por
Gus Iversen, Editor in Chief | June 07, 2024
A new study published in Radiology demonstrates the effectiveness of AI in improving breast cancer screening. Conducted in Denmark, the study highlights significant enhancements in cancer detection rates and reductions in false positives when using AI-assisted screening.
The research, led by Andreas D. Lauritzen, Ph.D., from the University of Copenhagen and Gentofte Hospital, analyzed data from two groups of women aged 50 to 69 who underwent biennial mammography screening in Denmark’s Capital Region. The study compared results from two periods: before and after the implementation of AI in the screening process.
In the pre-AI group, comprising 60,751 women screened between October 2020 and November 2021, two radiologists reviewed each mammogram. In the post-AI group, 58,246 women screened between November 2021 and October 2022 had their mammograms initially analyzed by an AI system. Mammograms deemed normal by AI were single-read by one of 19 specialized breast radiologists, while those flagged for potential issues underwent a double-read by two radiologists with AI assistance.
The AI system, Transpara (v.1.7.1) from ScreenPoint Medical, was trained through deep learning to identify suspicious lesions and calcifications and contributed to a significant improvement in screening outcomes. The AI-assisted screenings detected more breast cancers (0.82% compared to 0.70%) and had a lower false-positive rate (1.63% versus 2.39%) than those without AI assistance. Additionally, the recall rate decreased by 20.5%, and the radiologists' reading workload was reduced by 33.4%.
Lauritzen emphasized the impact on radiologists, stating, “Population-based screening with mammography reduces breast cancer mortality but places a substantial workload on radiologists. AI can substantially reduce this workload while improving screening performance.”
The study also found that the positive predictive value of AI-assisted screening was higher (33.5% versus 22.5%), and a greater proportion of detected invasive cancers were one centimeter or smaller in size (44.93% versus 36.60%).
However, Lauritzen noted the need for further research to evaluate long-term outcomes and potential overdiagnosis. He also mentioned a limitation of the current AI system: it does not access previous mammograms for comparison, unlike radiologists. Addressing this could further enhance AI performance in the future.
The study underscores the potential of AI to enhance breast cancer screening programs, though Lauritzen cautions that breast cancer screening protocols vary by country, and results might differ in other regions.
Back to HCB News
