Using AI to identify more breast cancers
Clinics and researchers are currently working to develop a method for using Artificial Intelligence, AI, in mammography screening. The aim of the project is for AI-based computer support to help perform an initial sorting of images obtained from screening.
How can AI help in mammography screening?
"Screening reduces mortality from breast cancer by about 30 per cent. We want to see if we can detect more early-stage breast cancer tumours using AI to support screening. This would increase the chances of treating more cancers in time. We also want to see if AI support can help to improve the work-flow in radiology", says Fredrik Strand, radiologist and project manager for AI as decision-making support in mammography screening.
Fredrik Strand believes that where it now takes two radiologists to check screening images, it may be that in certain cases, the analysis work could be done with AI support and one radiologist. This could also free up expertise – the specialists would be able to devote themselves more to following up cases requiring further analysis.
How will this happen?
"The idea is that the radiologist will use AI support to identify the results where there is a high risk of a hidden tumour. Our plan is to evaluate how MRI scanning of women at high risk influences the early detection of breast cancer. We will also investigate how it affects the number of women called back to provide tissue samples, and how many of these are negative or positive."
As part of a first stage, AI support, or the algorithm, will be evaluated using existing data. As part of the project, over one million screenings performed at Karolinska between 2008 and 2015 were collected. These are now being analysed and linked to information from the cancer register. The development work will also take in any unusual breast cancer types.
"There will be a lot of healthy images, as most people who undergo screening do not have any tumours, so it is important to use cancer register data in order to obtain comprehensive and reliable image analysis material. We also need a lot of examples of unusual breast cancers. And in order to cover multiple unusual variants, we are also planning to include image analysis material from other parts of the country. In this way we will produce a broad and relevant body of documentation on which to develop the decision-making support that can then be used in screening."
Can AI help in the earlier detection of breast cancer?
We believe it can. This is an exciting project and we will do our best to help develop mammography screening and to ensure that more cancers are detected at such an early stage as to be treatable, so that women can be free from this disease.
Mammography-based screening reduces mortalities from breast cancer by about 30 per cent. There is also increasing awareness that mammography does not work equally as well for all women, and that therefore additional examinations might be required in some women. At the same time however there is a specific shortage of breast radiologists. These factors feature prominently in the background to the project.
As part of mammography screening, a large part of radiologists' work involves finding images that are relatively easy to analyse.
"And this is where AI as decision-making support comes in. By using AI, a computer can be given the task of sorting out those images that are obviously easy to assess. The project aims to free up resources and allow radiologists to look at images that are harder to assess. As a result, it will be possible to screen more women faster. The aim is that this will become a tool that can be used by breast radiologists", says Tomas Borgegård, of Innovationsplatsen, who was involved in the start-up phase of the project that is now being carried out in the breast radiology activity.
Ai as decision-making support in mammography – About the project:
The idea is to use Artificial Intelligence, AI, to improve quality and to detect more breast cancers faster. There is a shortage of radiologists and breast radiologists, and support in decision-making can help to free up resources.
The project involves analysing over a million mammography examinations from Karolinska in the period 2008-2015. Researchers will then link image analysis from investigations to data from the cancer register. Image analysis from mammography screening together with register data in the cancer register is an important element in that the content for AI support will become more relevant.
There is also a plan to obtain information and images for unusual cancers from all over the country in order to ensure that the algorithm or AI-supported decision-making covers many different types of breast cancer.
Existing data is being used in the assessment of AI-supported decision-making, or the algorithm. The project also includes collaboration and evaluation of the algorithms together with the University of California, San Francisco, amongst other institutions.
A future phase will then see the method tested in parallel with ordinary investigations in order to compare results from the two methods. Another phase of the project involves finding out what those who perform mammography investigation think of AI-supported screening analysis.
Once the method has been optimised, it will be integrated into breast cancer screening. The project is being conducted over a 2-year period with the financial support of Vinnova.
Text: Susanne Bergqvist. Photo: Robert Sundberg.
Karolinska University Hospital has the lowest mortality rate for cardiac surgery
Karolinska University Hospital is among the best hospitals in the world in terms of survival after cardiac surgery.x
Grand display of color across Karolinska University Hospital – part of global manifestation for rare diagnoses
“We want to raise awareness of rare diagnoses in this spectacular way. We’re doing this within the scope of one of several national and international networks working to generate ...