Advantages and disadvantages of artificial intelligence in BreastScreen Norway
Through retrospective studies we will investigate advantages and disadvantages of using artificial intelligence in the assessment of screening mammograms in BreastScreen Norway.
Background
BreastScreen Norway is a public health service offering screening for breast cancer every other year for women aged 50 to 69. The aim of the program is to detect breast cancer at an early stage, so that fewer women will die from the disease.
Most of the women attending BreastScreen Norway do not have breast cancer. Radiologists working in the screening program thus spend much of their time assessing mammograms with no signs of breast cancer.
In later years several studies have shown that artificial intelligence (AI) can potentially improve today's screening program, for example as a support for the radiologists in image assessment. By AI we mean digital systems trained by analyzing large amounts of data over time, through which they have learned to recognize patterns in mammograms that could be signs of breast cancer.
The Cancer Registry of Norway at NIPH have initiated several studies to test different AI systems in the assessment of screening mammograms in BreastScreen Norway. In this project, we will use mammograms already assessed by radiologists, so-called retrospective testing.
Purpose
The aim of this project is to explore advantages and disadvantages of using AI in the assessment of screening mammograms in BreastScreen Norway. The long-term goal is to offer an improved screening service to women participating in the program.
Substudies will be performed as retrospective cohort studies, where results from several AI systems will be compared with the radiologists’ assessments. The studies use data from screening examinations performed at different breast centres in Norway, see below.
The results will provide knowledge about whether AI systems are of such quality that they can be used in BreastScreen Norway. The results will also be important when designing and performing studies where AI is used in a real screening situation (prospective testing), and to plan a future implementation in the program.
Data
The AI tools will use image data from screening mammograms in BreastScreen Norway to assess risk of breast cancer. In the studies, the Cancer Registry of Norway, NIPH, will register and use the information that the AI tools provide about the mammograms. This includes a score that indicates the probability of breast cancer being present in the images.
The studies will also include screening information from examinations performed at various breast centers in BreastScreen Norway. Screening information is information about attendance and results of the screening examination, including the radiologists' assessments.
Image data and screening information are obtained from the following breast centers and time periods:
- Western Norway Regional Health Authority:
- Haukeland University Hospital, 2008-2022
- Stavanger University Hospital, 2010-2019
- Førde Hospital Trust, 2017-2019
- Central Norway Regional Health Authority:
- St. Olavs hospital, 2009-2018
- Møre og Romsdal Hospital Trust, 2011-2018
- Northern Norway Regional Health Authority:
- University Hospital of North Norway, 2004-2018
- South Eastern Norway Regional Health Authority:
- Vestre Viken Hospital Trust, 2008-2019
- Akershus University Hospital, 2007-2021
- Vestfold Hospital Trust, 2004-2021
- Sørlandet Hospital Trust, 2010-2019
- Østfold Hospital Trust, 2008-2019
- Oslo University Hospital, 2005-2021
- Innlandet Hospital Trust, 2009-2019
The studies will only use data from women who have not refused storage of their personal information related to negative screening examinations in the Cancer Registry, in accordance with the Cancer Registry Regulation. The women will not be contacted regarding the studies, and it will not be possible to identify individuals in the published results.
Examples of sub-studies
We are conducting several sub-studies to shed light on advantages and disadvantages of using AI in the assessment of screening images. Examples of sub studies are:
- How does AI do risk assessment of screening examinations with normal results and screening examinations where breast cancer has been detected
- How does AI do risk assessment of screening examinations performed prior to a breast cancer diagnosis
- Comparisons of newer and older versions of AI tools in how they do risk assessment of screening examinations
- Is AI able to mark the "correct" area on the mammograms, i.e. where breast cancer has been detected
- How does AI tools classify mammographic density
Organisation
The Cancer Registry of Norway at NIPH is project leader and the responsible research institution. We are responsible for obtaining all necessary approvals, agreements with AI vendors, collect image data and screening information, installations of and images analyses in the AI systems, analyses of the data material, and publishing results.
The health trusts represented by the breast centers are collaborators in various sub-studies. They are the regional specialists in breast cancer screening and diagnostics. They contribute with image data for the studies, as well as radiologic expertise and knowledge of practical screening.
At Stavanger University Hospital and Østfold Hospital Trust there are currently two Ph.D. students, Henrik Wethe Koch (Stavanger University Hospital) and Marit Almenning Martiniussen (Østfold Hospital Trust), working on related studies.
The vendors of the different AI systems will assist with installation of the systems and training. The vendors do not have access to personal information from the cases included in the studies, and they cannot influence the published results.
