An interim safety analysis of a large randomized controlled trial published in The Lancet Oncology indicates that Artificial Intelligence (AI)-aided mammography can match the efficiency of two breast radiologists in detecting breast cancer, and almost halves the screen-reading workload, without an increase in false positives.
However, the complete study that aims to understand if AI interpretation reduces interval cancers (cancers detected between screenings), is not expected for several years.
The trial involved over 80,000 women in Sweden who were assigned in equal proportions to either AI-supported mammogram reading or standard analysis performed by two radiologists.
This is a part of the Mammography Screening with Artificial Intelligence (MASAI) trial, which is the first of its kind.
Interim Findings
The initial results are promising but should not be taken as a final green light for implementing AI in mammography screening, cautions Dr. Kristina Lång from Lund University, Sweden, the lead author of the study.
According to her, there are further questions about the patient outcomes, detecting interval cancers often missed by traditional screening, and the cost-effectiveness of the technology.
In the AI-assisted analysis, an AI system initially analyzed the mammograms and predicted cancer risk on a scale of 1 to 10. Based on this risk score, one or two radiologists further analyzed the images.
The AI also provided marks to assist radiologists in accurately interpreting images. In cases where AI failed to provide a risk score, those were referred for standard care.
Impact on Cancer Detection and Radiologists’ Workload
AI-supported screening detected one additional cancer per 1,000 screened women compared to the standard double reading.
This resulted in a cancer detection rate of six per 1,000 screened women for AI-supported analysis, compared to five per 1,000 for standard double reading without AI.
Importantly, AI-assisted screening resulted in a 44% reduction in the screen-reading workload for radiologists, which corresponds to a potential saving of 4.6 months for a single radiologist reading about 50 mammograms per hour.
This, according to Dr. Lång, could allow radiologists to focus more on advanced diagnostics and reduce patient waiting times.
Limitations and Future Considerations
Despite the promising results, the researchers pointed out several limitations.
The trial was conducted at a single center and was limited to one type of mammography device and one AI system, which may affect the generalizability of the results.
Furthermore, as the final decision to recall women was made by radiologists, the results depend on their performance and expertise.
The team will continue the trial to evaluate primary outcomes such as whether AI-assisted mammography can reduce interval cancers.
Dr. Nereo Segnan, an expert not involved in the study, highlighted the need for caution given the possibility of AI over-diagnosing or over-detecting indolent lesions.
He posed an important research question for future studies: “Is AI, when appropriately trained, able to capture relevant biological features such as the capacity of tumors to grow and disseminate?”
If you care about health, please read studies about supplements that could strongly reduce cancer death, and blood pressure drugs plus chemotherapy could reduce triple-negative breast cancer.
For more information about health, please see recent studies about a new way to halt excessive inflammation, and results showing dietary supplement that fights resistance in breast cancer.
The study was published in The Lancet Oncology.
Follow us on Twitter for more articles about this topic.
Copyright © 2023 Knowridge Science Report. All rights reserved.
