Study: Google AI outperforms radiologists in diagnosing lung cancer

A deep learning algorithm developed by members of Google's artificial intelligence team achieved "state-of-the-art performance" in detecting easily overlooked indicators of lung cancer in CT scans, according to a study published this week in Nature Medicine.

The AI was able to retroactively diagnose the presence or absence of lung cancer with 94 percent accuracy in 6,716 National Lung Cancer Screening Trial cases with known diagnoses, and with similar accuracy in 1,139 more cases from a clinical validation set. The images showed a variety of diagnoses, from fully developed cancer to the presence of pre-cancerous nodules to completely healthy lungs.

When multiple scans for the same patient were available, the model performed on-par with a group of six radiologists, but when only one scan was available, the model outperformed the human experts with 11 percent fewer false positives and 5 percent fewer false negatives.

Preliminary results of the study were presented earlier this month at Google's I/O developer conference. Though, at the time, Lily Peng, MD, PhD, a member of the research team and a product manager for Google Brain AI, described their findings as "promising," the technology is still largely experimental, and will not be available for widespread clinical use for quite some time.

More articles about AI:
Facebook CTO admits AI may not be the panacea he believed it was
Dr. Eric Topol: Integrating AI into medicine will 'restore the human mind'
Facebook releases disease prevention maps to fight flu, malaria outbreaks

Copyright © 2022 Becker's Healthcare. All Rights Reserved. Privacy Policy. Cookie Policy. Linking and Reprinting Policy.


Featured Whitepapers

Featured Webinars