AI Matches Radiologists’ Skill at Detecting Breast Cancer on Mammograms

An artificial intelligence (AI) system based on deep learning algorithms is able to detect breast cancer in digital mammography screenings as well as experienced radiologists, a study says.

The findings of the study, “Stand-Alone Artificial Intelligence for Breast Cancer Detection in Mammography: Comparison With 101 Radiologists,” were published in the Journal of the National Cancer Institute.

Despite significant improvements in treatment and screening methods, breast cancer is still the most prevalent type of cancer in women and a major cause of cancer-related mortality, accounting for approximately 500,000 annual deaths worldwide.

Previous studies have shown that large-scale screening mammography programs are highly effective at reducing breast cancer mortality.

“However, current screening programs are highly labor intensive due to the large number of women screened per detected cancer and the use of double reading (diagnosis confirmation by a second expert), especially in European screening programs, which also leads to additional economical costs. Moreover, despite this practice, up to 25% of mammographically visible cancers are still not detected at screening,” the investigators said.

Considering the lack of breast cancer screening expert radiologists in some countries, finding alternative approaches to provide screening programs worldwide is crucial.

Since the 1990s, specialized AI systems have been developed to identify and classify breast cancer tumors visible in screening mammograms. However, it has never been demonstrated that these systems have a direct impact on breast cancer screening performance or cost-effectiveness.

Now, a group of researchers from the Radboud University Medical Center in the Netherlands and their collaborators set out to compare, at a case level, the performance of a commercially available AI system to that of expert radiologists in cancer detection using digital mammography screenings.

The retrospective study included nine multi-reader, multi-case study datasets from seven different countries. Each dataset included digital mammography screenings obtained using four different acquisition systems, multiple radiologists’ exam assessments, and histopathological analysis of the tumors, yielding a total of 2,652 exams (653 malignant) and 28,296 independent medical assessments performed by 101 radiologists.

Results showed the AI system had higher accuracy and sensitivity than 61.4% and 57.9% of the radiologists, respectively. However, its performance never surpassed that of the best radiologist. No significant differences were found on cancer detection performance between the AI system and the average performance of the 101 expert radiologists.

According to the study’s authors, “the evaluated AI system achieved a cancer detection accuracy comparable to an average breast radiologist in this retrospective setting.”

“Before we could decide what is the best way for AI systems to be introduced in the realm of breast cancer screening with mammography, we wanted to know how good can these systems really be,” Ioannis Sechopoulos, one of the study’s authors, said in a press release. “It was exciting to see that these systems have reached the level of matching the performance of not just radiologists, but of radiologists who spend at least a substantial portion of their time reading screening mammograms.”

The scientists concluded, “These results were consistently observed across a large, heterogeneous, multi-center, multi-vendor, cancer-enriched cohort of mammograms. Although promising, the performance and fashion of implementation of such an AI system in a screening setting remains to be further investigated.”