Breakthrough AI method to select women for MRI enhances breast cancer detection and outperforms traditional methods in landmark study
New research shows that an AI-based method for selecting women for additional screening with magnetic resonance imaging (MRI) can significantly improve the detection of breast cancer missed by traditional mammography. The randomized clinical trial ScreenTrustMRI, led by Fredrik Strand at Karolinska University Hospital, demonstrates how innovative AI technology could potentially revolutionize current breast cancer screening practices.
Mammography is the most widely used method for breast cancer screening and has been shown to reduce mortality by early detection. Despite this, approximately 30 percent of cancer cases among participants in biannual screening are detected due to symptoms that arise between screening intervals. These cancers are often more aggressive and have worse prognosis.
Therefore, there is a need for complementary technical solutions. The randomized clinical trial ScreenTrustMRI, led by Fredrik Strand at Karolinska University Hospital and published in Nature Medicine, investigated a newly developed AI technology to assess mammogram images and identify women at high risk for missed cancer. Among the women who received a high AI-based score (the top 6.9 percent), half were randomly offered to participate in an additional MRI examination. The AI method has been developed in collaboration between research groups at KI and KTH and funded by Medtechlabs, the Breast Cancer Association, among others. Through deep learning, the AI technology may capture more complex image patterns compared to the current method based on mammographic density which is usually a choice between four categories as assessed by a radiologist or computer software. A total of 36 cancers were detected among 559 women who had previously received a clean bill of health from standard mammography.
"We had of course hoped that the AI method we developed would be good for selecting who would need an additional MRI examination, but that it would lead to us detecting such a large number of missed cancers, we could not have dreamed of," says Fredrik Strand, radiologist at Karolinska University Hospital and researcher at the Department of Oncology-Pathology at Karolinska Institute.
Compared to results from a prior trial based on mammographic density, the AI method was around four times more effective for detecting cancer, with 64 detected cases per 1,000 MRI examinations compared to 16.5 cases per 1,000 in the previous trial. This AI-based method proved to be very promising for improving the detection of invasive and multifocal cancer forms, underscoring its potential to complement traditional mammography. By focusing on approximately 7-8 percent of screening participants, this method can make it economically sustainable to use MRI as an additional examination, with a cost per detected cancer case comparable to mammography.
"Now that the study is completed, we must pause the method because it needs approval from the European Medicines Agency for routine use. Additionally, the software needs to be packaged and quality assured to become a product, and for this, we have received continued funding from the Wallenberg Foundations," says Fredrik Strand.
The results from the ScreenTrustMRI study offer hope that AI technology can play a crucial role in the future of breast cancer screening. By integrating AI-based assessments into the existing screening process, more cancer cases can be detected early, ultimately saving more lives.
Karolinska University Hospital
Press Officer
+46 8 517 740 10
presstjanst.karolinska@regionstockholm.se
Karolinska is one of Europe's largest university hospitals and together with Karolinska Institutet we have a leading role within the field of medical breakthroughs. Our aim is to always put the patient first by providing the best possible medical expertise, treatment and care. Through innovation and active collaboration with industry and academia, we are committed to being internationally prominent in medicine, research and education.
