News • Reliability of mammography analysis
AI still misses 14% of advanced breast cancer cases, study finds
A Korean research team has reported that artificial intelligence (AI) used in mammography missed 14% of invasive breast cancers, the type where delayed diagnosis can directly impact patient survival.
Image source: Korea University College of Medicine
A Korean research team has reported that artificial intelligence (AI) used in mammography missed 14% of invasive breast cancers, the type where delayed diagnosis can directly impact patient survival. The study appears in Radiology.
Breast cancer is classified into ductal carcinoma in situ (stage 0) and invasive cancer (stages 1–4). While prior studies have estimated the overall false‑negative rate of AI‑assisted mammography for all breast cancers at about 19.4%, its performance in invasive cancers has remained a critical question.
Researchers from the Breast Center at Korea University College of Medicine — Professor Sungeun Song (Department of Radiology, Korea University Anam Hospital) and Professor Okhee Woo (Department of Radiology, Korea University Guro Hospital) — analyzed 1,097 breast cancer cases diagnosed between 2014 and 2020 using a commercially available Korean AI program (Lunit Insight MMG).
While AI shows strong performance in detecting breast cancer, our findings highlight the need for continuous oversight and complementary work by radiologists
Sungeun Song
The AI missed 17.2% of luminal‑type cancers, 14.5% of triple‑negative cancers, and 9% of HER2‑positive cancers. Missed invasive cancers tended to occur in younger women, had tumors 2 cm or smaller, lower histologic grade, fewer lymph node metastases, low Ki‑67 expression, were located outside glandular areas, and often fell into BI‑RADS category 4. Dense breast tissue, non‑glandular tumor location, structural distortion, and microcalcifications were the most common reasons for missed detection. Notably, 61.7% of these cancers were considered detectable by radiologists.
“While AI shows strong performance in detecting breast cancer, our findings highlight the need for continuous oversight and complementary work by radiologists,” said Professor Song. “Knowing the features of invasive cancers that AI tends to miss will be key to improving both clinical use and future development of AI tools.”
Source: Korea University College of Medicine
03.09.2025
