June 4, 2024 — Danish breast radiologists are using artificial intelligence (AI) to improve the accuracy of breast cancer screening and reduce false positive rates. The results of the study were published today. RadiologyJournal of the Radiological Society of North America (RSNA).
Mammograms are effective at reducing breast cancer mortality, but they do carry the risk of false-positive results, and in recent years, researchers have been studying the use of AI systems in screening.
“We believe that AI has the potential to improve screening performance,” said Andreas D. Lauritzen, PhD, a postdoctoral researcher at the University of Copenhagen and researcher at Gentofte Hospital in Denmark.
AI can also significantly reduce radiologists’ workload by triaging potentially normal test results and using them for decision support.
“Mass screening with mammography reduces breast cancer mortality but places a significant burden on radiologists who must read large numbers of mammograms, the majority of which do not require patients to be retested,” said Dr. Lauritzen. “The reading burden increases further when screening programs employ double reading to improve cancer detection rates and reduce false-positive retests.”
Dr. Lauritzen and his colleagues decided to compare workload and screening performance in two cohorts of women who were screened before and after the implementation of AI.
This retrospective study compared two groups of women aged 50 to 69 years who underwent mammography screening every two years in the Capital Region of Denmark.
In the first group, two radiologists read the mammograms of women who underwent screening between October 2020 and November 2021, before the introduction of the AI. In the second group, screening mammograms performed between November 2021 and October 2022 were first analyzed by the AI. Mammograms that were judged by the AI to be likely normal were read by one of the 19 full-time breast radiologists (referred to as single reading). The remaining mammograms were read by two radiologists with AI-assisted decision support (referred to as double reading).
The commercially available AI system used for screening was trained with deep learning models to highlight and evaluate suspicious lesions and calcifications in mammograms. All women who underwent mammography screening were followed for at least 180 days. Screen-detected invasive cancers and ductal carcinoma in breast cancer (DCIS) were confirmed by needle biopsy or surgical specimens.
In total, 60,751 women were screened without AI and 58,246 women were screened using an AI system. In the AI-implemented group, 66.9% (38,977) of exams were single-read and 33.1% (19,269) were AI-assisted dual-read.
Compared with screening without AI, screening with the AI system detected significantly more breast cancers (0.82% vs. 0.70%) and had a lower false positive rate (1.63% vs. 2.39%).
“In the group that received the AI test, the retest rate decreased by 20.5 percent and radiologists' reading workload was reduced by 33.4 percent,” Dr. Lauritzen said.
The positive predictive value of AI screening was also higher than screening without AI (33.5% vs. 22.5%).A higher proportion of invasive cancers detected were 1 cm or smaller in size in the AI group (44.93% vs. 36.60%).
“All screening performance measures improved except for node-negative rates, which showed no evidence of change,” Dr. Lauritzen said.
Dr. Lauritzen said further studies are needed to assess long-term outcomes and to ensure there is no increased overdiagnosis.
“Radiologists typically have access to a woman's past mammogram data, but AI systems don't,” he says. “This is something we want to address in the future.”
It is also important to keep in mind that not all countries follow the same breast cancer screening protocols and intervals. Breast cancer screening protocols in the United States are different from those used in Denmark.
For more information: Click here for details
