Can AI Algorithms Accurately Diagnose Breast Cancer?
BACKGROUND AND PURPOSE:
Freeman et al. (BMJ, 2021) examined the accuracy of artificial intelligence (AI) in detecting breast cancer in mammography screening practice
Studies that reported test accuracy of AI algorithms, alone or in combination with radiologists, to detect cancer in women’s digital mammograms
The reference standard was biopsy with histology or follow-up (for screen negative women)
Cancer type detected
12 studies | 131,822 screened women
There were no prospective studies
Overall, studies were of poor methodological quality
3 retrospective studies compared AI with clinical decision of the original radiologist
79,910 women | 1878 detected cancers
94% of the AI systems were less accurate than a single radiologist
All were less accurate than the consensus of two or more radiologists
5 smaller studies evaluated AI as standalone to replace radiologist or as a reader aid
1086 women | 520 detected cancers
All evaluated AI systems were more accurate than a single radiologist reading a test set in the laboratory
High risk of bias and low generalizability to a real world setting as radiologists examined the mammographic images in a laboratory setting
Current evidence does not allow for conclusions to be drawn regarding accuracy of AI in breast cancer screening
No AI algorithm was able to match the efficacy of two radiologists’ consensus diagnosis
Prospective studies will be required to robustly assess the efficacy of AI in clinical practice
The authors state
Evidence is insufficient on the accuracy or clinical effect of introducing AI to examine mammograms anywhere on the screening pathway
It is not yet clear where on the clinical pathway AI might be of most benefit, but its use to redesign the pathway with AI complementing rather than competing with radiologists is a potentially promising way forward
OBG Project CME requires a modern web browser (Internet Explorer 10+, Mozilla Firefox, Apple Safari, Google Chrome, Microsoft Edge). Certain educational activities may require additional software to view multimedia, presentation, or printable versions of their content. These activities will be marked as such and will provide links to the required software. That software may be: Adobe Flash, Apple QuickTime, Adobe Acrobat, Microsoft PowerPoint, Windows Media Player, or Real Networks Real One Player.
Disclosure of Unlabeled Use
This educational activity may contain discussion of published and/or investigational uses of agents that are not indicated by the FDA. The planners of this activity do not recommend the use of any agent outside of the labeled indications.
The opinions expressed in the educational activity are those of the faculty and do not necessarily represent the views of the planners. Please refer to the official prescribing information for each product for discussion of approved indications, contraindications, and warnings.
Participants have an implied responsibility to use the newly acquired information to enhance patient outcomes and their own professional development. The information
presented in this activity is not meant to serve as a guideline for patient management. Any procedures, medications, or other courses of diagnosis or treatment discussed or suggested in this activity should not be used by clinicians without evaluation of their patient’s conditions and possible contraindications and/or dangers in use, review of any applicable manufacturer’s product information, and comparison with recommendations of other authorities.
Jointly provided by
NOT ENOUGH CME HOURS
It appears you don't have enough CME Hours to take this Post-Test. Feel free to buy additional CME hours or upgrade your current CME subscription plan