Artificial intelligence in BreastScreen Norway - a randomized controlled trial

In this study, we will investigate whether artificial intelligence in combination with radiologists is as good as or better at detecting breast cancer than the current standard procedure in BreastScreen Norway, where two radiologists evaluate the images.
Last updated:

Background and purpose

BreastScreen Norway offers screening mammography to all women aged 50 to 69 every two years. The aim is to detect breast cancer in an early stage of the disease, so that fewer women die from breast cancer.

Currently, all mammograms in BreastScreen Norway are assessed by two radiologists (doctors specialized in interpreting radiological images), who both interpret the images without knowing what the other has concluded (independent double reading).

However, very few participants in the screening program have signs of breast cancer, and radiologists therefore spend a lot of time assessing normal mammograms.

Artificial intelligence is a promising technology that has been introduced in several areas of medicine in recent years, including in the assessment of mammograms. These programs can be trained to detect changes in mammograms that could be breast cancer.

Studies have demonstrated very good results for the use of artificial intelligence in assessment of mammograms, as the programs find the same cancers as the radiologists, and also find them in an early stage of disease development. Studies also show that very few breast cancers are overlooked by these programs, fewer than three per 100. These studies have only used so-called retrospective data, meaning that the programs have analyzed the images after the radiologists have made their conclusions.

Time has now come for studies where artificial intelligence is included as part of the mammogram assessment in an otherwise normal screening situation. This study, called AIMS Norway (Artificial Intelligence in Mammography Screening in Norway) is one such study, where we will test the CE-marked artificial intelligence program Transpara in the assessment of screening images in BreastScreen Norway. CE-marking means that the program meets the safety requirements for medical equipment set by the EU.

The purpose of the study is to find out whether the artificial intelligence program, when used in a normal screening situation and in combination with one or two radiologists, is as good as or better at identifying breast cancers on mammograms than the standard procedure with independent double reading. This could mean a significant reduction of the need for radiological resources in BreastScreen Norway, without reducing the quality of the screening program.

Implementation

This is a randomized controlled trial where half of the study participants will have their mammograms analyzed by the artificial intelligence program in combination with one or two radiologists (study group). The other half will have their mammograms assessed by two radiologists according to standard procedure in BreastScreen Norway (control group). Results will be compared between the two groups to see if artificial intelligence is as good as or better than the standard procedure.

In the study group, the mammograms will be assessed by one or two radiologists, depending on the artificial intelligence program's assessment. Examinations that the program classifies as having a low risk of breast cancer will be assessed by one radiologist, while the remaining with medium or high risk will be assessed by two (standard procedure).

In the control group, the images will be assessed by two radiologists, according to standard procedure. The artificial intelligence program will also analyze the images in the control group, but these results will not be available until after data collection has ended. The results will be registered in the Cancer Register of Norway for later analyses in the study. 

Assigning participants into the two groups is done by randomization, and takes place automatically after signing a consent to participate. This means it is random whether a woman ends up in the study or control group, and that neither the women, the radiographers at the screening unit nor the radiologists who assess the images can influence the distribution.

The study will be conducted at the breast centers in the Western, Central and Northern Norway Regional Health Authorities. Everyone attending BreastScreen Norway at the included breast centers will be asked to participate in the study. Those who wish to participate must give their informed consent when attending screening.

Attendance and examination in BreastScreen Norway, as well as any follow-up examinations and treatment, are carried out as usual, regardless of whether the woman participates in the study or not, and to which group she is randomized.

Data

The study will only include information about women who have consented to participate. The Cancer Registry of Norway will register and use the information that the artificial intelligence program provides about their mammograms. This includes a score indicating the probability of breast cancer being present in the mammogram.

In addition, the study will use information related to attendance and findings on the mammograms. This includes information about the radiologists' assessments, including the results of the screening examinations (screening information). The mammograms, which are stored in the hospitals patient records (in Norwegian: pasientjournal), will be used to ensure the quality of the study.

It will not be possible to identify individuals in any published results.

Organization

The Cancer Registry of Norway is the coordinating institution with Solveig Hofvind as project manager. The study is a multi-center study, and both the Cancer Registry and the cooperating health trusts are responsible research institutions in accordance with the Health Research Act. The parties have an independent responsibility for organizing and carrying out the part of the study performed in their own institution, and that this is done in accordance with relevant regulations and formal approvals.

The Cancer Registry is the data controller in the study. This entails responsibility for obtaining the necessary approvals, collecting data, monitoring the study, quality assurance and analysis of the data material, as well as publishing and reporting results from the study.

The study will be carried out at all breast centers in the Western, Central and Northern Norway Regional Health Authorities. This includes 14 screening units in addition to mammography buses. The mammograms will be assessed by radiologists at the eight included breast centers. The IT departments in the three health regions will assist in the installation and setup of the artificial intelligence program in collaboration with the Cancer Registry of Norway and the PACS supplier (program for displaying radiological images).

ScreenPoint Medical (Netherlands) is the company that has developed the artificial intelligence program to be used in the study, Transpara. They will assist with installation and train the staff in how to use the system.

Status

Start-up of the study is planned in the Western Norway Regional Health Authority in the first half of 2024. The Central and Northern Norway Regional Health Authorities will be included gradually.

Recruitment of study participants will continue until the study has reached a number that provides sufficient statistical power to show whether artificial intelligence in combination with radiologists is as good as or better in assessing mammograms as radiologists alone.