Status per September 2025
The project has obtained all necessary approvals as well as agreements to test the AI tools from the three vendors ScreenPoint Medical, Lunit Inc. and Vara. The work of collecting image data, installing the AI tools and running the collected image data through these AI systems, is completed.
Most of of the sub-studies have started, while others are expected to finish soon.
So far, the work has resulted in 11 scientific publications, and more are in progress.
Based on the results of the first sub-studies, the Cancer Registry of Norway has started a study, called AIMS Norway, where AI is included as part of the image assessment in an otherwise normal screening situation.
Publications
Koch HW, Bergan MB, Gjesvik J, Larsen M, Bartsch H, Haldorsen IHS, et al. Mammographic features in screening mammograms with high AI scores but a true-negative screening result. Acta Radiol. 2025.
Martiniussen MA, Larsen M, Hovda T, Kristiansen MU, Dahl FA, Eikvil L, et al. Performance of Two Deep Learning-based AI Models for Breast Cancer Detection and Localization on Screening Mammograms from BreastScreen Norway. Radiol Artif Intell. 2025:e240039.
Hovda T, Larsen M, Bergan MB, Gjesvik J, Akslen LA, Hofvind S. Retrospective evaluation of a CE-marked AI system, including 1 017 208 mammography screening examinations. Eur. Radiol. 2025.
Gjesvik J, Moshina N, Lee CI, Miglioretti DL, Hofvind S. Artificial Intelligence Algorithm for Subclinical Breast Cancer Detection. JAMA Network Open. 2024;7(10):e2437402-e.
Larsen M, Olstad CF, Lee CI, Hovda T, Hoff SR, Martiniussen MA, et al. Performance of an Artificial Intelligence System for Breast Cancer Detection on Screening Mammograms from BreastScreen Norway. Radiol Artif Intell. 2024;6(3):e230375.
Koch HW, Larsen M, Bartsch H, Martiniussen MA, Styr BM, Fagerheim S, et al. How do AI markings on screening mammograms correspond to cancer location? An informed review of 270 breast cancer cases in BreastScreen Norway. Eur Radiol. 2024;34(9):6158-67.
Bergan MB, Larsen M, Moshina N, Bartsch H, Koch HW, Aase HS, et al. AI performance by mammographic density in a retrospective cohort study of 99,489 participants in BreastScreen Norway. Eur Radiol. 2024.
Larsen M, Olstad CF, Koch HW, Martiniussen MA, Hoff SR, Lund-Hanssen H, et al. AI Risk Score on Screening Mammograms Preceding Breast Cancer Diagnosis. Radiology. 2023;309(1):e230989.
Koch HW, Larsen M, Bartsch H, Kurz KD, Hofvind S. Artificial intelligence in BreastScreen Norway: a retrospective analysis of a cancer-enriched sample including 1254 breast cancer cases. Eur Radiol. 2023;33(5):3735-43.
Larsen M, Aglen CF, Lee CI, Hoff SR, Lund-Hanssen H, Lang K, et al. Artificial Intelligence Evaluation of 122 969 Mammography Examinations from a Population-based Screening Program. Radiology. 2022;303(3):502-11.
Larsen M, Aglen CF, Hoff SR, Lund-Hanssen H, Hofvind S. Possible strategies for use of artificial intelligence in screen-reading of mammograms, based on retrospective data from 122,969 screening examinations. Eur Radiol. 2022;32(12):8238-46.
About the project
Project leader: Solveig Hofvind
Project group at the Cancer Registry of Norway: Marthe Larsen, Åsne Holen, Marie Burns Bergan, Silje Sagstad, Nataliia Moshina
Financing: Parts of the project are funded by the Norwegian cancer society and the Norwegian breast cancer society through the Pink Ribbon campaign (#214931).
Project duration: March 1, 2019 – February 28, 2029
Ethical approvals: The project is approved by the Regional Committees for Medical and Health Research Ethics (REK), project number 2018/2574
Data protection: Information about data protection and privacy in the Cancer Registry of Norway (Norwegian).
Collaboration:
- Haukeland University Hospital
- Stavanger University Hospital
- Førde Hospital Trust
- Oslo University Hospital
- Vestre Viken Hospital Trust
- Vestfold Hospital Trust
- Sørlandet Hospital Trust
- University Hospital of North Norway
- Østfold Hospital Trust
- St. Olavs Hospital
- Innlandet Hospital Trust
- Møre og Romsdal Hospital Trust
- Akershus University Hospital
AI vendors