Artificial intelligence (AI) has emerged as a promising tool in the field of breast cancer screening, according to a recent study published in The Lancet Oncology journal. The study found that AI-supported mammography analysis is as effective as two breast radiologists working together and can reduce the screen-reading workload by half.
The research involved a safety analysis of a randomized controlled trial conducted with over 80,000 Swedish women. The use of AI in mammography screening has the potential to alleviate the burden on radiologists, who are already in scarce supply in many countries like the United Kingdom and Sweden.
Breast cancer screening through mammography plays a crucial role in detecting the disease at an early stage, making treatment more effective and reducing mortality rates. However, a significant percentage of interval cancers go undetected, and many suspicious findings turn out to be benign or non-cancerous.
To ensure a high level of sensitivity in breast cancer detection, European guidelines recommend involving two radiologists to screen mammograms. However, this practice is hindered by the shortage of breast radiologists and the lengthy training process required to become proficient in mammogram interpretation.
The study suggests that AI could serve as an automated second reader for mammograms, reducing the workload of radiologists and improving the accuracy of breast cancer screening. AI-based mammography provides computer-aided detection marks that highlight suspicious features, reducing the likelihood of false negative results. However, more robust evidence is needed from further trials to fully assess the effectiveness of this approach.
During the study, more than 80,000 women between the ages of 40 and 80 underwent mammogram screening across four sites in southwest Sweden. Some mammograms were analyzed using an AI-supported reading system before being reviewed by one or two radiologists, while others received standard analysis without AI.
The researchers compared early screening performance and the screen-reading workload between the intervention group (mammograms analyzed by radiologists after AI analysis) and the control group (mammograms analyzed without AI by two radiologists). The intervention group set a minimum acceptable limit for clinical safety, with a cancer detection rate of above three cancers per 1,000 screened women.
The AI system analyzed the mammography images and assigned a risk score from one to ten. If the risk score was less than ten, a radiologist conducted further analysis, and if it was ten, two radiologists reviewed the image. In 0.8% of cases, the AI failed to provide a risk score, leading to referral to standard care or double reading.
The study found that the average recall rates were 2.2% for AI-supported screening and 2% for standard double reading without AI. Furthermore, AI-supported screening detected 41 more cancers per 1,000 women than standard screening. It also reduced the screen-reading workload by 44%, with 36,886 fewer screen readings by radiologists in the AI-supported group.
While the study shows promising results for AI-supported mammography analysis, it acknowledges several limitations. The analysis was conducted at a single center and only involved one type of mammography device and AI system. Technical factors may also impact the system’s performance. Therefore, further research is needed to understand the implications of incorporating AI into mammography screening fully.
Despite the potential benefits of AI in breast cancer screening, researchers emphasize the importance of conducting new trials and program-based evaluations to address the shortage of radiologists in many countries. Additional research is necessary to determine the impact of AI on patient outcomes, cost-effectiveness, and the detection of interval cancers.
In conclusion, AI-supported mammography analysis has shown promising results in reducing radiologists’ workload and improving the accuracy of breast cancer screening. However, further research is needed to validate these findings and address the challenges associated with implementing AI technology in mammography screening programs.
