Artificial Intelligence (AI) can improve breast cancer detection and reduce workload on physicians, according to a new study featuring over 461,000 women. The research published in the journal Nature Medicine looked at integrating an AI tool as part of a national screening program for women without symptoms of breast cancer in Germany.
The majority of previous studies using AI for cancer diagnosis have been done on existing data after people have actually been diagnosed. This is a necessary step to validate the use of new diagnosis methods, basically to see how the AI algorithm would have done in terms of accuracy and sensitivity if it had been used, without any risk to the patients. But the new study used AI prospectively in a national screening study in Germany of women between the ages of 50-69 who had no symptoms, hoping to pick up cases of breast cancer at early stages.
The women were screened between 2021 and 2023 and split into two groups. In one group, the screening data was analyzed by two expert radiologists. In the other group, two radiologists also analyzed the data, but at least one of them also used an AI tool to assist.
In total, 2,881 women were diagnosed with breast cancer with the AI group having a 6.7% higher detection rate. When adjusting for potential variables in the groups, such as the age of the women diagnosed, this difference increased to 17.6% higher diagnoses in the AI- assisted group.
"We actually wanted to show with the study that AI reporting is equivalent to human reporting," explains Alexander Katalinic, MD, study leader and director of the Institute for Social Medicine and Epidemiology at the University of Lübeck and the UKSH, Lübeck Campus in Germany. "But the results surprised us positively: AI actually significantly improves the breast cancer detection rate," added Katalinic, in a press release.
The study also looked at "false positives" - when a suspicious scan is labelled as potentially cancerous triggering further investigations, when in reality, it isn't. AI cancer detection in the past has sometimes had a higher false-positive rate and these can lead to a lot of distress for patients and extra costs for healthcare systems. Both the AI-assisted and unassisted radiologists had a similar false-positive rate.
"The PRAIM study shows the enormous potential of AI to improve screening programs worldwide," said Stefan Bunk, CTO of the AI company Vara who designed the AI tool. "This evidence will take the discussion about the integration of AI into healthcare systems to a new level," added Bunk.
There is hope that AI tools may eventually be able to reduce the workload on doctors involved in diagnosing cancers, as well as improving the accuracy and reducing false positives. Most studies so far have been done with clinicians still validating all of the AI results to mitigate the risk of additional harm to patients caused by missed diagnoses or false positives.
"We hope that the higher hit rate with AI can further improve the prognosis for women with breast cancer. We will investigate this in the future," said Katalinic.