Page 9 - ARNM-3-3

Advances in Radiotherapy &

Nuclear Medicine

EDITORIAL
Navigating promise and pitfalls: Real-world

deployment of artificial intelligence in
mammography

Mehrshad Bakhshi*

Faculty of Medicine and Health Sciences, McGill University, Montreal, Quebec, Canada

We aim to provide a review of the article “Nationwide Real-World Implementation of Artificial
Intelligence (AI) for Cancer Detection in Population-Based Mammography Screening”
published in Nature Medicine in January 2025. This work builds on previous literature
1
concerning the use of AI in breast cancer screening. Breast cancer screening programs face
numerous challenges. Notably, there is a need for improved sensitivity and specificity in
diagnostic performance of mammography screening. Efficiency is another metric that could
be improved, given the intensive workload requirements of these nationwide screening
programs. This study aimed to compare the performance of AI-supported double reading to
standard double reading, specifically focusing on breast cancer detection rate and recall rate.
The study was conducted within the framework of Germany’s breast cancer screening
program, which targeted women aged 50 – 69 years. In this program, mammograms
were independently assessed by two radiologists. If either radiologist deemed the case
suspicious, a consensus conference was held. If suspicion persisted, the patient was
recalled for further investigations, often including a biopsy. This observational study
had two arms. Cases were assigned to the AI arm if radiologists used an AI-supported
viewer; otherwise, cases were assigned to the control group. The Vara MG AI system was
used, employing a deep convolutional neural network trained and validated on a dataset
*Corresponding author: of over two million mammograms.
Mehrshad Bakhshi
(mehrshad.bakhshi@mail.mcgill.ca) A total of 461,818 women formed the analytic sample, with 119 radiologists assessing
Citation: Bakhshi M. Navigating the mammograms. In terms of breast cancer detection rate, the use of the AI-supported
promise and pitfalls: Real-world program led to a 17.6% (95% confidence interval: 5.7 – 30.8%) higher detection rate,
deployment of artificial intelligence diagnosing one extra cancer per 1,000 women screened. There was no significant
in mammography. Adv Radiother
Nucl Med. 2025;3(3):1-2. difference in recall rate between groups. However, the positive predictive values of both
doi: 10.36922/ARNM025180019 recall and biopsy were statistically superior in the AI-assisted group.
Received: April 28, 2025 This study represents the largest to date on the use of AI in breast cancer screening. The
Accepted: May 7, 2025 AI decision referral approach is notable for reducing automation bias. The main finding
was that the program improved cancer detection rates. However, some considerations
Published online: May 21, 2025
must be taken when interpreting these findings. It was found that the additional cancers
Copyright: © 2025 Author(s). detected were low-grade, raising concerns about the potential for overdiagnosis. Hence,
This is an Open-Access article
distributed under the terms of the follow-up studies will be needed to clarify these issues. There were also several limitations.
Creative Commons Attribution The observational design limits the strength of the findings. In addition, reader behavior
License, permitting distribution, bias was present, as radiologists had the option to choose whether to use the AI decision
and reproduction in any medium,
provided the original work is tool. Finally, we noted potential conflicts of interest. Most of the co-authors in this
properly cited. publication are stakeholders at the company that owns the AI model, and they also
Publisher’s Note: AccScience contributed significantly to the study design, data collection, and writing of the report.
Publishing remains neutral with
regard to jurisdictional claims in Conflict of interest
published maps and institutional
affiliations. The author declares no conflict of interest.

Volume 3 Issue 3 (2025) 1 doi: 10.36922/ARNM025180019