Artificial intelligence (AI) in breast cancer screening has come a long way in a short period of time; however, it is not yet infallible. In a recent study, Invasive Breast Cancers Missed by AI Screening of Mammograms, 14% of invasive breast cancers (154 of 1097) were missed by AI, with the human epidermal growth factor receptor 2–enriched subtype showing the lowest false-negative rate and highest abnormality score. Among the AI-missed cancers, 61.7% (95 of 154) were actionable, often due to overlapping dense tissue, nonmammary zone locations, architectural distortions, or amorphous microcalcifications. Nonmammary zone locations are areas in the breast where cancer can occur outside of the typical mammary gland. Architectural distortion refers to an abnormality in breast tissue where the normal structure appears pulled or distorted without necessarily forming a distinct mass. Amorphous or indistinct calcifications are defined as ‘without a clearly defined shape or form’.

Compared with AI-detected cancers, AI-missed cancers were associated with younger age, a tumor size less than or equal to 2 cm, a lower histologic grade, fewer lymph node metastasis, low Ki-67 expression, fewer HER2-positive tumors, luminal subtype, more BI-RADS category 4 interpretations, and frequent nonmammary zone locations. These results are consistent with the known characteristics of AI-missed invasive cancers, which are more luminal subtypes, size smaller than 1 cm, and lack of axillary lymph node metastasis. Also in the study, sentinel lymph node biopsy, rather than axillary lymph node dissection, was more frequently performed in patients with AI-missed cancers than in those with AI-detected cancers. Accordingly, patients with AI-missed cancers may have a favorable prognosis, except for two factors, younger age and dense breasts, and may receive less invasive axillary surgery.
The most common reason for AI-missed cancers but actionable findings was the lesion being obscured by overlapping dense breast tissue. Interestingly, the second most common reason for AI misses was nonmammary zone locations. The retromammary fat layer and subareolar area were the two most common locations of missed lesions.

In conclusion, although AI is useful for detecting advanced-stage invasive cancers, it may be inadequate on its own for identifying cancers with some of the features revealed in this study. To avoid AI-missed invasive cancers on mammograms, your radiologist will perform meticulous assessments for luminal cancer, dense breasts, lesions located outside the mammary zone, architectural distortion, and amorphous microcalcifications. Patients should ask their radiologist about how they are assessing for these factors when AI is used.

Reference: Woo OH et al. Invasive Breast Cancers Missed by AI Screening of Mammograms. Radiology 2025; 315(3):e242408. https://doi.org/10.1148/radiol.242408.