Background and purpose
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.
The vast majority of women who attend BreastScreen Norway have normal mammograms with no signs of breast cancer. The radiologists working in BreastScreen Norway thus spend a lot of time evaluating normal mammograms, i.e. where there are no signs of breast cancer.
This project will investigate whether artificial intelligence can be used to support the interpretation of screening mammograms so that radiologists can find the right breast cancer cases more quickly and easily.
The long-term goal is to increase the quality of BreastScreen Norway by obtaining more precise and effective screening, diagnostics and treatment of women.
The machine learning system we will test, Transpara, has been developed by ScreenPoint Medical B.V., and is FDA approved. The system will be tested through a retrospective cohort study, where the goal is to compare performance of this machine learning system with single interpretation of mammograms performed by radiologists in BreastScreen Norway.
The machine learning system will evaluate screening mammograms (image data), giving each image a score that indicates the likelyhood of breast cancer - similar to how the radiologists normally score mammograms in BreastScreen Norway. Then, the interpretation score from the machine learning system can be compared to the radiologists' interpretation scores, to find out whether the machine learning system will detect the same breast tumors as the radiologists do, and whether the system detects breast tumors that the radiologists have not seen.
Later, if the analyzes on the retrospective data show that the machine learning system performs as well, or better than, the radiologists, the Cancer Registry of Norway will conduct a follow-up randomized controlled trial for further investigating the advantages and disadvantages of the machine learning system. We will describe the randomized controlled trial closer to start-up.
The project will use data from screening examinations already conducted in BreastScreen Norway, consisting of mammograms (image data) and information on radiological assessments, including any positive and negative findings in the screening program (screening data).
The study population is women who have been screened with digital mammography at breast centers in Østfold, Agder, Trøndelag and Møre og Romsdal in BreastScreen Norway from 2008 and later. The project will only use data from women who have allowed that their personal data related to negative screening results be permanently stored at the Cancer Registry, in accordance with the Cancer Registry Regulations (Kreftregisterforskriften).
The retrospective part is a register-based study and the participants will not be contacted about the project. It will not be possible to identify individuals from any published study results.
The Cancer Registry is data controller in the project. We are responsible for obtaining the necessary approvals, data collection, use and testing of the machine learning system, as well as analyzing the data material.
ScreenPoint Medical is the company that has developed the machine learning system to be tested, Transpara. They will assist with the installation of Transpara, as well as training the staff in how to use the system.
The breast centers in Østfold, Agder, Trøndelag and Møre og Romsdal are the regional specialists on breast cancer screening and diagnosis. They contribute with image data in the project and expertise about practical aspects of screening.
Necessary approvals have been obtained, the machine learning system has been installed at the Cancer Registry of Norway and image data from the breast centers in Trøndelag and Møre og Romsdal have been collected at the Cancer Registry. The process of making Transpara interpret the data is ongoing at the Cancer Registry, and we can soon start the initial analyses. Data collection from the breast centers in Østfold and Agder is still ongoing.