Use of artificial intelligence for image analysis in breast cancer screening (bmj.com)

Article summary

A study evaluated 36 AI systems for breast cancer screening and found that 34 of them were less accurate than a single radiologist, and all were less accurate than the consensus of two or more radiologists. The study's findings suggest that the current state of AI in radiology may not be as effective as claimed. The results have implications for the development and investment in AI-powered breast cancer screening tools. The study's methodology compared AI system results to radiology reports and follow-up biopsies.

Main themes

• AI accuracy in radiology
• Comparison to human radiologists
• Overstated research claims
• Investment in AI-powered tools
• Deep learning limitations
• Variable human accuracy
• Collaboration in radiology
• Real-world applications and reliability

What commenters say

• Some AI systems may be lucky rather than genuinely accurate, which could lead to overestimation of their effectiveness.
• The state of the art in deep learning may not be sufficient for radiology, resulting in overstated research claims.
• Investing in AI-powered radiology tools may not be wise due to the current lack of accuracy and reliability.
• The comparison between AI systems and human radiologists may not be straightforward, as human accuracy can also be variable and dependent on factors like experience and collaboration.
• The fact that some AI systems perform better than others does not necessarily mean they are reliable or effective in real-world applications.
• The study's findings do not necessarily prove that the task is impossible, but rather that current approaches may not be sufficient.
• The goal of AI in radiology should not be solely focused on beating human performance, but also on providing consistent and reliable results.
• The development of AI-powered radiology tools requires more rigorous testing and evaluation to ensure their accuracy and effectiveness.