Study Reveals Key Traits of Breast Cancers Often Missed by AI Tools
A new study sheds light on why artificial intelligence may fail to detect certain breast cancers, highlighting the need for cautious interpretation of AI results in clinical practice.
AI has shown promise in identifying early-stage breast cancer and easing radiologists’ workloads. Yet, prior studies suggest that AI can overlook specific cancers, particularly in certain patient populations. A recent analysis published in Radiology by the RSNA further explores this issue.
South Korean researchers evaluated breast imaging data from nearly 1,100 women screened between 2014 and 2020. Their findings point to a need for heightened scrutiny in cases featuring luminal cancer, dense breast tissue, or lesions outside the mammary zone.
“Although artificial intelligence is useful for detecting advanced-stage invasive cancers, it is inadequate for identifying cancers with some of the features revealed in this study,” noted lead author Ok Hee Woo, MD, of Korea University Guro Hospital in Seoul. “Understanding the features of AI-missed invasive cancers on mammograms can help readers use AI appropriately in clinical practice, thus contributing to its further optimization.”
In the study, a breast radiologist with 14 years of experience reviewed the mammograms and identified lesions. The research team used commercial AI software from Seoul-based Lunit to analyze consecutive screening exams. A cancer was labeled “AI-missed” if the software failed to mark the correct location as confirmed by the expert. Three additional radiologists reviewed these missed cases to assess whether the cancers could have been identified and to understand why AI had failed.
Overall, AI failed to detect approximately 14% of cancers (154 of 1,097 cases). These missed cases were more likely to involve younger patients, tumors 2 cm or smaller, low histologic grade, limited lymph node involvement, and Breast Imaging Reporting and Data System (BI-RADS) category 4 assessments. Radiologist-blinded reviews identified dense breast tissue (56 cases), nonmammary lesion locations (22), architectural distortion (12), and amorphous microcalcifications (5) as leading factors behind AI’s oversights.
An accompanying editorial by Lisa A. Mullen, MD, of Johns Hopkins, acknowledged the study’s limitations, including its exclusively Asian patient population and a high rate of dense breasts (over 70%). Still, she emphasized the study's importance in guiding radiologists on how to approach AI findings.
“When using AI, it is critical for radiologists to understand what could be potentially missed by the software so that the radiologist can use the information to decrease the chance of missed cancers,” Mullen wrote. “The radiologist should pay close attention to dense breasts and nonmammary zone areas, as well as search carefully for architectural distortion, microcalcifications, and small lesions. Areas of future research should include similar studies with more diverse patient populations and evaluation of other AI algorithms. Continued improvement of AI algorithms is also indicated to increase cancer detection and limit false-negative interpretation.”