Artificial Intelligence Detects 20% More Breast Cancers in Mammograms, According to Swedish Trial
Early results from a Swedish trial have shown that artificial intelligence (AI) is able to accurately detect 20 percent more breast cancers from mammograms than traditional screening by radiologists. This study, the first randomized controlled trial to examine the use of AI in breast cancer screening, comes at a time when the landscape for AI technology and its regulation is rapidly evolving.
The interim results of the trial, published in the Lancet Oncology, revealed that using AI-supported analysis of mammograms, in conjunction with one or two radiologists, was as effective as using two radiologists alone. Additionally, the use of AI led to a 20 percent increase in the detection of cancers. Furthermore, the workload for radiologists was significantly reduced, with a 44 percent decrease in time spent reading mammograms.
The trial, which is still ongoing, involved over 80,000 women in Sweden. Half of the participants had their mammograms analyzed by two radiologists without the assistance of AI, while the other half had their mammograms assessed by AI and a radiographer, except in cases where AI generated the highest risk score, in which two radiologists assessed the screening.
These results come at a time when there is increasing interest in the opportunities and risks associated with AI in medicine and other industries. While AI is being increasingly utilized in medical settings, concerns regarding the training and validation of algorithms, as well as potential biases and over-diagnosis, have been raised.
Meanwhile, the European Union (EU) is planning to introduce stringent regulations on the use of AI, and the European Medicines Agency is currently evaluating the risks and benefits of implementing AI in drug development.
Despite the positive findings of the trial, lead author Kristina Lång from Lund University in Sweden cautioned that the interim safety results alone are not enough to confirm whether AI is ready for implementation in mammography screening. Lång and her colleagues are awaiting further results from the trial that will determine if the use of AI reduces the number of cancers detected between screenings and if it is worth implementing on a larger scale.
Lång highlighted that the greatest potential of AI in this context is its ability to alleviate the burden on radiologists’ time, allowing them to assist more patients. However, Stephen Duffy, a professor of cancer screening at Queen Mary University of London, raised concerns about the possibility of AI over-detecting harmless lesions.
Not only is this study the first randomized trial exploring the use of AI in reading mammograms, it is also one of the first to focus on AI in the field of radiology. An editorial in European Radiology earlier this year stressed the importance of randomized controlled trials in monitoring the safety of AI systems, which can produce errors that are unpredictable and undetectable by human logic.