Artificial intelligence—experience from the breast imaging field

In the city of Malmö where Dr. Lång works, about 64,000 women attend mammography screenings every year.

“The double reading procedure means about 130,000 readings per year,” she says. “That’s quite a lot, but the Transpara application has helped us to prioritize exams in the PACS that require our immediate attention. The Transpara presorting tool uses the Exam Score, a 10-point scale to classify the exams according to cancer risk, enabling our radiologists to focus on the high-risk cases first. The seamless integration of AI applications with the PACS is vital for efficiency and safety, and the integration of Transpara with Sectra’s PACS works perfectly,” says Dr. Lång.

Dr. Lång working in the Sectra diagnostic application.

Human and machine in symbiosis

In addition to presorting, Transpara can be used as decision support when suspicious findings and their corresponding AI scores are highlighted using the CAD (computer-aided detection) functionality of Sectra’s PACS. The radiologists in Malmö have developed a simple strategy to enhance the reading of screenings using these CAD marks.

“We read the screenings without CAD marks first to ensure a holistic interpretation,” says Dr. Lång. “Then we turn on the CAD marks, and if there are any regions of interest, these are carefully assessed using a second-opinion approach. By working this way, we are combining human perception with computerized image analysis—human and machine in symbiosis.”

The future of mammography screening

Dr. Lång describes the discussions that have taken place—both in her department and internationally—about the benefits of AI and how AI tools can be further integrated with their daily routines.

“We’ve been discussing this for some time and have come up with a method that we believe can be effective and help to solve some of the challenges we are now facing in the breast imaging field. One of these is the shortage of breast radiologists in Sweden. Screening services will be at serious risk if we can’t come up with an alternative to the time-consuming double-reading procedure, and that would be detrimental to women,” she says.

In an attempt to address this challenge, Dr. Lång and her colleagues have initiated a trial to determine whether the integration of AI into mammography screening can improve efficacy and reduce the reading workload. The study called MASAI (Mammography Screening With Artificial Intelligence) will compare AI-integrated mammography screening reading with conventional double reading in a large-scale randomized trial with 100,000 women. This is the first prospective randomized trial focused on the use of AI in breast screening as an alternative for double reading.

The Transpara Exam Score will be used to identify low-risk screening exams for single reading, and high-risk screenings for double reading. Additionally, Transpara markers will be available during reading to provide decision support to radiologists.

“In addition to reducing the workload, the integration of AI into mammography screening has other potential benefits, most importantly a reduction of the interval cancer rate. Interval cancers are cancers that are missed when the last screening was read and diagnosed in the interval between the next round of screening,” Dr. Lång explains.

“Interval cancers can be very aggressive and, if detected at an earlier stage, we could further reduce the breast cancer mortality rate. We have shown how AI has the potential to lower the interval cancer rate by 20%, but this has yet to be tested prospectively. The primary aim of the MASAI trial is to measure the effect on interval cancer rates. However, if we discover that it is not possible to reduce interval cancer rates with AI, we may still be able to maintain a similar level of cancer detection with a considerably lower screen-reading volume for radiologists.”

“We also hypothesize that AI-integrated screening has the potential to reduce false positives, which is a major drawback with current mammography screening,” says Dr. Lång. “A false positive result can cause significant stress and anxiety for women who are recalled for further testing, and that should be kept as low as possible.”

This MASAI trial commenced in April 2021 and the research team expects to present its first findings by the autumn of 2022.

How to choose an AI vendor

When asked what to consider when selecting an AI vendor, Dr. Lång emphasizes the importance of working with companies that are transparent.

“We want to know how the algorithms have been developed and what datasets they are based on, and most importantly, that the performance of the AI system has been evaluated clinically by independent researchers. The company we are currently working with is transparent and has a long research tradition. I also recommend checking whether the application is compatible with your PACS.”

She also appreciates the fact that Sectra now offers its customers the Sectra Amplifier Marketplace, where a wide range of validated AI tools from various vendors are ready for integration with the PACS from the start.

“I think this way of working sounds fantastic. We’ve spent a lot of time making everything work for us, and the ability to just go online, buy what you need from a single interface, and then go back to your PACS and start working seems so much easier.”

Communicating AI to the women

As AI applications become more common in clinical practice and the technology continues to evolve, there is a need to communicate to women how AI is used in a credible manner. Dr. Lång tells how women have previously been skeptical of AI, but the trend is now turning.

“New studies show that women are now more in favor of AI, a trend that we have also noticed here in Malmö. There is more positivity and curiosity about these new techniques in general, and I also think it depends on how we communicate about the tools we use. We emphasize how the AI technology we use supports the doctor, but can never replace them.”

Overall, breast imaging is one of the areas in which AI will probably have a significant impact on medical imaging in the coming years. But there are still steps to climb before AI becomes a standard of care. In her previous work, Dr. Lång has shown that AI has the potential to lower interval cancer rates and reduce the mammography screening workload. The results of the MASAI trial will contribute to a better understanding of the role of AI in mammography screening and determine whether it is a safe and efficient method for improving the efficacy of breast cancer screening.
   

###

Update per August 15, 2023

The first results of the MASAI study was published in August 2023 in The Lancet Oncology. To summarize, the results showed that AI-supported mammography screening resulted in a similar cancer detection rate (6.1 per 1,000) compared with standard double reading (5.1 per 1,000), with a reduced screen-reading workload by a remarkable 44.3%. This indicates that the use of AI in mammography screening is safe. The trial was thus not halted, and the primary endpoint of interval cancer rate will be assessed in 100,000 enrolled participants after 2-years of follow up.