Study: Crowdsourced AI faster than humans in segmenting cancerous lung tumors

A new study published in JAMA Oncology shows that artificial intelligence systems crowdsourced from data scientists can segment lung tumors as accurately as human experts can, and at much higher speeds.

Thirty-four contestants submitted 45 algorithms to Indianapolis-based data crowdsourcing platform Topcoder and researchers from the Dana-Farber Cancer Institute and Brigham and Women's Hospital, both in Boston, as well as Cambridge, Mass.-based Harvard Catalyst, Harvard Business School and the Laboratory for Innovation Science at Harvard.

After 10 weeks, researchers found that several of the AI solutions were able to perform the tumor segmentation work of a human expert not only with equal accuracy, but also 75 percent to 96.8 percent faster. According to the study, the algorithms replicated the segmentation work of an expert in the span of 15 seconds to two minutes, compared to the average of eight minutes spent on the task by humans.

Tumor segmentation is a critical and time-intensive task in the application of radiation therapy that typically requires highly trained experts to complete. As the study notes, there is currently a shortage of radiation oncologists trained in segmentation, so AI methods like those tested are necessary to "transfer expert skill and knowledge to under-resourced healthcare settings and improve the quality of radiation therapy globally."

More articles about AI:
IBM stops development, sales of Watson for Drug Discovery
Gilead, machine learning startup test AI's drug development potential
AI-driven apps are coaching digital-first millennial managers

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


Featured Whitepapers

Featured Webinars