AI can replace every third specialist in breast cancer screening

Reducing the workload of radiologists when it comes to the interpretation of medical images is one of the most obvious successes of AI in healthcare. Illustration: Lars Wallin/Region Hovedstaden

Since November, the Capital Region of Denmark has been testing an AI solution to assess mammograms within the breast cancer screening program. According to preliminary results, it is believed to be able to relieve specialists of 30 percent of their work and provide the women in the program with a faster response.

The method has great prospects—including abroad, says the Danish Cancer Society.

Every two years, women between the ages of 50 and 69 in Denmark are offered an X-ray examination of the breast with a view to detecting breast cancer in the early stages. In the Capital Region alone, 75,000 women accept the offer every year. At each breast cancer screening, four X-rays are taken, each of which must be assessed by two independent radiologists—the so-called “double reading”.

Artificial intelligence replaces one of two specialists

In November, the Capital Region of Denmark entered into an agreement with Danish company Human Bytes in order to supply the tool Transpara AI for use in breast cancer screening in the region. The goal was to eliminate the first radiologist from the analysis of mammograms in so-called low-risk cases. These are women without a history of breast cancer and without symptoms. Transpara AI has been developed by Dutch company ScreenPoint Medical and has been tested in clinical trials, just as the Capital Region has carried out pilot trials in the clinic.

But over the next few years, the AI will need to be used on a much larger scale. However, still under close scrutiny to determine whether the accuracy remains the same as that of the radiologists.

Saves 30 percent of radiologists’ screen time

During the first three months, the AI tool was used for the screening of 15,000 women—the mammograms were analysed by the algorithm and then looked at by a specialist. This so-called “single reading” method is widely used abroad, but has not been considered safe enough at home.

“The AI Mammo project is an algorithm that is based on pattern recognition, and it replaces one of the otherwise two specialists who assess the X-rays that are taken in connection with screening for breast cancer. The tool has been in use since November 2021. Already within the first two months, it has relieved the doctors who look at the X-rays of approximately 30% of their work,” says Bodil Ørkild, deputy director at Herlev and Gentofte Hospital, in the region’s press release. Bodil Ørkild is also chairwoman of a national working group that works to spread the use of artificial intelligence in the breast cancer screening programs on a national scale.

Faster response to screening

Chairman of the Health Committee in the Capital Region Christoffer Buster Reinhardt (C) expects that the use of artificial intelligence will help reduce waiting times in the screening program.

“It’s really good that we can help remedy the great shortage of staff with the use of new technology. Unfortunately, we lack breast radiologists, so them getting help from artificial intelligence means that we are able to give women faster answers to their X-rays than we otherwise would be. In this way, women who need to be examined for possibility of cancer can start receiving treatment faster,” said Christoffer Buster Reinhardt in a press release.

The tool being able to relieve the very limited number of radiologists by as much as 30 percent of their workload is better than expected. Ulrik Juul Rokkedal Therkildsen, partnership director at Human Bytes, said in a press release in November, when the agreement with the Capital Region was established, that one could expect a 25 percent reduction in the workload of breast radiologists.

Ready for use after extensive testing

Computer aided detection or CAD is not a new phenomenon in interpretation of medical images. Radiologists began using computer programs to aid them in diagnostics as early as the 1990s. The transition from X-rays on film to digital mammograms greatly contributed to the spread of CAD. However, the hope for earlier and better detection of breast cancer has not yet been fulfilled. On the other hand, many programs have tended to overdiagnose, not only affecting patients who are then overtreated. The many false positive results have also drawn resources from the system, so any savings in radiologist hours were offset by bottlenecks further down the process. Therefore, double reading is still recommended as standard in Europe.

Since the “primitive” algorithms in the traditional CAD systems, however, more computing power and new algorithms for interpretation of medical images have come into play.

“Modern image processing algorithms make use of machine learning (ML) techniques, predominantly of the deep learning (DL) type with large neural networks, which enable automatic training solely on the basis of large amounts of data. Potentiated by new high-performance graphics processing units, these algorithms are far superior to the classic ones, both in terms of data processing and pattern recognition,” read a status article on artificial intelligence for cancer diagnosis in breast cancer screening published in the Danish medical journal Ugeskrift for Læger in 2020.

The Capital Region of Denmark has in recent years tested several methods for AI analysis of mammograms and finds that artificial intelligence is as good and accurate as the current double reading involving two radiologists.
In addition to breast cancer screening, the region has shown interest in several of the so-called signature projects using AI in the public sector. Two of the projects focused on optimising the interpretation of medical images belong to the Radiological Artificial Intelligence Testcenter (RAIT), which is a consortium between the departments of radiology at Bispebjerg/Frederiksberg Hospital and Herlev/Gentofte Hospital. Among other things, RAIT seeks to use AI to prioritize and unify descriptions of X-rays of the lungs. Another project will apply already acquired experience with decision support to analyse X-rays of knees for osteoarthritis to all emergency radiology departments in the region.

The Danish Cancer Society is following the breast cancer project in the Capital Region.

“I have been presented with the results of using artificial intelligence in the clinic, and I am impressed by the new possibilities that help support and relieve the radiologists. There are great prospects in using artificial intelligence as a supplement. The Danish Cancer Society wants to see more of this in the future, and I hope we can learn from other areas. Artificial intelligence may be able to help solve the screening problem in the future, and that is good news—for every single woman who is waiting for screening,” says Jesper Fisker, managing director of the Danish Cancer Society.