A new study led by researchers at the UCLA Health Jonsson Comprehensive Cancer Center shows that so-called interval breast cancer can be detected earlier using AI. Interval breast cancer is the form of breast cancer that develops in the period between routine checkups, such as periodic screenings. It involves tumors that are sometimes visible on a mammogram but not detected by radiologists. For example, because the tumors show very subtle signs that are demonstrably below the level of detection by the human eye.
The study, published in the Journal of the National Cancer Institute, found that AI was able to identify “mammographically visible” types of interval cancers earlier by marking them at the time of screening. Researchers estimate that integrating AI into screening could help reduce the number of interval breast cancers by 30 percent. “For patients, early detection of cancer can make a big difference. It can lead to less aggressive treatment and increase the likelihood of a better outcome,” said Tiffany Yu, assistant professor of radiology at UCLA's David Geffen School of Medicine.
Similar research in Europe
Although similar research has been conducted in Europe, including by Swedish scientists, this study is one of the first to examine the use of AI to detect interval breast cancers in the United States. Researchers point out that there are important differences between U.S. and European screening practices.
In the US, most mammograms are performed using digital breast tomosynthesis (DBT), also known as 3D mammography, and patients are usually screened every year. In contrast, European programs usually use digital mammography (DM), also called 2D mammography, and screen patients every two to three years.
185,000 mammograms analyzed
The retrospective study analyzed data from nearly 185,000 past mammograms from 2010-2019 that included both DM and DBT. From the data, the team looked at 148 cases in which a woman was diagnosed with interval breast cancer. Radiologists then looked at these cases to determine why the cancer had not been detected earlier. The new study adapted a European classification system to categorize interval cancers. These include: Missed reading error, minimal signs-actionable, minimal signs-not-actionable, true interval cancer, occult (which is really invisible on a mammogram), and missed by technical error.
Researchers then applied a commercially available AI software called Transpara to initial screening mammograms taken before cancer diagnosis to determine whether it could detect subtle signs of cancer that radiologists missed during initial screenings, or at least flag them as suspicious. The tool gave each mammogram a cancer risk score from 1 to 10. A score of 8 or higher was considered potentially worrisome.
Key findings
- The team found that the AI marked more than three-quarters (76%) of mammograms that originally read as normal but were later linked to an interval breast cancer.
- The AI tool was able to flag missed reading errors, where the cancer was visible on the mammogram but missed or misinterpreted by the radiologist, at 90 percent.
- For cancers that were occult or completely invisible on the mammogram, the AI flagged 69 percent of cases.
- The AI did prove slightly less effective - 50 percent - in identifying true interval cancers, cancers that were not present at the time of screening but developed later.
“While we had some exciting results, we also uncovered many inaccuracies of the AI and problems that need to be investigated further in the real world,” he said. For example, despite the AI tool being invisible on mammograms, it still flagged 69% of screening mammograms with occult cancers. However, when we looked at the specific areas on the images that the AI marked as suspicious, the AI did not do as well and only marked the actual cancer 22% of the time,” said Dr. Hannah Milch, assistant professor of Radiology at the David Geffen School of Medicine and lead author of the study.
