[RSNA 2024] Lunit AI+1 radiologist more accurate than 2 radiologists in Swedish breast cancer screening

CHICAGO, Ill. -- By Lee Han-soo/Korea Biomedical correspondent -- At the ongoing 2024 annual meeting of the Radiological Society of North America (RSNA), Lunit presented follow-up findings from its ScreenTrustCAD trial on AI-assisted breast cancer screening.

The study showed the successful implementation of Lunit INSIGHT MMG, an AI-powered mammography analysis solution, at Capio S:t Göran Hospital in Sweden. The hospital has become the "first institution worldwide" to fully integrate AI as an independent reader, replacing one of two radiologists in their breast cancer screening protocol, according to Lunit.

The original ScreenTrustCAD trial (2021-2022), published in The Lancet Digital Health, was the first prospective interventional study to demonstrate that AI could match or surpass traditional two-radiologist systems. Following these results, Capio S:t Göran Hospital fully integrated Lunit INSIGHT MMG in mid-2023, pioneering a new approach to breast cancer screening.

The study, comparing nine-month periods before and after AI adoption, revealed a 36 percent reduction in reading time, an 11 percent decrease in recall rates, and a notable improvement in cancer detection rates from 4.8 to 5.5 per 1,000 screenings.

The positive predictive value for recalls increased from 16.9 percent to 22.1 percent, indicating more efficient and accurate screening processes.

In an exclusive interview, Korea Biomedical Review met with Dr. Karin Dembrower, Senior Physician of Capio S:t Göran Hospital Mammography Clinic, who shared insights into the implementation process and its impact.

"When we had the promising results from the ScreenTrustCAD trial, I was sitting together with my boss and discussing what to do with these results," she said. "It would be unethical not to implement that strategy when we know that it's better."

The transition from trial to implementation wasn't without challenges.

Dembrower revealed that the process took approximately 10 months, requiring careful attention to IT security, compliance with general data protection regulations (GDPR), and legal considerations.

"We had to look at the agreement between the region that gives us the assignment to perform screening at the hospital," she noted. "The agreement says that there should be two readers, and then we have to determine if AI could be considered a reader."

However, since the hospital began the research, it has confirmed positive results.

One of the most significant improvements has been the reduction in waiting time.

"Before, the waiting time could be six to eight weeks before biopsy and receiving results," Dr. Dembrower shared. "Patients can become extremely worried during that period and we believe also that the waiting time should be as short as possible."

This anxiety is something Dembrower witnesses firsthand in her practice. However, with the new AI-assisted workflow, these stressful waiting periods have been dramatically reduced to zero, allowing for more timely care and less anxiety for patients.

The implementation has also led to a meaningful shift in how radiologists utilize their expertise.

"We've experienced a task shifting for breast radiologists," Dembrower explained. "From assessing a lot of healthy women - we're inviting 80,000 women yearly at our facility - to performing more advanced diagnostics for diagnosed women, allowing us to use our skills better.”

Dembrower explained that the results may mean that the impact of this AI integration could be even more significant for smaller facilities.

"At many breast centers, like the smaller rural centers, there may be a huge lack of breast radiologists,” she said. “They really have to make efforts to use their time as efficiently as possible, and in those places, I think such a software could make a really huge difference."

This efficiency gain allows radiologists to focus on more complex cases and advanced diagnostics, better utilizing their specialized training and expertise, she added.

The success has attracted attention from patients across Sweden.

"Many women have been contacting us saying they really want their breasts to be screened at our facility since it seems to be working even better with AI," Dembrower noted. "We didn't have those kind of phone calls before."

Dembrower and her team are also planning to analyze interval cancer cases from both the study period and post-implementation phase, which will be unveiled soon.

“This comprehensive analysis of interval cancers—cases that develop between regular screening appointments—will provide crucial insights into the AI system's effectiveness in real-world clinical practice,” she said.

Looking toward the future, Dembrower emphasized the importance of simplicity in implementation.

"The more simple a solution is, the more likely it is to be implemented," she stated.

A company would need evidence and simple solutions because screening involves thousands of women, which means that a solution should not have complicated tracks, she added.