the Hb1Ac test for diabetes.

Dubbed “Project Nightingale,” the efforts were announced amid concerns and federal inquiries into the data’s safekeeping and patient consent for use.