Approximately 250,000 mammography examinations are conducted annually as part of BreastScreen Norway [1]. More than 99% of these are negative for breast cancer, which means that large radiological resources are spent on interpreting normal images [2]. Today, all screening examinations are interpreted by two independent breast radiologists, and if at least one of them finds suspicious changes, the images are further discussed in a consensus meeting. Here, radiologists decide whether the woman should be recalled for further assessments.
Retrospective studies using artificial intelligence (AI) to classify examinations as negative or positive have shown very promising results[3,4,5]. However, there is a need for prospective studies to define the role of AI as a support or potential replacement for radiologists in screening programs.
AIMS is a collaborative project between UNN, Nordlandssykehuset (NLSH), and the Cancer Registry of Norway, aiming to conduct a randomized controlled trial (RTC) to test an AI tool as an integrated part of the breast screening program. To measure the effectiveness of the AI tool, the examinations are divided into two groups: a control group that follows the standard procedure with independent double reading, and a study group where the AI tool is integrated into the process. The way the AI tool is integrated is that each survey receives an AI-based risk score. Examinations with a high score will still be interpreted by two independent radiologists (standard procedure), while examinations with a low risk score will only be interpreted by one radiologist.
By comparing the number of detected breast cancer cases in the two groups, the project will investigate whether the interpretation of screening examinations with an AI tool combined with one or two radiologists can replace the current independent double reading without compromising the breast cancer detection rate. If AI can free up resources without affecting screening quality, it will provide radiologists with more time for complex breast cancer diagnoses, in addition to helping with the shortage of breast radiologists, which is a significant challenge in Northern Norway.
The project is also part of a larger multi-center study, administered by the Norwegian Cancer Registry and involving breast centers in Helse Nord, Helse Midt, and Helse Vest. The breast centers at UNN and NLSH are involved in the study. Two people will be employed in 30% positions at NLSH and UNN, respectively, to coordinate the data collection. The Cancer Registry will be responsible for project administration and guidance, data collection/establishment of databases, quality assurance, and analyses. SPKI contributes with project management and AI competence.
[1] https://www.kreftregisteret.no/screening/mammografiprogrammet/Nokkeltall/
[2] Larsen, Marthe, et al. "Artificial intelligence evaluation of 122 969 mammography examinations from a population-based screening program." Radiology 303.3 (2022): 502-511.
[3] Rodriguez-Ruiz, Alejandro, et al. "Stand-alone artificial intelligence for breast cancer detection in mammography: comparison with 101 radiologists." JNCI: Journal of the National Cancer Institute 111.9 (2019): 916-922.
[4] Salim, Mattie, et al. "External evaluation of 3 commercial artificial intelligence algorithms for independent assessment of screening mammograms." JAMA oncology 6.10 (2020): 1581-1588.
[5] McKinney, Scott Mayer, et al. "International evaluation of an AI system for breast cancer screening." Nature 577.7788 (2020): 89-94.
